From 5e72deb7d4229716465ca6bf5dc7b8da26e83c76 Mon Sep 17 00:00:00 2001
From: feitianbubu <feitianbubu@qq.com>
Date: Fri, 1 May 2026 14:16:49 +0800
Subject: [PATCH 01/68] =?UTF-8?q?feat:=20ops=20=E9=94=99=E8=AF=AF=E8=AF=A6?=
 =?UTF-8?q?=E6=83=85=E5=BC=B9=E7=AA=97=E6=94=AF=E6=8C=81=E8=87=AA=E5=AE=9A?=
 =?UTF-8?q?=E4=B9=89=E6=97=B6=E9=97=B4=E5=8C=BA=E9=97=B4?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 frontend/src/views/admin/ops/OpsDashboard.vue     |  2 ++
 .../admin/ops/components/OpsErrorDetailsModal.vue | 15 ++++++++++++++-
 2 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/frontend/src/views/admin/ops/OpsDashboard.vue b/frontend/src/views/admin/ops/OpsDashboard.vue
index 50bc52494d2a..8dc3c68eb402 100644
--- a/frontend/src/views/admin/ops/OpsDashboard.vue
+++ b/frontend/src/views/admin/ops/OpsDashboard.vue
@@ -114,6 +114,8 @@
         <OpsErrorDetailsModal
           :show="showErrorDetails"
           :time-range="timeRange"
+          :custom-start-time="customStartTime"
+          :custom-end-time="customEndTime"
           :platform="platform"
           :group-id="groupId"
           :error-type="errorDetailsType"
diff --git a/frontend/src/views/admin/ops/components/OpsErrorDetailsModal.vue b/frontend/src/views/admin/ops/components/OpsErrorDetailsModal.vue
index e9e4401001df..4c17b0be8534 100644
--- a/frontend/src/views/admin/ops/components/OpsErrorDetailsModal.vue
+++ b/frontend/src/views/admin/ops/components/OpsErrorDetailsModal.vue
@@ -9,6 +9,8 @@ import { opsAPI, type OpsErrorLog } from '@/api/admin/ops'
 interface Props {
   show: boolean
   timeRange: string
+  customStartTime?: string | null
+  customEndTime?: string | null
   platform?: string
   groupId?: number | null
   errorType: 'request' | 'upstream'
@@ -96,6 +98,17 @@ async function fetchErrorLogs() {
       view: viewMode.value
     }
 
+    if (props.timeRange === 'custom') {
+      if (props.customStartTime && props.customEndTime) {
+        params.start_time = props.customStartTime
+        params.end_time = props.customEndTime
+        delete params.time_range
+      } else {
+        // Safety fallback: avoid sending time_range=custom (backend doesn't support it)
+        params.time_range = '1h'
+      }
+    }
+
     const platform = String(props.platform || '').trim()
     if (platform) params.platform = platform
     if (typeof props.groupId === 'number' && props.groupId > 0) params.group_id = props.groupId
@@ -147,7 +160,7 @@ watch(
 )
 
 watch(
-  () => [props.timeRange, props.platform, props.groupId] as const,
+  () => [props.timeRange, props.customStartTime, props.customEndTime, props.platform, props.groupId] as const,
   () => {
     if (!props.show) return
     page.value = 1

From 3bff4b64b1c6a5460ed9c53889db777cbf0053f7 Mon Sep 17 00:00:00 2001
From: feitianbubu <feitianbubu@qq.com>
Date: Fri, 10 Jul 2026 23:43:16 +0800
Subject: [PATCH 02/68] fix(ui): localize user role label in app header

---
 frontend/src/components/layout/AppHeader.vue | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/frontend/src/components/layout/AppHeader.vue b/frontend/src/components/layout/AppHeader.vue
index d0e45260551f..aa8b0cffb21d 100644
--- a/frontend/src/components/layout/AppHeader.vue
+++ b/frontend/src/components/layout/AppHeader.vue
@@ -111,8 +111,8 @@
               <div class="text-sm font-medium text-gray-900 dark:text-white">
                 {{ displayName }}
               </div>
-              <div class="text-xs capitalize text-gray-500 dark:text-dark-400">
-                {{ user.role }}
+              <div class="text-xs text-gray-500 dark:text-dark-400">
+                {{ t('admin.users.roles.' + user.role) }}
               </div>
             </div>
             <Icon name="chevronDown" size="sm" class="hidden text-gray-400 md:block" />

From 7d796f1119a36f0cb48541de54e07f443def1a21 Mon Sep 17 00:00:00 2001
From: feitianbubu <feitianbubu@qq.com>
Date: Sat, 11 Jul 2026 00:08:26 +0800
Subject: [PATCH 03/68] fix(ui): adapt native form controls to dark mode via
 color-scheme

---
 frontend/src/components/common/DateRangePicker.vue | 2 +-
 frontend/src/style.css                             | 5 +++++
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/frontend/src/components/common/DateRangePicker.vue b/frontend/src/components/common/DateRangePicker.vue
index b00334a67a1b..56861979bfc4 100644
--- a/frontend/src/components/common/DateRangePicker.vue
+++ b/frontend/src/components/common/DateRangePicker.vue
@@ -405,7 +405,7 @@ onUnmounted(() => {
 }
 
 .dark .date-picker-input::-webkit-calendar-picker-indicator {
-  filter: invert(0.7);
+  filter: none;
 }
 
 .date-picker-separator {
diff --git a/frontend/src/style.css b/frontend/src/style.css
index 1bd13b5f95a7..bf518e93252b 100644
--- a/frontend/src/style.css
+++ b/frontend/src/style.css
@@ -9,6 +9,11 @@
 
   html {
     @apply scroll-smooth antialiased;
+    color-scheme: light;
+  }
+
+  .dark {
+    color-scheme: dark;
   }
 
   body {

From a6d868f27b64e6b2e2dd96a69560f82b7243846a Mon Sep 17 00:00:00 2001
From: feitianbubu <feitianbubu@qq.com>
Date: Sat, 11 Jul 2026 17:18:25 +0800
Subject: [PATCH 04/68] fix(dashboard): include cache tokens in token card
 breakdown

---
 frontend/src/components/user/dashboard/UserDashboardStats.vue | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/frontend/src/components/user/dashboard/UserDashboardStats.vue b/frontend/src/components/user/dashboard/UserDashboardStats.vue
index 7a0fcc202e16..7cc013b8207f 100644
--- a/frontend/src/components/user/dashboard/UserDashboardStats.vue
+++ b/frontend/src/components/user/dashboard/UserDashboardStats.vue
@@ -78,7 +78,7 @@
         <div>
           <p class="text-xs font-medium text-gray-500 dark:text-gray-400">{{ t('dashboard.todayTokens') }}</p>
           <p class="text-xl font-bold text-gray-900 dark:text-white">{{ formatTokens(stats?.today_tokens || 0) }}</p>
-          <p class="text-xs text-gray-500 dark:text-gray-400">{{ t('dashboard.input') }}: {{ formatTokens(stats?.today_input_tokens || 0) }} / {{ t('dashboard.output') }}: {{ formatTokens(stats?.today_output_tokens || 0) }}</p>
+          <p class="text-xs text-gray-500 dark:text-gray-400">{{ t('dashboard.input') }}: {{ formatTokens(stats?.today_input_tokens || 0) }} / {{ t('dashboard.output') }}: {{ formatTokens(stats?.today_output_tokens || 0) }} / {{ t('dashboard.cache') }}: {{ formatTokens((stats?.today_cache_creation_tokens || 0) + (stats?.today_cache_read_tokens || 0)) }}</p>
         </div>
       </div>
     </div>
@@ -92,7 +92,7 @@
         <div>
           <p class="text-xs font-medium text-gray-500 dark:text-gray-400">{{ t('dashboard.totalTokens') }}</p>
           <p class="text-xl font-bold text-gray-900 dark:text-white">{{ formatTokens(stats?.total_tokens || 0) }}</p>
-          <p class="text-xs text-gray-500 dark:text-gray-400">{{ t('dashboard.input') }}: {{ formatTokens(stats?.total_input_tokens || 0) }} / {{ t('dashboard.output') }}: {{ formatTokens(stats?.total_output_tokens || 0) }}</p>
+          <p class="text-xs text-gray-500 dark:text-gray-400">{{ t('dashboard.input') }}: {{ formatTokens(stats?.total_input_tokens || 0) }} / {{ t('dashboard.output') }}: {{ formatTokens(stats?.total_output_tokens || 0) }} / {{ t('dashboard.cache') }}: {{ formatTokens((stats?.total_cache_creation_tokens || 0) + (stats?.total_cache_read_tokens || 0)) }}</p>
         </div>
       </div>
     </div>

From 35e8ba2a3ea1e8b351f958f49df66eef80b407d1 Mon Sep 17 00:00:00 2001
From: feitianbubu <feitianbubu@qq.com>
Date: Sat, 11 Jul 2026 18:28:21 +0800
Subject: [PATCH 05/68] fix(announcements): use proper empty-state copy instead
 of error message

---
 frontend/src/i18n/locales/en/admin/resources.ts | 1 +
 frontend/src/i18n/locales/zh/admin/resources.ts | 1 +
 frontend/src/views/admin/AnnouncementsView.vue  | 2 +-
 3 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/frontend/src/i18n/locales/en/admin/resources.ts b/frontend/src/i18n/locales/en/admin/resources.ts
index 8e9e34c0ec1f..a25fc9cd7831 100644
--- a/frontend/src/i18n/locales/en/admin/resources.ts
+++ b/frontend/src/i18n/locales/en/admin/resources.ts
@@ -375,6 +375,7 @@ export default {
     announcements: {
       title: 'Announcements',
       description: 'Create announcements and target by conditions',
+      createFirstAnnouncement: 'No announcements yet. Create your first one.',
       createAnnouncement: 'Create Announcement',
       editAnnouncement: 'Edit Announcement',
       deleteAnnouncement: 'Delete Announcement',
diff --git a/frontend/src/i18n/locales/zh/admin/resources.ts b/frontend/src/i18n/locales/zh/admin/resources.ts
index 2019a0fe0c7c..12bcab4d9e33 100644
--- a/frontend/src/i18n/locales/zh/admin/resources.ts
+++ b/frontend/src/i18n/locales/zh/admin/resources.ts
@@ -372,6 +372,7 @@ export default {
     announcements: {
       title: '公告管理',
       description: '创建公告并按条件投放',
+      createFirstAnnouncement: '还没有公告，创建您的第一条公告。',
       createAnnouncement: '创建公告',
       editAnnouncement: '编辑公告',
       deleteAnnouncement: '删除公告',
diff --git a/frontend/src/views/admin/AnnouncementsView.vue b/frontend/src/views/admin/AnnouncementsView.vue
index a39e323f0d45..d1846d4f1899 100644
--- a/frontend/src/views/admin/AnnouncementsView.vue
+++ b/frontend/src/views/admin/AnnouncementsView.vue
@@ -141,7 +141,7 @@
           <template #empty>
             <EmptyState
               :title="t('empty.noData')"
-              :description="t('admin.announcements.failedToLoad')"
+              :description="t('admin.announcements.createFirstAnnouncement')"
               :action-text="t('admin.announcements.createAnnouncement')"
               @action="openCreateDialog"
             />

From 0d5e3ca9b03bd555329110ba9dffa7275f5866ec Mon Sep 17 00:00:00 2001
From: feitianbubu <feitianbubu@qq.com>
Date: Sat, 11 Jul 2026 19:48:08 +0800
Subject: [PATCH 06/68] fix(ops): show neutral SLA card when window has no
 requests

---
 frontend/src/views/admin/ops/components/OpsDashboardHeader.vue | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/frontend/src/views/admin/ops/components/OpsDashboardHeader.vue b/frontend/src/views/admin/ops/components/OpsDashboardHeader.vue
index ab95cd306c5f..8ca67ac84daf 100644
--- a/frontend/src/views/admin/ops/components/OpsDashboardHeader.vue
+++ b/frontend/src/views/admin/ops/components/OpsDashboardHeader.vue
@@ -393,6 +393,7 @@ const tpsAvgLabel = computed(() => {
 const slaPercent = computed(() => {
   const v = overview.value?.sla
   if (typeof v !== 'number') return null
+  if ((overview.value?.request_count_sla ?? 0) <= 0) return null
   return v * 100
 })
 
@@ -1271,7 +1272,7 @@ function handleToolbarRefresh() {
           <div class="mt-3 text-xs">
             <div class="flex justify-between">
               <span class="text-gray-500">{{ t('admin.ops.exceptions') }}:</span>
-              <span class="font-bold text-red-600 dark:text-red-400">{{ formatNumber((overview.request_count_sla ?? 0) - (overview.success_count ?? 0)) }}</span>
+              <span class="font-bold text-gray-900 dark:text-white">{{ formatNumber((overview.request_count_sla ?? 0) - (overview.success_count ?? 0)) }}</span>
             </div>
           </div>
         </div>

From 9c36b75a7d2feaf20db7de25b3ab33f108e04279 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Tue, 28 Jul 2026 16:56:17 +0800
Subject: [PATCH 07/68] fix(claude): strip cache control from deferred tools

---
 ...teway_anthropic_apikey_passthrough_test.go | 24 ++++++++++
 .../service/gateway_anthropic_passthrough.go  |  1 +
 .../gateway_context_management_test.go        | 46 +++++++++++++++++++
 .../internal/service/gateway_count_tokens.go  |  2 +
 .../internal/service/gateway_tool_rewrite.go  | 43 +++++++++++++++--
 .../service/gateway_tool_rewrite_test.go      | 28 +++++++++++
 .../service/gateway_upstream_request.go       |  1 +
 7 files changed, 141 insertions(+), 4 deletions(-)

diff --git a/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go b/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
index 4fa52cc9ce80..102c2b29048f 100644
--- a/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
+++ b/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
@@ -769,6 +769,30 @@ func TestGatewayService_AnthropicAPIKeyPassthrough_BuildRequestRejectsInvalidBas
 	require.Error(t, err)
 }
 
+func TestGatewayService_AnthropicAPIKeyPassthrough_StripsDeferredToolCacheControl(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
+	svc := &GatewayService{cfg: &config.Config{Security: config.SecurityConfig{URLAllowlist: config.URLAllowlistConfig{Enabled: false}}}}
+	account := &Account{Platform: PlatformAnthropic, Type: AccountTypeAPIKey}
+	body := []byte(`{"tools":[{"name":"deferred","custom":{"defer_loading":true},"cache_control":{"type":"ephemeral"}},{"name":"ordinary","custom":{"defer_loading":false},"cache_control":{"type":"ephemeral"}},{"name":"malformed","custom":{"defer_loading":"true"},"cache_control":{"type":"ephemeral"}}]}`)
+
+	_, wireBody, err := svc.buildUpstreamRequestAnthropicAPIKeyPassthrough(context.Background(), c, account, body, "k")
+	require.NoError(t, err)
+	require.False(t, gjson.GetBytes(wireBody, "tools.0.cache_control").Exists())
+	require.True(t, gjson.GetBytes(wireBody, "tools.1.cache_control").Exists())
+	require.True(t, gjson.GetBytes(wireBody, "tools.2.cache_control").Exists())
+
+	countReq, err := svc.buildCountTokensRequestAnthropicAPIKeyPassthrough(context.Background(), c, account, body, "k")
+	require.NoError(t, err)
+	countBody, err := io.ReadAll(countReq.Body)
+	require.NoError(t, err)
+	require.False(t, gjson.GetBytes(countBody, "tools.0.cache_control").Exists())
+	require.True(t, gjson.GetBytes(countBody, "tools.1.cache_control").Exists())
+	require.True(t, gjson.GetBytes(countBody, "tools.2.cache_control").Exists())
+}
+
 func TestGatewayService_AnthropicOAuth_NotAffectedByAPIKeyPassthroughToggle(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	rec := httptest.NewRecorder()
diff --git a/backend/internal/service/gateway_anthropic_passthrough.go b/backend/internal/service/gateway_anthropic_passthrough.go
index c18d5bf047cd..568d560bb3d8 100644
--- a/backend/internal/service/gateway_anthropic_passthrough.go
+++ b/backend/internal/service/gateway_anthropic_passthrough.go
@@ -310,6 +310,7 @@ func (s *GatewayService) buildUpstreamRequestAnthropicAPIKeyPassthrough(
 	body []byte,
 	token string,
 ) (*http.Request, []byte, error) {
+	body = stripDeferredToolCacheControl(body)
 	targetURL := claudeAPIURL
 	baseURL := account.GetBaseURL()
 	if baseURL != "" {
diff --git a/backend/internal/service/gateway_context_management_test.go b/backend/internal/service/gateway_context_management_test.go
index bdeb2c600b8b..b7ab3e942976 100644
--- a/backend/internal/service/gateway_context_management_test.go
+++ b/backend/internal/service/gateway_context_management_test.go
@@ -4,6 +4,7 @@ package service
 
 import (
 	"context"
+	"fmt"
 	"io"
 	"net/http"
 	"net/http/httptest"
@@ -650,6 +651,51 @@ func TestBuildCountTokensRequest_APIKeyHaiku_StripsContextManagementEndToEnd(t *
 		"count_tokens API-key + 客户端未带 beta token → body strip")
 }
 
+func TestBuildCountTokensRequest_StripsCacheControlOnlyFromLiteralDeferredTools(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	body := []byte(`{"model":"claude-haiku-4-5","messages":[],"tools":[{"name":"deferred","custom":{"defer_loading":true},"cache_control":{"type":"ephemeral"}},{"name":"ordinary","custom":{"defer_loading":false},"cache_control":{"type":"ephemeral"}},{"name":"string","custom":{"defer_loading":"true"},"cache_control":{"type":"ephemeral"}},{"name":"number","custom":{"defer_loading":1},"cache_control":{"type":"ephemeral"}},{"name":"object","custom":{"defer_loading":{}},"cache_control":{"type":"ephemeral"}}]}`)
+
+	tests := []struct {
+		name      string
+		account   *Account
+		token     string
+		tokenType string
+	}{
+		{
+			name:      "generic API key",
+			account:   &Account{Platform: PlatformAnthropic, Type: AccountTypeAPIKey},
+			token:     "sk-ant-test",
+			tokenType: "apikey",
+		},
+		{
+			name:      "recognized Claude Code OAuth without mimicry",
+			account:   &Account{Platform: PlatformAnthropic, Type: AccountTypeOAuth},
+			token:     "oauth-token",
+			tokenType: "oauth",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			rec := httptest.NewRecorder()
+			c, _ := gin.CreateTestContext(rec)
+			c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages/count_tokens", nil)
+			svc := &GatewayService{cfg: &config.Config{}}
+
+			req, wireBody, err := svc.buildCountTokensRequest(
+				context.Background(), c, tt.account, body,
+				tt.token, tt.tokenType, "claude-haiku-4-5", false,
+			)
+			require.NoError(t, err)
+			require.False(t, gjson.GetBytes(wireBody, "tools.0.cache_control").Exists())
+			for idx := 1; idx < 5; idx++ {
+				require.Equal(t, "ephemeral", gjson.GetBytes(wireBody, fmt.Sprintf("tools.%d.cache_control.type", idx)).String())
+			}
+			require.JSONEq(t, string(wireBody), string(readUpstreamBodyForTest(t, req)))
+		})
+	}
+}
+
 // count_tokens passthrough preserve 测试
 func TestBuildCountTokensRequestAnthropicAPIKeyPassthrough_PreservesContextManagementWhenClientHeaderHasBeta(t *testing.T) {
 	gin.SetMode(gin.TestMode)
diff --git a/backend/internal/service/gateway_count_tokens.go b/backend/internal/service/gateway_count_tokens.go
index 7b66ac9596e4..003cbb7c1e99 100644
--- a/backend/internal/service/gateway_count_tokens.go
+++ b/backend/internal/service/gateway_count_tokens.go
@@ -366,6 +366,7 @@ func (s *GatewayService) buildCountTokensRequestAnthropicAPIKeyPassthrough(
 	body []byte,
 	token string,
 ) (*http.Request, error) {
+	body = stripDeferredToolCacheControl(body)
 	targetURL := claudeAPICountTokensURL
 	baseURL := account.GetBaseURL()
 	if baseURL != "" {
@@ -429,6 +430,7 @@ func (s *GatewayService) buildCountTokensRequestAnthropicAPIKeyPassthrough(
 
 // buildCountTokensRequest 构建 count_tokens 上游请求
 func (s *GatewayService) buildCountTokensRequest(ctx context.Context, c *gin.Context, account *Account, body []byte, token, tokenType, modelID string, mimicClaudeCode bool) (*http.Request, []byte, error) {
+	body = stripDeferredToolCacheControl(body)
 	// 确定目标 URL
 	targetURL := claudeAPICountTokensURL
 	if account.Type == AccountTypeAPIKey {
diff --git a/backend/internal/service/gateway_tool_rewrite.go b/backend/internal/service/gateway_tool_rewrite.go
index da62daf96172..42d786fe3eda 100644
--- a/backend/internal/service/gateway_tool_rewrite.go
+++ b/backend/internal/service/gateway_tool_rewrite.go
@@ -248,14 +248,17 @@ func applyToolNameRewriteToBody(body []byte, rw *ToolNameRewrite) []byte {
 	return body
 }
 
-// applyToolsLastCacheBreakpoint 在 tools 数组最后一个工具上注入 cache_control
-// 断点，对齐 Parrot `tools[-1]["cache_control"] = {"type":"ephemeral","ttl":"1h"}`
-// 行为，但 ttl 按本仓规则：
+// applyToolsLastCacheBreakpoint 在最后一个非延迟加载工具上注入 cache_control
+// 断点。Anthropic 不允许 custom.defer_loading=true 的工具携带 cache_control，
+// 因此会先清理所有延迟加载工具上的客户端断点。其余行为对齐 Parrot
+// `tools[-1]["cache_control"] = {"type":"ephemeral","ttl":"1h"}`，
+// 但 ttl 按本仓规则：
 //   - 客户端已为该 tool 显式设置 cache_control.ttl → 完全透传不覆盖
 //   - 否则注入 {"type":"ephemeral","ttl": claude.DefaultCacheControlTTL}
 //
 // 纯副作用函数，tools 不存在或为空数组时 no-op。
 func applyToolsLastCacheBreakpoint(body []byte) []byte {
+	body = stripDeferredToolCacheControl(body)
 	tools := gjson.GetBytes(body, "tools")
 	if !tools.IsArray() {
 		return body
@@ -264,7 +267,17 @@ func applyToolsLastCacheBreakpoint(body []byte) []byte {
 	if len(arr) == 0 {
 		return body
 	}
-	lastIdx := len(arr) - 1
+	lastIdx := -1
+	for idx, tool := range arr {
+		if isDeferredLoadingTool(tool) {
+			continue
+		}
+		lastIdx = idx
+	}
+	if lastIdx == -1 {
+		return body
+	}
+
 	existingCC := arr[lastIdx].Get("cache_control")
 
 	if existingCC.Exists() && existingCC.Get("ttl").String() != "" {
@@ -285,6 +298,28 @@ func applyToolsLastCacheBreakpoint(body []byte) []byte {
 	return body
 }
 
+func isDeferredLoadingTool(tool gjson.Result) bool {
+	return tool.Get("custom.defer_loading").Type == gjson.True
+}
+
+// stripDeferredToolCacheControl removes the cache marker Anthropic rejects on
+// deferred tools. Only the literal JSON boolean true enables deferred loading.
+func stripDeferredToolCacheControl(body []byte) []byte {
+	tools := gjson.GetBytes(body, "tools")
+	if !tools.IsArray() {
+		return body
+	}
+	for idx, tool := range tools.Array() {
+		if !isDeferredLoadingTool(tool) || !tool.Get("cache_control").Exists() {
+			continue
+		}
+		if next, err := sjson.DeleteBytes(body, fmt.Sprintf("tools.%d.cache_control", idx)); err == nil {
+			body = next
+		}
+	}
+	return body
+}
+
 // restoreToolNamesInBytes 对 bytes chunk 做逆向还原：假名 → 真名。
 // 按 ReverseOrdered 的假名长度倒序逐个 bytes.Replace，防止子串冲突
 // （与 Parrot _restore_tool_names_in_chunk 的 sorted(..., reverse=True) 等价）。
diff --git a/backend/internal/service/gateway_tool_rewrite_test.go b/backend/internal/service/gateway_tool_rewrite_test.go
index 9e6f6806da53..d35f0181a77a 100644
--- a/backend/internal/service/gateway_tool_rewrite_test.go
+++ b/backend/internal/service/gateway_tool_rewrite_test.go
@@ -2,6 +2,7 @@ package service
 
 import (
 	"context"
+	"fmt"
 	"strings"
 	"testing"
 
@@ -149,6 +150,33 @@ func TestApplyToolsLastCacheBreakpoint_PassesThroughClientTTL(t *testing.T) {
 	require.Equal(t, "1h", gjson.GetBytes(out, "tools.0.cache_control.ttl").String())
 }
 
+func TestApplyToolsLastCacheBreakpoint_StripsDeferredToolCacheControl(t *testing.T) {
+	body := []byte(`{"tools":[{"name":"a","custom":{"defer_loading":true},"cache_control":{"type":"ephemeral","ttl":"1h"}},{"name":"b","custom":{"defer_loading":true}}]}`)
+	out := applyToolsLastCacheBreakpoint(body)
+
+	require.False(t, gjson.GetBytes(out, "tools.0.cache_control").Exists())
+	require.False(t, gjson.GetBytes(out, "tools.1.cache_control").Exists())
+}
+
+func TestApplyToolsLastCacheBreakpoint_SkipsDeferredFinalTool(t *testing.T) {
+	body := []byte(`{"tools":[{"name":"a","input_schema":{}},{"name":"b","custom":{"defer_loading":true}}]}`)
+	out := applyToolsLastCacheBreakpoint(body)
+
+	require.Equal(t, "ephemeral", gjson.GetBytes(out, "tools.0.cache_control.type").String())
+	require.Equal(t, "5m", gjson.GetBytes(out, "tools.0.cache_control.ttl").String())
+	require.False(t, gjson.GetBytes(out, "tools.1.cache_control").Exists())
+}
+
+func TestApplyToolsLastCacheBreakpoint_OnlyLiteralTrueIsDeferred(t *testing.T) {
+	body := []byte(`{"tools":[{"name":"true","custom":{"defer_loading":true},"cache_control":{"type":"ephemeral"}},{"name":"false","custom":{"defer_loading":false},"cache_control":{"type":"ephemeral"}},{"name":"string","custom":{"defer_loading":"true"},"cache_control":{"type":"ephemeral"}},{"name":"number","custom":{"defer_loading":1},"cache_control":{"type":"ephemeral"}},{"name":"object","custom":{"defer_loading":{}},"cache_control":{"type":"ephemeral"}}]}`)
+	out := stripDeferredToolCacheControl(body)
+
+	require.False(t, gjson.GetBytes(out, "tools.0.cache_control").Exists())
+	for idx := 1; idx < 5; idx++ {
+		require.Equal(t, "ephemeral", gjson.GetBytes(out, fmt.Sprintf("tools.%d.cache_control.type", idx)).String())
+	}
+}
+
 func TestStripMessageCacheControl(t *testing.T) {
 	body := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi","cache_control":{"type":"ephemeral"}}]}]}`)
 	out := stripMessageCacheControl(body)
diff --git a/backend/internal/service/gateway_upstream_request.go b/backend/internal/service/gateway_upstream_request.go
index c5d962f20bc1..0a8efbdd5e57 100644
--- a/backend/internal/service/gateway_upstream_request.go
+++ b/backend/internal/service/gateway_upstream_request.go
@@ -19,6 +19,7 @@ import (
 )
 
 func (s *GatewayService) buildUpstreamRequest(ctx context.Context, c *gin.Context, account *Account, body []byte, token, tokenType, modelID string, reqStream bool, mimicClaudeCode bool) (*http.Request, []byte, error) {
+	body = stripDeferredToolCacheControl(body)
 	if account.Platform == PlatformAnthropic && account.Type == AccountTypeServiceAccount {
 		req, err := s.buildUpstreamRequestAnthropicVertex(ctx, c, account, body, token, modelID, reqStream)
 		return req, body, err

From 0b35370a7ab9fd419574922411bf94531e4350ee Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Wed, 29 Jul 2026 16:41:36 +0800
Subject: [PATCH 08/68] fix(claude): support top-level deferred tools

---
 .../service/gateway_anthropic_apikey_passthrough_test.go  | 8 +++++---
 backend/internal/service/gateway_tool_rewrite.go          | 8 +++++---
 backend/internal/service/gateway_tool_rewrite_test.go     | 7 ++++---
 3 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go b/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
index 102c2b29048f..acd6c6136f8e 100644
--- a/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
+++ b/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
@@ -776,21 +776,23 @@ func TestGatewayService_AnthropicAPIKeyPassthrough_StripsDeferredToolCacheContro
 	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
 	svc := &GatewayService{cfg: &config.Config{Security: config.SecurityConfig{URLAllowlist: config.URLAllowlistConfig{Enabled: false}}}}
 	account := &Account{Platform: PlatformAnthropic, Type: AccountTypeAPIKey}
-	body := []byte(`{"tools":[{"name":"deferred","custom":{"defer_loading":true},"cache_control":{"type":"ephemeral"}},{"name":"ordinary","custom":{"defer_loading":false},"cache_control":{"type":"ephemeral"}},{"name":"malformed","custom":{"defer_loading":"true"},"cache_control":{"type":"ephemeral"}}]}`)
+	body := []byte(`{"tools":[{"name":"deferred","custom":{"defer_loading":true},"cache_control":{"type":"ephemeral"}},{"name":"top-level-deferred","defer_loading":true,"cache_control":{"type":"ephemeral"}},{"name":"ordinary","defer_loading":false,"cache_control":{"type":"ephemeral"}},{"name":"malformed","defer_loading":"true","cache_control":{"type":"ephemeral"}}]}`)
 
 	_, wireBody, err := svc.buildUpstreamRequestAnthropicAPIKeyPassthrough(context.Background(), c, account, body, "k")
 	require.NoError(t, err)
 	require.False(t, gjson.GetBytes(wireBody, "tools.0.cache_control").Exists())
-	require.True(t, gjson.GetBytes(wireBody, "tools.1.cache_control").Exists())
+	require.False(t, gjson.GetBytes(wireBody, "tools.1.cache_control").Exists())
 	require.True(t, gjson.GetBytes(wireBody, "tools.2.cache_control").Exists())
+	require.True(t, gjson.GetBytes(wireBody, "tools.3.cache_control").Exists())
 
 	countReq, err := svc.buildCountTokensRequestAnthropicAPIKeyPassthrough(context.Background(), c, account, body, "k")
 	require.NoError(t, err)
 	countBody, err := io.ReadAll(countReq.Body)
 	require.NoError(t, err)
 	require.False(t, gjson.GetBytes(countBody, "tools.0.cache_control").Exists())
-	require.True(t, gjson.GetBytes(countBody, "tools.1.cache_control").Exists())
+	require.False(t, gjson.GetBytes(countBody, "tools.1.cache_control").Exists())
 	require.True(t, gjson.GetBytes(countBody, "tools.2.cache_control").Exists())
+	require.True(t, gjson.GetBytes(countBody, "tools.3.cache_control").Exists())
 }
 
 func TestGatewayService_AnthropicOAuth_NotAffectedByAPIKeyPassthroughToggle(t *testing.T) {
diff --git a/backend/internal/service/gateway_tool_rewrite.go b/backend/internal/service/gateway_tool_rewrite.go
index 42d786fe3eda..d1b69937232e 100644
--- a/backend/internal/service/gateway_tool_rewrite.go
+++ b/backend/internal/service/gateway_tool_rewrite.go
@@ -249,8 +249,9 @@ func applyToolNameRewriteToBody(body []byte, rw *ToolNameRewrite) []byte {
 }
 
 // applyToolsLastCacheBreakpoint 在最后一个非延迟加载工具上注入 cache_control
-// 断点。Anthropic 不允许 custom.defer_loading=true 的工具携带 cache_control，
-// 因此会先清理所有延迟加载工具上的客户端断点。其余行为对齐 Parrot
+// 断点。Anthropic 不允许 defer_loading=true 的工具携带 cache_control，
+// 因此会先清理所有延迟加载工具上的客户端断点。兼容官方顶层字段和
+// Claude Code 使用的 custom.defer_loading 字段。其余行为对齐 Parrot
 // `tools[-1]["cache_control"] = {"type":"ephemeral","ttl":"1h"}`，
 // 但 ttl 按本仓规则：
 //   - 客户端已为该 tool 显式设置 cache_control.ttl → 完全透传不覆盖
@@ -299,7 +300,8 @@ func applyToolsLastCacheBreakpoint(body []byte) []byte {
 }
 
 func isDeferredLoadingTool(tool gjson.Result) bool {
-	return tool.Get("custom.defer_loading").Type == gjson.True
+	return tool.Get("defer_loading").Type == gjson.True ||
+		tool.Get("custom.defer_loading").Type == gjson.True
 }
 
 // stripDeferredToolCacheControl removes the cache marker Anthropic rejects on
diff --git a/backend/internal/service/gateway_tool_rewrite_test.go b/backend/internal/service/gateway_tool_rewrite_test.go
index d35f0181a77a..c894ef708c63 100644
--- a/backend/internal/service/gateway_tool_rewrite_test.go
+++ b/backend/internal/service/gateway_tool_rewrite_test.go
@@ -159,7 +159,7 @@ func TestApplyToolsLastCacheBreakpoint_StripsDeferredToolCacheControl(t *testing
 }
 
 func TestApplyToolsLastCacheBreakpoint_SkipsDeferredFinalTool(t *testing.T) {
-	body := []byte(`{"tools":[{"name":"a","input_schema":{}},{"name":"b","custom":{"defer_loading":true}}]}`)
+	body := []byte(`{"tools":[{"name":"a","input_schema":{}},{"name":"b","defer_loading":true}]}`)
 	out := applyToolsLastCacheBreakpoint(body)
 
 	require.Equal(t, "ephemeral", gjson.GetBytes(out, "tools.0.cache_control.type").String())
@@ -168,11 +168,12 @@ func TestApplyToolsLastCacheBreakpoint_SkipsDeferredFinalTool(t *testing.T) {
 }
 
 func TestApplyToolsLastCacheBreakpoint_OnlyLiteralTrueIsDeferred(t *testing.T) {
-	body := []byte(`{"tools":[{"name":"true","custom":{"defer_loading":true},"cache_control":{"type":"ephemeral"}},{"name":"false","custom":{"defer_loading":false},"cache_control":{"type":"ephemeral"}},{"name":"string","custom":{"defer_loading":"true"},"cache_control":{"type":"ephemeral"}},{"name":"number","custom":{"defer_loading":1},"cache_control":{"type":"ephemeral"}},{"name":"object","custom":{"defer_loading":{}},"cache_control":{"type":"ephemeral"}}]}`)
+	body := []byte(`{"tools":[{"name":"custom-true","custom":{"defer_loading":true},"cache_control":{"type":"ephemeral"}},{"name":"top-level-true","defer_loading":true,"cache_control":{"type":"ephemeral"}},{"name":"false","defer_loading":false,"cache_control":{"type":"ephemeral"}},{"name":"string","defer_loading":"true","cache_control":{"type":"ephemeral"}},{"name":"number","defer_loading":1,"cache_control":{"type":"ephemeral"}},{"name":"object","defer_loading":{},"cache_control":{"type":"ephemeral"}}]}`)
 	out := stripDeferredToolCacheControl(body)
 
 	require.False(t, gjson.GetBytes(out, "tools.0.cache_control").Exists())
-	for idx := 1; idx < 5; idx++ {
+	require.False(t, gjson.GetBytes(out, "tools.1.cache_control").Exists())
+	for idx := 2; idx < 6; idx++ {
 		require.Equal(t, "ephemeral", gjson.GetBytes(out, fmt.Sprintf("tools.%d.cache_control.type", idx)).String())
 	}
 }

From 2321c0e771a7638127f9931cd964ad5d84ab5277 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Wed, 29 Jul 2026 16:53:44 +0800
Subject: [PATCH 09/68] chore: retry CI checks


From 1afb8264e53c4952cdd1a860a3605351e5ac330a Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Wed, 29 Jul 2026 23:25:08 +0800
Subject: [PATCH 10/68] fix(lint): use require.NotNil for staticcheck SA5011

Replace if-then-t.Fatal nil checks with testify require.NotNil so
staticcheck can prove the pointer is non-nil before field access.
Fixes SA5011 in payment_handler_test, parse_test, and
openai_images_incomplete_test.
---
 backend/internal/handler/admin/payment_handler_test.go    | 5 ++---
 backend/internal/pkg/proxyurl/parse_test.go               | 6 +++---
 backend/internal/service/openai_images_incomplete_test.go | 5 ++---
 3 files changed, 7 insertions(+), 9 deletions(-)

diff --git a/backend/internal/handler/admin/payment_handler_test.go b/backend/internal/handler/admin/payment_handler_test.go
index d0752ec09a6e..c72132382ffb 100644
--- a/backend/internal/handler/admin/payment_handler_test.go
+++ b/backend/internal/handler/admin/payment_handler_test.go
@@ -8,6 +8,7 @@ import (
 
 	dbent "github.com/Wei-Shaw/sub2api/ent"
 	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/stretchr/testify/require"
 )
 
 func TestSanitizeAdminPaymentOrderForResponseAddsCurrency(t *testing.T) {
@@ -32,9 +33,7 @@ func TestSanitizeAdminPaymentOrderForResponseAddsCurrency(t *testing.T) {
 	}
 
 	got := sanitizeAdminPaymentOrderForResponse(order)
-	if got == nil {
-		t.Fatal("expected sanitized order")
-	}
+	require.NotNil(t, got)
 	if got.Currency != "USD" {
 		t.Fatalf("expected currency USD, got %q", got.Currency)
 	}
diff --git a/backend/internal/pkg/proxyurl/parse_test.go b/backend/internal/pkg/proxyurl/parse_test.go
index 5fb57c16f7bb..05fde1284db7 100644
--- a/backend/internal/pkg/proxyurl/parse_test.go
+++ b/backend/internal/pkg/proxyurl/parse_test.go
@@ -3,6 +3,8 @@ package proxyurl
 import (
 	"strings"
 	"testing"
+
+	"github.com/stretchr/testify/require"
 )
 
 func TestParse_空字符串直连(t *testing.T) {
@@ -39,9 +41,7 @@ func TestParse_有效HTTP代理(t *testing.T) {
 	if trimmed != "http://proxy.example.com:8080" {
 		t.Errorf("trimmed 不匹配: got %q", trimmed)
 	}
-	if parsed == nil {
-		t.Fatal("parsed 不应为 nil")
-	}
+	require.NotNil(t, parsed)
 	if parsed.Host != "proxy.example.com:8080" {
 		t.Errorf("Host 不匹配: got %q", parsed.Host)
 	}
diff --git a/backend/internal/service/openai_images_incomplete_test.go b/backend/internal/service/openai_images_incomplete_test.go
index 6a8ba56225d0..aa2f92b42d66 100644
--- a/backend/internal/service/openai_images_incomplete_test.go
+++ b/backend/internal/service/openai_images_incomplete_test.go
@@ -9,6 +9,7 @@ import (
 	"testing"
 
 	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
 )
 
 // response.incomplete（生成超时/截断）应被识别为可重试的 502 上游错误，触发 failover。
@@ -16,9 +17,7 @@ func TestExtractImagesUpstreamError_IncompleteIsRetryable(t *testing.T) {
 	body := "data: {\"type\":\"response.created\",\"response\":{\"id\":\"resp_1\"}}\n\n" +
 		"data: {\"type\":\"response.incomplete\",\"response\":{\"id\":\"resp_1\",\"status\":\"incomplete\",\"incomplete_details\":{\"reason\":\"max_output_tokens\"}}}\n\n"
 	got := extractOpenAIImagesUpstreamError([]byte(body))
-	if got == nil {
-		t.Fatal("incomplete event should produce an upstream error, got nil")
-	}
+	require.NotNil(t, got)
 	if got.StatusCode != http.StatusBadGateway {
 		t.Fatalf("incomplete(max_output_tokens) should be 502 retryable, got %d", got.StatusCode)
 	}

From 5ea03c178df54d0e8d7346c72935df6633fb634e Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Thu, 30 Jul 2026 00:06:44 +0800
Subject: [PATCH 11/68] fix(lint): resolve remaining nil dereference warnings

---
 backend/internal/service/openai_images_incomplete_test.go | 4 +---
 backend/internal/service/ops_service_user_error_test.go   | 5 ++---
 backend/internal/service/ops_user_error_test.go           | 6 +++---
 3 files changed, 6 insertions(+), 9 deletions(-)

diff --git a/backend/internal/service/openai_images_incomplete_test.go b/backend/internal/service/openai_images_incomplete_test.go
index aa2f92b42d66..e707fbf75844 100644
--- a/backend/internal/service/openai_images_incomplete_test.go
+++ b/backend/internal/service/openai_images_incomplete_test.go
@@ -36,9 +36,7 @@ func TestExtractImagesUpstreamError_IncompleteIsRetryable(t *testing.T) {
 func TestExtractImagesUpstreamError_IncompleteContentFilterNotRetryable(t *testing.T) {
 	body := "data: {\"type\":\"response.incomplete\",\"response\":{\"id\":\"r\",\"status\":\"incomplete\",\"incomplete_details\":{\"reason\":\"content_filter\"}}}\n\n"
 	got := extractOpenAIImagesUpstreamError([]byte(body))
-	if got == nil {
-		t.Fatal("content_filter incomplete should produce error")
-	}
+	require.NotNil(t, got)
 	if got.StatusCode != http.StatusBadRequest {
 		t.Fatalf("content_filter should be 400 (non-retryable), got %d", got.StatusCode)
 	}
diff --git a/backend/internal/service/ops_service_user_error_test.go b/backend/internal/service/ops_service_user_error_test.go
index b1f6152702af..3fa88739ad6a 100644
--- a/backend/internal/service/ops_service_user_error_test.go
+++ b/backend/internal/service/ops_service_user_error_test.go
@@ -6,6 +6,7 @@ import (
 	"testing"
 
 	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
+	"github.com/stretchr/testify/require"
 )
 
 type stubOpsRepoForUserErr struct {
@@ -119,9 +120,7 @@ func TestGetUserErrorRequestDetail_OwnershipEnforced(t *testing.T) {
 	if err2 != nil {
 		t.Fatalf("expected no error for legitimate access, got %v", err2)
 	}
-	if got2 == nil {
-		t.Fatal("expected non-nil detail for legitimate access")
-	}
+	require.NotNil(t, got2)
 	if got2.ID != 42 {
 		t.Errorf("want ID=42, got %d", got2.ID)
 	}
diff --git a/backend/internal/service/ops_user_error_test.go b/backend/internal/service/ops_user_error_test.go
index c72cf5dbfb60..e02842d3d21f 100644
--- a/backend/internal/service/ops_user_error_test.go
+++ b/backend/internal/service/ops_user_error_test.go
@@ -5,6 +5,8 @@ import (
 	"strings"
 	"testing"
 	"time"
+
+	"github.com/stretchr/testify/require"
 )
 
 func TestMapUserErrorCategory(t *testing.T) {
@@ -132,9 +134,7 @@ func TestToUserErrorRequestDetail_WhitelistAndRedacts(t *testing.T) {
 	}
 
 	out := ToUserErrorRequestDetail(src)
-	if out == nil {
-		t.Fatal("expected non-nil detail")
-	}
+	require.NotNil(t, out)
 
 	// 基础字段正确映射
 	if out.ID != 999 {

From 145b0ac35e06b168cddfab754b074afb20d234d5 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Thu, 30 Jul 2026 00:28:14 +0800
Subject: [PATCH 12/68] fix(lint): make remaining pointer assertions explicit

---
 .../internal/service/payment_order_result_test.go   | 13 ++++---------
 .../service/ratelimit_service_anthropic_test.go     | 10 ++++------
 2 files changed, 8 insertions(+), 15 deletions(-)

diff --git a/backend/internal/service/payment_order_result_test.go b/backend/internal/service/payment_order_result_test.go
index e77fbce4822c..ce405e0a23dd 100644
--- a/backend/internal/service/payment_order_result_test.go
+++ b/backend/internal/service/payment_order_result_test.go
@@ -9,6 +9,7 @@ import (
 	dbent "github.com/Wei-Shaw/sub2api/ent"
 	"github.com/Wei-Shaw/sub2api/internal/payment"
 	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
+	"github.com/stretchr/testify/require"
 )
 
 func TestShouldUseAlipayMobilePrecreate(t *testing.T) {
@@ -389,15 +390,11 @@ func TestMaybeBuildWeChatOAuthRequiredResponse(t *testing.T) {
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}
-	if resp == nil {
-		t.Fatal("expected oauth_required response, got nil")
-	}
+	require.NotNil(t, resp)
 	if resp.ResultType != payment.CreatePaymentResultOAuthRequired {
 		t.Fatalf("result type = %q, want %q", resp.ResultType, payment.CreatePaymentResultOAuthRequired)
 	}
-	if resp.OAuth == nil {
-		t.Fatal("expected oauth payload, got nil")
-	}
+	require.NotNil(t, resp.OAuth)
 	if resp.OAuth.AppID != "wx123456" {
 		t.Fatalf("appid = %q, want %q", resp.OAuth.AppID, "wx123456")
 	}
@@ -503,9 +500,7 @@ func TestMaybeBuildWeChatOAuthRequiredResponseFallsBackToConfiguredLegacySigning
 	if err != nil {
 		t.Fatalf("expected nil error, got %v", err)
 	}
-	if resp == nil {
-		t.Fatal("expected oauth-required response, got nil")
-	}
+	require.NotNil(t, resp)
 	if resp.ResultType != payment.CreatePaymentResultOAuthRequired {
 		t.Fatalf("result type = %q, want %q", resp.ResultType, payment.CreatePaymentResultOAuthRequired)
 	}
diff --git a/backend/internal/service/ratelimit_service_anthropic_test.go b/backend/internal/service/ratelimit_service_anthropic_test.go
index 0e75b4f914c0..c02ad00e1fd4 100644
--- a/backend/internal/service/ratelimit_service_anthropic_test.go
+++ b/backend/internal/service/ratelimit_service_anthropic_test.go
@@ -5,6 +5,8 @@ import (
 	"strconv"
 	"testing"
 	"time"
+
+	"github.com/stretchr/testify/require"
 )
 
 func TestCalculateAnthropic429ResetTime_Only5hExceeded(t *testing.T) {
@@ -193,9 +195,7 @@ func TestSelectAnthropicFableWindowLimit_RejectedStatus(t *testing.T) {
 	headers.Set("anthropic-ratelimit-unified-7d_oi-reset", strconv.FormatInt(reset.Unix(), 10))
 
 	limit := selectAnthropicFableWindowLimit(headers, now)
-	if limit == nil {
-		t.Fatal("expected non-nil limit")
-	}
+	require.NotNil(t, limit)
 	if !limit.resetAt.Equal(reset) {
 		t.Errorf("expected resetAt=%v, got %v", reset, limit.resetAt)
 	}
@@ -214,9 +214,7 @@ func TestSelectAnthropicFableWindowLimit_UtilizationOnly(t *testing.T) {
 	headers.Set("anthropic-ratelimit-unified-7d_oi-reset", strconv.FormatInt(reset.Unix(), 10))
 
 	limit := selectAnthropicFableWindowLimit(headers, now)
-	if limit == nil {
-		t.Fatal("expected non-nil limit")
-	}
+	require.NotNil(t, limit)
 	if !limit.resetAt.Equal(reset) {
 		t.Errorf("expected resetAt=%v, got %v", reset, limit.resetAt)
 	}

From e087f1b73699d31a1077b03e10dae0e8c3992a9c Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Thu, 30 Jul 2026 00:49:48 +0800
Subject: [PATCH 13/68] fix(lint): use fatal pointer assertions in usage tests

---
 .../internal/service/account_usage_service_test.go | 14 +++++---------
 .../service/ratelimit_service_anthropic_test.go    |  4 +---
 2 files changed, 6 insertions(+), 12 deletions(-)

diff --git a/backend/internal/service/account_usage_service_test.go b/backend/internal/service/account_usage_service_test.go
index ca153a3c4245..6d7fa8fb5f36 100644
--- a/backend/internal/service/account_usage_service_test.go
+++ b/backend/internal/service/account_usage_service_test.go
@@ -5,6 +5,8 @@ import (
 	"net/http"
 	"testing"
 	"time"
+
+	"github.com/stretchr/testify/require"
 )
 
 type accountUsageCodexProbeRepo struct {
@@ -216,9 +218,7 @@ func TestBuildCodexUsageProgressFromExtra_ZerosExpiredWindow(t *testing.T) {
 			"codex_5h_reset_at":     "2026-03-16T10:00:00Z", // 2h ago
 		}
 		progress := buildCodexUsageProgressFromExtra(extra, "5h", now)
-		if progress == nil {
-			t.Fatal("expected non-nil progress")
-		}
+		require.NotNil(t, progress)
 		if progress.Utilization != 0 {
 			t.Fatalf("expected Utilization=0 for expired window, got %v", progress.Utilization)
 		}
@@ -234,9 +234,7 @@ func TestBuildCodexUsageProgressFromExtra_ZerosExpiredWindow(t *testing.T) {
 			"codex_5h_reset_at":     resetAt,
 		}
 		progress := buildCodexUsageProgressFromExtra(extra, "5h", now)
-		if progress == nil {
-			t.Fatal("expected non-nil progress")
-		}
+		require.NotNil(t, progress)
 		if progress.Utilization != 42.0 {
 			t.Fatalf("expected Utilization=42, got %v", progress.Utilization)
 		}
@@ -248,9 +246,7 @@ func TestBuildCodexUsageProgressFromExtra_ZerosExpiredWindow(t *testing.T) {
 			"codex_7d_reset_at":     "2026-03-15T00:00:00Z", // yesterday
 		}
 		progress := buildCodexUsageProgressFromExtra(extra, "7d", now)
-		if progress == nil {
-			t.Fatal("expected non-nil progress")
-		}
+		require.NotNil(t, progress)
 		if progress.Utilization != 0 {
 			t.Fatalf("expected Utilization=0 for expired 7d window, got %v", progress.Utilization)
 		}
diff --git a/backend/internal/service/ratelimit_service_anthropic_test.go b/backend/internal/service/ratelimit_service_anthropic_test.go
index c02ad00e1fd4..0755a02565c9 100644
--- a/backend/internal/service/ratelimit_service_anthropic_test.go
+++ b/backend/internal/service/ratelimit_service_anthropic_test.go
@@ -248,9 +248,7 @@ func TestSelectAnthropicFableWindowLimit_FallsBackToAggregateReset(t *testing.T)
 	headers.Set("anthropic-ratelimit-unified-reset", strconv.FormatInt(reset.Unix(), 10))
 
 	limit := selectAnthropicFableWindowLimit(headers, now)
-	if limit == nil {
-		t.Fatal("expected non-nil limit via aggregate reset fallback")
-	}
+	require.NotNil(t, limit)
 	if !limit.resetAt.Equal(reset) {
 		t.Errorf("expected resetAt=%v, got %v", reset, limit.resetAt)
 	}

From c3fc4331c1b150998a9b2dd9dfc3fe5d4d4d64bf Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Thu, 30 Jul 2026 01:12:42 +0800
Subject: [PATCH 14/68] fix(lint): make compatibility test assertions fatal

---
 .../service/gemini_messages_compat_service_test.go     |  4 +---
 .../openai_gateway_service_codex_snapshot_test.go      | 10 ++++------
 2 files changed, 5 insertions(+), 9 deletions(-)

diff --git a/backend/internal/service/gemini_messages_compat_service_test.go b/backend/internal/service/gemini_messages_compat_service_test.go
index 59f83a683541..acf9075c5477 100644
--- a/backend/internal/service/gemini_messages_compat_service_test.go
+++ b/backend/internal/service/gemini_messages_compat_service_test.go
@@ -782,9 +782,7 @@ func TestExtractGeminiUsage(t *testing.T) {
 				}
 				return
 			}
-			if got == nil {
-				t.Fatalf("期望返回非 nil，实际返回 nil")
-			}
+			require.NotNil(t, got)
 			if got.InputTokens != tt.wantUsage.InputTokens {
 				t.Errorf("InputTokens: 期望 %d，实际 %d", tt.wantUsage.InputTokens, got.InputTokens)
 			}
diff --git a/backend/internal/service/openai_gateway_service_codex_snapshot_test.go b/backend/internal/service/openai_gateway_service_codex_snapshot_test.go
index 27208b58d921..59568b33c9d4 100644
--- a/backend/internal/service/openai_gateway_service_codex_snapshot_test.go
+++ b/backend/internal/service/openai_gateway_service_codex_snapshot_test.go
@@ -3,6 +3,8 @@ package service
 import (
 	"testing"
 	"time"
+
+	"github.com/stretchr/testify/require"
 )
 
 func TestCodexSnapshotBaseTime(t *testing.T) {
@@ -50,9 +52,7 @@ func TestCodexResetAtRFC3339(t *testing.T) {
 	t.Run("positive seconds", func(t *testing.T) {
 		sec := 90
 		got := codexResetAtRFC3339(base, &sec)
-		if got == nil {
-			t.Fatal("expected non-nil")
-		}
+		require.NotNil(t, got)
 		if *got != "2026-02-16T10:01:30Z" {
 			t.Fatalf("got %s, want %s", *got, "2026-02-16T10:01:30Z")
 		}
@@ -61,9 +61,7 @@ func TestCodexResetAtRFC3339(t *testing.T) {
 	t.Run("negative seconds clamp to base", func(t *testing.T) {
 		sec := -3
 		got := codexResetAtRFC3339(base, &sec)
-		if got == nil {
-			t.Fatal("expected non-nil")
-		}
+		require.NotNil(t, got)
 		if *got != "2026-02-16T10:00:00Z" {
 			t.Fatalf("got %s, want %s", *got, "2026-02-16T10:00:00Z")
 		}

From c91fbeb0c008b2323e98ca5b73cedf074c8543dd Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Tue, 11 Aug 2026 16:01:06 +0800
Subject: [PATCH 15/68] chore: remove unrelated test refactors

---
 .../internal/handler/admin/payment_handler_test.go |  5 +++--
 backend/internal/pkg/proxyurl/parse_test.go        |  6 +++---
 .../internal/service/account_usage_service_test.go | 14 +++++++++-----
 .../service/gemini_messages_compat_service_test.go |  4 +++-
 .../openai_gateway_service_codex_snapshot_test.go  | 10 ++++++----
 .../service/openai_images_incomplete_test.go       |  9 ++++++---
 .../service/ops_service_user_error_test.go         |  5 +++--
 backend/internal/service/ops_user_error_test.go    |  6 +++---
 .../internal/service/payment_order_result_test.go  | 13 +++++++++----
 .../service/ratelimit_service_anthropic_test.go    | 14 +++++++++-----
 10 files changed, 54 insertions(+), 32 deletions(-)

diff --git a/backend/internal/handler/admin/payment_handler_test.go b/backend/internal/handler/admin/payment_handler_test.go
index c72132382ffb..d0752ec09a6e 100644
--- a/backend/internal/handler/admin/payment_handler_test.go
+++ b/backend/internal/handler/admin/payment_handler_test.go
@@ -8,7 +8,6 @@ import (
 
 	dbent "github.com/Wei-Shaw/sub2api/ent"
 	"github.com/Wei-Shaw/sub2api/internal/service"
-	"github.com/stretchr/testify/require"
 )
 
 func TestSanitizeAdminPaymentOrderForResponseAddsCurrency(t *testing.T) {
@@ -33,7 +32,9 @@ func TestSanitizeAdminPaymentOrderForResponseAddsCurrency(t *testing.T) {
 	}
 
 	got := sanitizeAdminPaymentOrderForResponse(order)
-	require.NotNil(t, got)
+	if got == nil {
+		t.Fatal("expected sanitized order")
+	}
 	if got.Currency != "USD" {
 		t.Fatalf("expected currency USD, got %q", got.Currency)
 	}
diff --git a/backend/internal/pkg/proxyurl/parse_test.go b/backend/internal/pkg/proxyurl/parse_test.go
index 05fde1284db7..5fb57c16f7bb 100644
--- a/backend/internal/pkg/proxyurl/parse_test.go
+++ b/backend/internal/pkg/proxyurl/parse_test.go
@@ -3,8 +3,6 @@ package proxyurl
 import (
 	"strings"
 	"testing"
-
-	"github.com/stretchr/testify/require"
 )
 
 func TestParse_空字符串直连(t *testing.T) {
@@ -41,7 +39,9 @@ func TestParse_有效HTTP代理(t *testing.T) {
 	if trimmed != "http://proxy.example.com:8080" {
 		t.Errorf("trimmed 不匹配: got %q", trimmed)
 	}
-	require.NotNil(t, parsed)
+	if parsed == nil {
+		t.Fatal("parsed 不应为 nil")
+	}
 	if parsed.Host != "proxy.example.com:8080" {
 		t.Errorf("Host 不匹配: got %q", parsed.Host)
 	}
diff --git a/backend/internal/service/account_usage_service_test.go b/backend/internal/service/account_usage_service_test.go
index 6d7fa8fb5f36..ca153a3c4245 100644
--- a/backend/internal/service/account_usage_service_test.go
+++ b/backend/internal/service/account_usage_service_test.go
@@ -5,8 +5,6 @@ import (
 	"net/http"
 	"testing"
 	"time"
-
-	"github.com/stretchr/testify/require"
 )
 
 type accountUsageCodexProbeRepo struct {
@@ -218,7 +216,9 @@ func TestBuildCodexUsageProgressFromExtra_ZerosExpiredWindow(t *testing.T) {
 			"codex_5h_reset_at":     "2026-03-16T10:00:00Z", // 2h ago
 		}
 		progress := buildCodexUsageProgressFromExtra(extra, "5h", now)
-		require.NotNil(t, progress)
+		if progress == nil {
+			t.Fatal("expected non-nil progress")
+		}
 		if progress.Utilization != 0 {
 			t.Fatalf("expected Utilization=0 for expired window, got %v", progress.Utilization)
 		}
@@ -234,7 +234,9 @@ func TestBuildCodexUsageProgressFromExtra_ZerosExpiredWindow(t *testing.T) {
 			"codex_5h_reset_at":     resetAt,
 		}
 		progress := buildCodexUsageProgressFromExtra(extra, "5h", now)
-		require.NotNil(t, progress)
+		if progress == nil {
+			t.Fatal("expected non-nil progress")
+		}
 		if progress.Utilization != 42.0 {
 			t.Fatalf("expected Utilization=42, got %v", progress.Utilization)
 		}
@@ -246,7 +248,9 @@ func TestBuildCodexUsageProgressFromExtra_ZerosExpiredWindow(t *testing.T) {
 			"codex_7d_reset_at":     "2026-03-15T00:00:00Z", // yesterday
 		}
 		progress := buildCodexUsageProgressFromExtra(extra, "7d", now)
-		require.NotNil(t, progress)
+		if progress == nil {
+			t.Fatal("expected non-nil progress")
+		}
 		if progress.Utilization != 0 {
 			t.Fatalf("expected Utilization=0 for expired 7d window, got %v", progress.Utilization)
 		}
diff --git a/backend/internal/service/gemini_messages_compat_service_test.go b/backend/internal/service/gemini_messages_compat_service_test.go
index acf9075c5477..59f83a683541 100644
--- a/backend/internal/service/gemini_messages_compat_service_test.go
+++ b/backend/internal/service/gemini_messages_compat_service_test.go
@@ -782,7 +782,9 @@ func TestExtractGeminiUsage(t *testing.T) {
 				}
 				return
 			}
-			require.NotNil(t, got)
+			if got == nil {
+				t.Fatalf("期望返回非 nil，实际返回 nil")
+			}
 			if got.InputTokens != tt.wantUsage.InputTokens {
 				t.Errorf("InputTokens: 期望 %d，实际 %d", tt.wantUsage.InputTokens, got.InputTokens)
 			}
diff --git a/backend/internal/service/openai_gateway_service_codex_snapshot_test.go b/backend/internal/service/openai_gateway_service_codex_snapshot_test.go
index 59568b33c9d4..27208b58d921 100644
--- a/backend/internal/service/openai_gateway_service_codex_snapshot_test.go
+++ b/backend/internal/service/openai_gateway_service_codex_snapshot_test.go
@@ -3,8 +3,6 @@ package service
 import (
 	"testing"
 	"time"
-
-	"github.com/stretchr/testify/require"
 )
 
 func TestCodexSnapshotBaseTime(t *testing.T) {
@@ -52,7 +50,9 @@ func TestCodexResetAtRFC3339(t *testing.T) {
 	t.Run("positive seconds", func(t *testing.T) {
 		sec := 90
 		got := codexResetAtRFC3339(base, &sec)
-		require.NotNil(t, got)
+		if got == nil {
+			t.Fatal("expected non-nil")
+		}
 		if *got != "2026-02-16T10:01:30Z" {
 			t.Fatalf("got %s, want %s", *got, "2026-02-16T10:01:30Z")
 		}
@@ -61,7 +61,9 @@ func TestCodexResetAtRFC3339(t *testing.T) {
 	t.Run("negative seconds clamp to base", func(t *testing.T) {
 		sec := -3
 		got := codexResetAtRFC3339(base, &sec)
-		require.NotNil(t, got)
+		if got == nil {
+			t.Fatal("expected non-nil")
+		}
 		if *got != "2026-02-16T10:00:00Z" {
 			t.Fatalf("got %s, want %s", *got, "2026-02-16T10:00:00Z")
 		}
diff --git a/backend/internal/service/openai_images_incomplete_test.go b/backend/internal/service/openai_images_incomplete_test.go
index e707fbf75844..6a8ba56225d0 100644
--- a/backend/internal/service/openai_images_incomplete_test.go
+++ b/backend/internal/service/openai_images_incomplete_test.go
@@ -9,7 +9,6 @@ import (
 	"testing"
 
 	"github.com/gin-gonic/gin"
-	"github.com/stretchr/testify/require"
 )
 
 // response.incomplete（生成超时/截断）应被识别为可重试的 502 上游错误，触发 failover。
@@ -17,7 +16,9 @@ func TestExtractImagesUpstreamError_IncompleteIsRetryable(t *testing.T) {
 	body := "data: {\"type\":\"response.created\",\"response\":{\"id\":\"resp_1\"}}\n\n" +
 		"data: {\"type\":\"response.incomplete\",\"response\":{\"id\":\"resp_1\",\"status\":\"incomplete\",\"incomplete_details\":{\"reason\":\"max_output_tokens\"}}}\n\n"
 	got := extractOpenAIImagesUpstreamError([]byte(body))
-	require.NotNil(t, got)
+	if got == nil {
+		t.Fatal("incomplete event should produce an upstream error, got nil")
+	}
 	if got.StatusCode != http.StatusBadGateway {
 		t.Fatalf("incomplete(max_output_tokens) should be 502 retryable, got %d", got.StatusCode)
 	}
@@ -36,7 +37,9 @@ func TestExtractImagesUpstreamError_IncompleteIsRetryable(t *testing.T) {
 func TestExtractImagesUpstreamError_IncompleteContentFilterNotRetryable(t *testing.T) {
 	body := "data: {\"type\":\"response.incomplete\",\"response\":{\"id\":\"r\",\"status\":\"incomplete\",\"incomplete_details\":{\"reason\":\"content_filter\"}}}\n\n"
 	got := extractOpenAIImagesUpstreamError([]byte(body))
-	require.NotNil(t, got)
+	if got == nil {
+		t.Fatal("content_filter incomplete should produce error")
+	}
 	if got.StatusCode != http.StatusBadRequest {
 		t.Fatalf("content_filter should be 400 (non-retryable), got %d", got.StatusCode)
 	}
diff --git a/backend/internal/service/ops_service_user_error_test.go b/backend/internal/service/ops_service_user_error_test.go
index 3fa88739ad6a..b1f6152702af 100644
--- a/backend/internal/service/ops_service_user_error_test.go
+++ b/backend/internal/service/ops_service_user_error_test.go
@@ -6,7 +6,6 @@ import (
 	"testing"
 
 	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
-	"github.com/stretchr/testify/require"
 )
 
 type stubOpsRepoForUserErr struct {
@@ -120,7 +119,9 @@ func TestGetUserErrorRequestDetail_OwnershipEnforced(t *testing.T) {
 	if err2 != nil {
 		t.Fatalf("expected no error for legitimate access, got %v", err2)
 	}
-	require.NotNil(t, got2)
+	if got2 == nil {
+		t.Fatal("expected non-nil detail for legitimate access")
+	}
 	if got2.ID != 42 {
 		t.Errorf("want ID=42, got %d", got2.ID)
 	}
diff --git a/backend/internal/service/ops_user_error_test.go b/backend/internal/service/ops_user_error_test.go
index e02842d3d21f..c72cf5dbfb60 100644
--- a/backend/internal/service/ops_user_error_test.go
+++ b/backend/internal/service/ops_user_error_test.go
@@ -5,8 +5,6 @@ import (
 	"strings"
 	"testing"
 	"time"
-
-	"github.com/stretchr/testify/require"
 )
 
 func TestMapUserErrorCategory(t *testing.T) {
@@ -134,7 +132,9 @@ func TestToUserErrorRequestDetail_WhitelistAndRedacts(t *testing.T) {
 	}
 
 	out := ToUserErrorRequestDetail(src)
-	require.NotNil(t, out)
+	if out == nil {
+		t.Fatal("expected non-nil detail")
+	}
 
 	// 基础字段正确映射
 	if out.ID != 999 {
diff --git a/backend/internal/service/payment_order_result_test.go b/backend/internal/service/payment_order_result_test.go
index ce405e0a23dd..e77fbce4822c 100644
--- a/backend/internal/service/payment_order_result_test.go
+++ b/backend/internal/service/payment_order_result_test.go
@@ -9,7 +9,6 @@ import (
 	dbent "github.com/Wei-Shaw/sub2api/ent"
 	"github.com/Wei-Shaw/sub2api/internal/payment"
 	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
-	"github.com/stretchr/testify/require"
 )
 
 func TestShouldUseAlipayMobilePrecreate(t *testing.T) {
@@ -390,11 +389,15 @@ func TestMaybeBuildWeChatOAuthRequiredResponse(t *testing.T) {
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}
-	require.NotNil(t, resp)
+	if resp == nil {
+		t.Fatal("expected oauth_required response, got nil")
+	}
 	if resp.ResultType != payment.CreatePaymentResultOAuthRequired {
 		t.Fatalf("result type = %q, want %q", resp.ResultType, payment.CreatePaymentResultOAuthRequired)
 	}
-	require.NotNil(t, resp.OAuth)
+	if resp.OAuth == nil {
+		t.Fatal("expected oauth payload, got nil")
+	}
 	if resp.OAuth.AppID != "wx123456" {
 		t.Fatalf("appid = %q, want %q", resp.OAuth.AppID, "wx123456")
 	}
@@ -500,7 +503,9 @@ func TestMaybeBuildWeChatOAuthRequiredResponseFallsBackToConfiguredLegacySigning
 	if err != nil {
 		t.Fatalf("expected nil error, got %v", err)
 	}
-	require.NotNil(t, resp)
+	if resp == nil {
+		t.Fatal("expected oauth-required response, got nil")
+	}
 	if resp.ResultType != payment.CreatePaymentResultOAuthRequired {
 		t.Fatalf("result type = %q, want %q", resp.ResultType, payment.CreatePaymentResultOAuthRequired)
 	}
diff --git a/backend/internal/service/ratelimit_service_anthropic_test.go b/backend/internal/service/ratelimit_service_anthropic_test.go
index 0755a02565c9..0e75b4f914c0 100644
--- a/backend/internal/service/ratelimit_service_anthropic_test.go
+++ b/backend/internal/service/ratelimit_service_anthropic_test.go
@@ -5,8 +5,6 @@ import (
 	"strconv"
 	"testing"
 	"time"
-
-	"github.com/stretchr/testify/require"
 )
 
 func TestCalculateAnthropic429ResetTime_Only5hExceeded(t *testing.T) {
@@ -195,7 +193,9 @@ func TestSelectAnthropicFableWindowLimit_RejectedStatus(t *testing.T) {
 	headers.Set("anthropic-ratelimit-unified-7d_oi-reset", strconv.FormatInt(reset.Unix(), 10))
 
 	limit := selectAnthropicFableWindowLimit(headers, now)
-	require.NotNil(t, limit)
+	if limit == nil {
+		t.Fatal("expected non-nil limit")
+	}
 	if !limit.resetAt.Equal(reset) {
 		t.Errorf("expected resetAt=%v, got %v", reset, limit.resetAt)
 	}
@@ -214,7 +214,9 @@ func TestSelectAnthropicFableWindowLimit_UtilizationOnly(t *testing.T) {
 	headers.Set("anthropic-ratelimit-unified-7d_oi-reset", strconv.FormatInt(reset.Unix(), 10))
 
 	limit := selectAnthropicFableWindowLimit(headers, now)
-	require.NotNil(t, limit)
+	if limit == nil {
+		t.Fatal("expected non-nil limit")
+	}
 	if !limit.resetAt.Equal(reset) {
 		t.Errorf("expected resetAt=%v, got %v", reset, limit.resetAt)
 	}
@@ -248,7 +250,9 @@ func TestSelectAnthropicFableWindowLimit_FallsBackToAggregateReset(t *testing.T)
 	headers.Set("anthropic-ratelimit-unified-reset", strconv.FormatInt(reset.Unix(), 10))
 
 	limit := selectAnthropicFableWindowLimit(headers, now)
-	require.NotNil(t, limit)
+	if limit == nil {
+		t.Fatal("expected non-nil limit via aggregate reset fallback")
+	}
 	if !limit.resetAt.Equal(reset) {
 		t.Errorf("expected resetAt=%v, got %v", reset, limit.resetAt)
 	}

From 76a13a5a8d56dca2befb8a3d70daf356c540f336 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Thu, 13 Aug 2026 02:43:03 +0800
Subject: [PATCH 16/68] fix(gateway): handle Anthropic SSE overload errors

---
 backend/internal/service/gateway_forward.go   |  18 ++-
 .../gateway_forward_partial_usage_test.go     | 114 ++++++++++++++++++
 2 files changed, 128 insertions(+), 4 deletions(-)

diff --git a/backend/internal/service/gateway_forward.go b/backend/internal/service/gateway_forward.go
index d10cb1ea3be5..863e4d2e8340 100644
--- a/backend/internal/service/gateway_forward.go
+++ b/backend/internal/service/gateway_forward.go
@@ -14,6 +14,7 @@ import (
 
 	"github.com/Wei-Shaw/sub2api/internal/pkg/claude"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/tidwall/gjson"
 
 	"github.com/gin-gonic/gin"
 )
@@ -800,14 +801,23 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 	var firstTokenMs *int
 	var clientDisconnect bool
 	if reqStream {
+		writerSizeBeforeStream := c.Writer.Size()
 		streamResult, err := s.handleStreamingResponse(ctx, resp, c, account, startTime, originalModel, reqModel, shouldMimicClaudeCode)
 		if err != nil {
 			var sseErr *sseStreamErrorEventError
 			if errors.As(err, &sseErr) {
 				// 上游 HTTP 200 + SSE 流体内出现 event:error 帧。
-				// 保留 StatusCode=403 以兼容既有 failover/客户端响应语义，
-				// 但补全 ResponseBody 与 ops 上下文，让运维日志能反映上游真实错误。
 				body := []byte(sseErr.RawData)
+				semanticStatus := http.StatusForbidden
+				if c.Writer.Size() == writerSizeBeforeStream && gjson.GetBytes(body, "error.type").String() == "overloaded_error" {
+					semanticStatus = 529
+					syntheticResp := &http.Response{
+						StatusCode: semanticStatus,
+						Header:     resp.Header.Clone(),
+						Body:       io.NopCloser(bytes.NewReader(body)),
+					}
+					s.handleFailoverSideEffects(ctx, syntheticResp, account, reqModel)
+				}
 
 				upstreamMsg := sanitizeUpstreamErrorMessage(
 					strings.TrimSpace(extractUpstreamErrorMessage(body)),
@@ -826,7 +836,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 					Platform:           account.Platform,
 					AccountID:          account.ID,
 					AccountName:        account.Name,
-					UpstreamStatusCode: 403,
+					UpstreamStatusCode: semanticStatus,
 					UpstreamRequestID:  resp.Header.Get("x-request-id"),
 					Kind:               "stream_error",
 					Message:            upstreamMsg,
@@ -840,7 +850,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 				)
 
 				return nil, &UpstreamFailoverError{
-					StatusCode:   403,
+					StatusCode:   semanticStatus,
 					ResponseBody: body,
 				}
 			}
diff --git a/backend/internal/service/gateway_forward_partial_usage_test.go b/backend/internal/service/gateway_forward_partial_usage_test.go
index f6701ca8b008..2b7666378e74 100644
--- a/backend/internal/service/gateway_forward_partial_usage_test.go
+++ b/backend/internal/service/gateway_forward_partial_usage_test.go
@@ -8,12 +8,37 @@ import (
 	"net/http/httptest"
 	"strings"
 	"testing"
+	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/config"
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/require"
 )
 
+type gatewayForwardErrorPolicyRepoStub struct {
+	AccountRepository
+	tempCalls           int
+	modelRateLimitCalls []gatewayForwardModelRateLimitCall
+}
+
+type gatewayForwardModelRateLimitCall struct {
+	accountID int64
+	scope     string
+}
+
+func (r *gatewayForwardErrorPolicyRepoStub) SetTempUnschedulable(context.Context, int64, time.Time, string) error {
+	r.tempCalls++
+	return nil
+}
+
+func (r *gatewayForwardErrorPolicyRepoStub) SetModelRateLimit(_ context.Context, id int64, scope string, _ time.Time, _ ...string) error {
+	r.modelRateLimitCalls = append(r.modelRateLimitCalls, gatewayForwardModelRateLimitCall{
+		accountID: id,
+		scope:     scope,
+	})
+	return nil
+}
+
 // 本文件覆盖 issue #5148：流式转发中途出错（缺失 terminal 事件、读错误等）时，
 // 已观测到的上游 usage 不得随错误一起被丢弃，Forward 必须把部分结果与错误一同
 // 返回，供 handler 照常提交 usage 记录。
@@ -183,6 +208,95 @@ func TestGatewayService_Forward_FailoverErrorKeepsNilResult(t *testing.T) {
 	require.Nil(t, result, "failover 错误必须保持 result=nil，防止重试成功后双重计费")
 }
 
+func TestGatewayService_Forward_PreOutputSSEOverloadedErrorUsesSemantic529(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
+
+	body := []byte(`{"model":"claude-3-5-sonnet-latest","stream":true,"messages":[{"role":"user","content":"hello"}]}`)
+	parsed, err := ParseGatewayRequest(NewRequestBodyRef(body), PlatformAnthropic)
+	require.NoError(t, err)
+
+	const errorJSON = `{"type":"error","error":{"details":null,"type":"overloaded_error","message":"Overloaded"},"request_id":"req_01"}`
+	upstream := &anthropicHTTPUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}},
+		Body:       io.NopCloser(strings.NewReader("event: error\ndata: " + errorJSON + "\n\n")),
+	}}
+	repo := &gatewayForwardErrorPolicyRepoStub{}
+	cfg := &config.Config{Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize}}
+	svc := &GatewayService{
+		cfg:                  cfg,
+		responseHeaderFilter: compileResponseHeaderFilter(cfg),
+		httpUpstream:         upstream,
+		rateLimitService:     NewRateLimitService(repo, nil, cfg, nil, nil),
+		deferredService:      &DeferredService{},
+	}
+	account := newAnthropicOAuthAccountForPartialUsageTest()
+	account.Credentials["temp_unschedulable_enabled"] = true
+	account.Credentials["temp_unschedulable_rules"] = []any{map[string]any{
+		"error_code":       float64(529),
+		"keywords":         []any{"Overloaded"},
+		"duration_minutes": float64(10),
+	}}
+
+	result, err := svc.Forward(context.Background(), c, account, parsed)
+	require.Error(t, err)
+	require.Nil(t, result)
+
+	var failoverErr *UpstreamFailoverError
+	require.ErrorAs(t, err, &failoverErr)
+	require.Equal(t, 529, failoverErr.StatusCode)
+	require.JSONEq(t, errorJSON, string(failoverErr.ResponseBody))
+	require.Len(t, repo.modelRateLimitCalls, 1, "synthetic 529 must participate in temp-unschedulable rules")
+	require.Equal(t, account.ID, repo.modelRateLimitCalls[0].accountID)
+	require.Equal(t, parsed.Model, repo.modelRateLimitCalls[0].scope)
+	require.Empty(t, rec.Body.String(), "pre-output overload must remain eligible for account failover")
+}
+
+func TestGatewayService_Forward_PostOutputSSEOverloadedErrorKeepsExistingStatus(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
+
+	body := []byte(`{"model":"claude-3-5-sonnet-latest","stream":true,"messages":[{"role":"user","content":"hello"}]}`)
+	parsed, err := ParseGatewayRequest(NewRequestBodyRef(body), PlatformAnthropic)
+	require.NoError(t, err)
+
+	const errorJSON = `{"type":"error","error":{"type":"overloaded_error","message":"Overloaded"}}`
+	fixture := "event: message_start\ndata: {\"type\":\"message_start\",\"message\":{\"usage\":{\"input_tokens\":1}}}\n\n" +
+		"event: error\ndata: " + errorJSON + "\n\n"
+	upstream := &anthropicHTTPUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}},
+		Body:       io.NopCloser(strings.NewReader(fixture)),
+	}}
+	repo := &gatewayForwardErrorPolicyRepoStub{}
+	cfg := &config.Config{Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize}}
+	svc := &GatewayService{
+		cfg:                  cfg,
+		responseHeaderFilter: compileResponseHeaderFilter(cfg),
+		httpUpstream:         upstream,
+		rateLimitService:     NewRateLimitService(repo, nil, cfg, nil, nil),
+		deferredService:      &DeferredService{},
+	}
+
+	result, err := svc.Forward(context.Background(), c, newAnthropicOAuthAccountForPartialUsageTest(), parsed)
+	require.Error(t, err)
+	require.Nil(t, result)
+
+	var failoverErr *UpstreamFailoverError
+	require.ErrorAs(t, err, &failoverErr)
+	require.Equal(t, http.StatusForbidden, failoverErr.StatusCode)
+	require.JSONEq(t, errorJSON, string(failoverErr.ResponseBody))
+	require.Zero(t, repo.tempCalls)
+	require.Contains(t, rec.Body.String(), "message_start")
+}
+
 func TestGatewayService_AnthropicAPIKeyPassthrough_ForwardStreamMissingTerminalPreservesPartialUsage(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 

From a288bab73ace3ec789dd3af26ca8e98f3277c766 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Thu, 13 Aug 2026 12:40:06 +0800
Subject: [PATCH 17/68] fix(gateway): align passthrough model discovery

---
 .../gateway_hotpath_optimization_test.go      | 91 +++++++++++++++++++
 backend/internal/service/gateway_service.go   | 11 +++
 2 files changed, 102 insertions(+)

diff --git a/backend/internal/service/gateway_hotpath_optimization_test.go b/backend/internal/service/gateway_hotpath_optimization_test.go
index 48cfc358a6fd..d9a00710a38c 100644
--- a/backend/internal/service/gateway_hotpath_optimization_test.go
+++ b/backend/internal/service/gateway_hotpath_optimization_test.go
@@ -602,6 +602,97 @@ func TestGetAvailableModels_ErrorAndGlobalListBranches(t *testing.T) {
 	require.Equal(t, int64(1), okRepo.listAllCalls.Load())
 }
 
+func TestGetAvailableModels_OpenAIPassthroughUsesDefaultFallback(t *testing.T) {
+	groupID := int64(10)
+
+	tests := []struct {
+		name     string
+		accounts []Account
+		want     []string
+	}{
+		{
+			name: "passthrough only ignores stale mapping",
+			accounts: []Account{
+				{
+					ID:          1,
+					Platform:    PlatformOpenAI,
+					Credentials: map[string]any{"model_mapping": map[string]any{"stale-model": "upstream-model"}},
+					Extra:       map[string]any{"openai_passthrough": true},
+				},
+			},
+			want: nil,
+		},
+		{
+			name: "passthrough wins over ordinary account mapping",
+			accounts: []Account{
+				{
+					ID:          2,
+					Platform:    PlatformOpenAI,
+					Credentials: map[string]any{"model_mapping": map[string]any{"configured-model": "configured-upstream"}},
+				},
+				{
+					ID:          3,
+					Platform:    PlatformOpenAI,
+					Credentials: map[string]any{"model_mapping": map[string]any{"stale-model": "upstream-model"}},
+					Extra:       map[string]any{"openai_passthrough": true},
+				},
+			},
+			want: nil,
+		},
+		{
+			name: "ordinary accounts preserve mapped whitelist",
+			accounts: []Account{
+				{
+					ID:          4,
+					Platform:    PlatformOpenAI,
+					Credentials: map[string]any{"model_mapping": map[string]any{"configured-model": "configured-upstream"}},
+				},
+			},
+			want: []string{"configured-model"},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			repo := &modelsListAccountRepoStub{byGroup: map[int64][]Account{groupID: tt.accounts}}
+			svc := &GatewayService{
+				accountRepo:        repo,
+				modelsListCache:    gocache.New(time.Minute, time.Minute),
+				modelsListCacheTTL: time.Minute,
+			}
+
+			require.Equal(t, tt.want, svc.GetAvailableModels(context.Background(), &groupID, PlatformOpenAI))
+		})
+	}
+}
+
+func TestGetAvailableModels_GlobalListPreservesMappedModelsWithOpenAIPassthrough(t *testing.T) {
+	groupID := int64(11)
+	repo := &modelsListAccountRepoStub{
+		byGroup: map[int64][]Account{
+			groupID: {
+				{
+					ID:       1,
+					Platform: PlatformOpenAI,
+					Extra:    map[string]any{"openai_passthrough": true},
+				},
+				{
+					ID:          2,
+					Platform:    PlatformAnthropic,
+					Credentials: map[string]any{"model_mapping": map[string]any{"claude-mapped": "claude-upstream"}},
+				},
+			},
+		},
+	}
+	svc := &GatewayService{
+		accountRepo:        repo,
+		modelsListCache:    gocache.New(time.Minute, time.Minute),
+		modelsListCacheTTL: time.Minute,
+	}
+
+	require.Equal(t, []string{"claude-mapped"}, svc.GetAvailableModels(context.Background(), &groupID, ""))
+}
+
 func TestGatewayHotpathHelpers_CacheTTLAndStickyContext(t *testing.T) {
 	t.Run("resolve_user_group_rate_cache_ttl", func(t *testing.T) {
 		require.Equal(t, defaultUserGroupRateCacheTTL, resolveUserGroupRateCacheTTL(nil))
diff --git a/backend/internal/service/gateway_service.go b/backend/internal/service/gateway_service.go
index 0f3415fab513..7e42a74cbd92 100644
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@@ -1379,6 +1379,17 @@ func (s *GatewayService) GetAvailableModels(ctx context.Context, groupID *int64,
 	hasAnyMapping := false
 
 	for _, acc := range accounts {
+		// Passthrough routing accepts models independently of model_mapping. A stale
+		// mapping on any eligible passthrough account therefore cannot define the
+		// public whitelist; return nil so the handler uses its default model set.
+		if platform == PlatformOpenAI && acc.IsOpenAIPassthroughEnabled() {
+			if s.modelsListCache != nil {
+				s.modelsListCache.Set(cacheKey, []string(nil), s.modelsListCacheTTL)
+				modelsListCacheStoreTotal.Add(1)
+			}
+			return nil
+		}
+
 		mapping := acc.GetModelMapping()
 		if len(mapping) > 0 {
 			hasAnyMapping = true

From 674570ca17be336f2015d229495c8ceeda9d6c04 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Fri, 14 Aug 2026 00:25:55 +0800
Subject: [PATCH 18/68] fix: preserve group pricing in auth snapshots

---
 .../internal/service/api_key_auth_cache.go    |  2 +
 .../service/api_key_auth_cache_impl.go        |  6 ++-
 .../api_key_auth_cache_pricing_test.go        | 50 +++++++++++++++++++
 .../service/api_key_auth_cache_profit_test.go |  2 +-
 4 files changed, 58 insertions(+), 2 deletions(-)
 create mode 100644 backend/internal/service/api_key_auth_cache_pricing_test.go

diff --git a/backend/internal/service/api_key_auth_cache.go b/backend/internal/service/api_key_auth_cache.go
index 5fdc96b7937e..30a41eece6cc 100644
--- a/backend/internal/service/api_key_auth_cache.go
+++ b/backend/internal/service/api_key_auth_cache.go
@@ -84,6 +84,8 @@ type APIKeyAuthGroupSnapshot struct {
 	AudioRealtimePricePerMin        *float64                      `json:"audio_realtime_price_per_min,omitempty"`
 	AudioTTSPricePerMillionChars    *float64                      `json:"audio_tts_price_per_million_chars,omitempty"`
 	AudioSTTPricePerHour            *float64                      `json:"audio_stt_price_per_hour,omitempty"`
+	LongContextPricingEnabled       bool                          `json:"long_context_pricing_enabled"`
+	ModelPricing                    []ChannelModelPricing         `json:"model_pricing,omitempty"`
 	ClaudeCodeOnly                  bool                          `json:"claude_code_only"`
 	FallbackGroupID                 *int64                        `json:"fallback_group_id,omitempty"`
 	FallbackGroupIDOnInvalidRequest *int64                        `json:"fallback_group_id_on_invalid_request,omitempty"`
diff --git a/backend/internal/service/api_key_auth_cache_impl.go b/backend/internal/service/api_key_auth_cache_impl.go
index 696f21bb257e..7cb37baf4874 100644
--- a/backend/internal/service/api_key_auth_cache_impl.go
+++ b/backend/internal/service/api_key_auth_cache_impl.go
@@ -14,7 +14,7 @@ import (
 	"github.com/dgraph-io/ristretto"
 )
 
-const apiKeyAuthSnapshotVersion = 19 // v19: group search/audio/video_model_prices billing fields (force refresh of pre-fix snapshots)
+const apiKeyAuthSnapshotVersion = 20 // v20: group long-context and model pricing fields (force refresh of pre-fix snapshots)
 
 type apiKeyAuthCacheConfig struct {
 	l1Size        int
@@ -406,6 +406,8 @@ func (s *APIKeyService) snapshotFromAPIKey(ctx context.Context, apiKey *APIKey)
 			AudioRealtimePricePerMin:        apiKey.Group.AudioRealtimePricePerMin,
 			AudioTTSPricePerMillionChars:    apiKey.Group.AudioTTSPricePerMillionChars,
 			AudioSTTPricePerHour:            apiKey.Group.AudioSTTPricePerHour,
+			LongContextPricingEnabled:       apiKey.Group.LongContextPricingEnabled,
+			ModelPricing:                    apiKey.Group.ModelPricing,
 			ClaudeCodeOnly:                  apiKey.Group.ClaudeCodeOnly,
 			FallbackGroupID:                 apiKey.Group.FallbackGroupID,
 			FallbackGroupIDOnInvalidRequest: apiKey.Group.FallbackGroupIDOnInvalidRequest,
@@ -501,6 +503,8 @@ func (s *APIKeyService) snapshotToAPIKey(key string, snapshot *APIKeyAuthSnapsho
 			AudioRealtimePricePerMin:        snapshot.Group.AudioRealtimePricePerMin,
 			AudioTTSPricePerMillionChars:    snapshot.Group.AudioTTSPricePerMillionChars,
 			AudioSTTPricePerHour:            snapshot.Group.AudioSTTPricePerHour,
+			LongContextPricingEnabled:       snapshot.Group.LongContextPricingEnabled,
+			ModelPricing:                    snapshot.Group.ModelPricing,
 			ClaudeCodeOnly:                  snapshot.Group.ClaudeCodeOnly,
 			FallbackGroupID:                 snapshot.Group.FallbackGroupID,
 			FallbackGroupIDOnInvalidRequest: snapshot.Group.FallbackGroupIDOnInvalidRequest,
diff --git a/backend/internal/service/api_key_auth_cache_pricing_test.go b/backend/internal/service/api_key_auth_cache_pricing_test.go
new file mode 100644
index 000000000000..22aa636aa725
--- /dev/null
+++ b/backend/internal/service/api_key_auth_cache_pricing_test.go
@@ -0,0 +1,50 @@
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestAPIKeyAuthSnapshotGroupPricingRoundtrip(t *testing.T) {
+	groupID := int64(50)
+	inputPrice := 1e-6
+	outputPrice := 2e-6
+	apiKey := &APIKey{
+		ID: 82, UserID: 40, GroupID: &groupID, Key: "sk-pricing-roundtrip", Status: StatusActive,
+		User: &User{ID: 40, Status: StatusActive},
+		Group: &Group{
+			ID: groupID, Name: "pricing-roundtrip", Platform: PlatformAnthropic, Status: StatusActive,
+			LongContextPricingEnabled: true,
+			ModelPricing: []ChannelModelPricing{{
+				Models: []string{"claude-sonnet-*"}, BillingMode: BillingModeToken,
+				InputPrice: &inputPrice, OutputPrice: &outputPrice,
+			}},
+		},
+	}
+	svc := &APIKeyService{}
+
+	payload, err := json.Marshal(&APIKeyAuthCacheEntry{Snapshot: svc.snapshotFromAPIKey(context.Background(), apiKey)})
+	require.NoError(t, err)
+	var cached APIKeyAuthCacheEntry
+	require.NoError(t, json.Unmarshal(payload, &cached))
+
+	materialized, used, err := svc.applyAuthCacheEntry(apiKey.Key, &cached)
+	require.NoError(t, err)
+	require.True(t, used)
+	require.NotNil(t, materialized.Group)
+	require.True(t, materialized.Group.LongContextPricingEnabled)
+	require.Equal(t, apiKey.Group.ModelPricing, materialized.Group.ModelPricing)
+
+	billing := &BillingService{fallbackPrices: map[string]*ModelPricing{
+		"claude-sonnet-4": {InputPricePerToken: 3e-6, OutputPricePerToken: 15e-6},
+	}}
+	resolver := NewModelPricingResolver(nil, billing)
+	resolved := resolver.Resolve(context.Background(), PricingInput{Model: "claude-sonnet-4", Group: materialized.Group})
+	require.Equal(t, PricingSourceGroup, resolved.Source)
+	require.True(t, resolved.longContextPricingEnabled)
+	require.InDelta(t, inputPrice, resolved.BasePricing.InputPricePerToken, 1e-12)
+	require.InDelta(t, outputPrice, resolved.BasePricing.OutputPricePerToken, 1e-12)
+}
diff --git a/backend/internal/service/api_key_auth_cache_profit_test.go b/backend/internal/service/api_key_auth_cache_profit_test.go
index 40519f28967f..fbcd96a42ae6 100644
--- a/backend/internal/service/api_key_auth_cache_profit_test.go
+++ b/backend/internal/service/api_key_auth_cache_profit_test.go
@@ -53,7 +53,7 @@ func TestAPIKeyAuthSnapshotProfitControlRoundtrip(t *testing.T) {
 	snapshot := svc.snapshotFromAPIKey(context.Background(), apiKey)
 	require.NotNil(t, snapshot)
 	require.Equal(t, apiKeyAuthSnapshotVersion, snapshot.Version)
-	require.Equal(t, 19, snapshot.Version, "v19 起认证快照携带 search/audio/video_model_prices 计费字段")
+	require.Equal(t, 20, snapshot.Version, "v20 起认证快照携带分组长上下文与模型定价字段")
 
 	// 模拟 L2 缓存的完整 JSON 往返（与 apiKeyCache.SetAuthCache/GetAuthCache 同构）。
 	payload, err := json.Marshal(&APIKeyAuthCacheEntry{Snapshot: snapshot})

From 44ef88f659a02ca9b725ce11b110e03cd43d814d Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Sat, 15 Aug 2026 11:02:59 +0800
Subject: [PATCH 19/68] fix(openai): restore API-key custom tools

---
 .../service/openai_gateway_forward.go         |   1 +
 .../service/openai_gateway_passthrough.go     | 102 ++++++++++++
 ...nai_gateway_responses_client_tools_test.go | 146 ++++++++++++++++++
 3 files changed, 249 insertions(+)
 create mode 100644 backend/internal/service/openai_gateway_responses_client_tools_test.go

diff --git a/backend/internal/service/openai_gateway_forward.go b/backend/internal/service/openai_gateway_forward.go
index f7eb75934606..31beb344a0f6 100644
--- a/backend/internal/service/openai_gateway_forward.go
+++ b/backend/internal/service/openai_gateway_forward.go
@@ -21,6 +21,7 @@ import (
 func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, account *Account, body []byte) (*OpenAIForwardResult, error) {
 	beginUpstreamResponseModelObservation(c)
 	clearGrokResponsesClientToolMapping(c)
+	clearOpenAIResponsesClientToolMapping(c)
 	clearOpenAIResponsesNamespaceNames(c)
 	startTime := time.Now()
 	// 固定渠道映射后的请求级 canonical body；账号 normalize/strip 不得改写跨 failover hint。
diff --git a/backend/internal/service/openai_gateway_passthrough.go b/backend/internal/service/openai_gateway_passthrough.go
index 8c029f0aa863..4fc694f17f86 100644
--- a/backend/internal/service/openai_gateway_passthrough.go
+++ b/backend/internal/service/openai_gateway_passthrough.go
@@ -16,6 +16,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
@@ -25,6 +26,84 @@ import (
 	"go.uber.org/zap"
 )
 
+const openAIResponsesClientToolMappingContextKey = "openai_responses_client_tool_mapping"
+
+func hasOpenAIResponsesClientToolMapping(mapping apicompat.ResponsesClientToolMapping) bool {
+	return len(mapping.CustomTools) > 0 || mapping.ToolSearch || len(mapping.NamespaceTools) > 0
+}
+
+func adaptOpenAIResponsesClientTools(body []byte) ([]byte, apicompat.ResponsesClientToolMapping, error) {
+	if !needsOpenAIResponsesClientToolAdaptation(body) {
+		return body, apicompat.ResponsesClientToolMapping{}, nil
+	}
+
+	decoder := json.NewDecoder(bytes.NewReader(body))
+	decoder.UseNumber()
+	var requestBody map[string]any
+	if err := decoder.Decode(&requestBody); err != nil {
+		return body, apicompat.ResponsesClientToolMapping{}, fmt.Errorf("decode OpenAI Responses client tools: %w", err)
+	}
+	var trailingValue any
+	if err := decoder.Decode(&trailingValue); !errors.Is(err, io.EOF) {
+		if err == nil {
+			err = errors.New("multiple JSON values")
+		}
+		return body, apicompat.ResponsesClientToolMapping{}, fmt.Errorf("decode OpenAI Responses client tools trailing data: %w", err)
+	}
+	mapping, changed, err := apicompat.AdaptResponsesClientTools(requestBody)
+	if err != nil || !changed {
+		return body, mapping, err
+	}
+	rebuilt, err := marshalOpenAIUpstreamJSON(requestBody)
+	if err != nil {
+		return body, apicompat.ResponsesClientToolMapping{}, fmt.Errorf("encode OpenAI Responses client tools: %w", err)
+	}
+	return rebuilt, mapping, nil
+}
+
+func needsOpenAIResponsesClientToolAdaptation(body []byte) bool {
+	needsAdaptation := false
+	var visit func(gjson.Result) bool
+	visit = func(value gjson.Result) bool {
+		if value.IsObject() {
+			switch strings.TrimSpace(value.Get("type").String()) {
+			case "custom", "custom_tool_call", "custom_tool_call_output",
+				"tool_search", "tool_search_call", "tool_search_output":
+				needsAdaptation = true
+				return false
+			}
+		}
+		if value.IsObject() || value.IsArray() {
+			value.ForEach(func(_, child gjson.Result) bool {
+				return visit(child)
+			})
+		}
+		return !needsAdaptation
+	}
+	visit(gjson.ParseBytes(body))
+	return needsAdaptation
+}
+
+func openAIResponsesClientToolMapping(c *gin.Context) (apicompat.ResponsesClientToolMapping, bool) {
+	if c == nil {
+		return apicompat.ResponsesClientToolMapping{}, false
+	}
+	value, ok := c.Get(openAIResponsesClientToolMappingContextKey)
+	mapping, typed := value.(apicompat.ResponsesClientToolMapping)
+	return mapping, ok && typed && hasOpenAIResponsesClientToolMapping(mapping)
+}
+
+// clearOpenAIResponsesClientToolMapping removes mapping state from the prior
+// forwarding attempt. Forward retries accounts on the same Gin context.
+func clearOpenAIResponsesClientToolMapping(c *gin.Context) {
+	if c == nil {
+		return
+	}
+	if _, exists := c.Get(openAIResponsesClientToolMappingContextKey); exists {
+		c.Set(openAIResponsesClientToolMappingContextKey, apicompat.ResponsesClientToolMapping{})
+	}
+}
+
 func (s *OpenAIGatewayService) forwardOpenAIPassthrough(
 	ctx context.Context,
 	c *gin.Context,
@@ -82,6 +161,16 @@ func (s *OpenAIGatewayService) forwardOpenAIPassthrough(
 		reqStream = gjson.GetBytes(body, "stream").Bool()
 	}
 
+	if account != nil && account.Platform == PlatformOpenAI && account.Type == AccountTypeAPIKey &&
+		!isOpenAIResponsesCompactPath(c) && needsOpenAIResponsesClientToolAdaptation(body) {
+		adaptedBody, mapping, adaptErr := adaptOpenAIResponsesClientTools(body)
+		if adaptErr != nil {
+			return nil, adaptErr
+		}
+		body = adaptedBody
+		c.Set(openAIResponsesClientToolMappingContextKey, mapping)
+	}
+
 	sanitizedBody, sanitized, err := sanitizeEmptyBase64InputImagesInOpenAIBody(body)
 	if err != nil {
 		return nil, err
@@ -236,6 +325,13 @@ func (s *OpenAIGatewayService) forwardOpenAIPassthrough(
 		return nil, s.handleErrorResponsePassthrough(ctx, resp, c, account, body, probeBody)
 	}
 	defer func() { _ = resp.Body.Close() }()
+	if mapping, ok := openAIResponsesClientToolMapping(c); ok && isEventStreamResponse(resp.Header) {
+		maxLineSize := defaultMaxLineSize
+		if s.cfg != nil && s.cfg.Gateway.MaxLineSize > 0 {
+			maxLineSize = s.cfg.Gateway.MaxLineSize
+		}
+		resp.Body = newGrokResponsesClientToolStreamBody(resp.Body, mapping, maxLineSize)
+	}
 
 	serviceTier := extractOpenAIServiceTierFromBody(body)
 
@@ -1520,6 +1616,12 @@ func (s *OpenAIGatewayService) handleNonStreamingResponsePassthrough(
 	if err != nil {
 		return nil, fmt.Errorf("restore OpenAI passthrough namespace response: %w", err)
 	}
+	if mapping, ok := openAIResponsesClientToolMapping(c); ok && json.Valid(body) {
+		body, _, err = apicompat.RestoreResponsesClientToolPayload(body, mapping)
+		if err != nil {
+			return nil, fmt.Errorf("restore OpenAI Responses client tools: %w", err)
+		}
+	}
 	if !writeOpenAICompactSSEBridge(c, resp.StatusCode, body) {
 		c.Data(resp.StatusCode, contentType, body)
 	}
diff --git a/backend/internal/service/openai_gateway_responses_client_tools_test.go b/backend/internal/service/openai_gateway_responses_client_tools_test.go
new file mode 100644
index 000000000000..cfb4f6eb624f
--- /dev/null
+++ b/backend/internal/service/openai_gateway_responses_client_tools_test.go
@@ -0,0 +1,146 @@
+package service
+
+import (
+	"bytes"
+	"context"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+func openAIClientToolsRequest(stream bool) []byte {
+	streamValue := "false"
+	if stream {
+		streamValue = "true"
+	}
+	return []byte(`{"model":"gpt-5.4","input":"fix it","stream":` + streamValue + `,"tools":[{"type":"custom","name":"exec"},{"type":"custom","name":"apply_patch"}]}`)
+}
+
+func assertOpenAIClientToolsLowered(t *testing.T, body []byte) {
+	t.Helper()
+	for index, name := range []string{"exec", "apply_patch"} {
+		tool := gjson.GetBytes(body, "tools."+string(rune('0'+index)))
+		require.Equal(t, "function", tool.Get("type").String())
+		require.Equal(t, name, tool.Get("name").String())
+		require.Equal(t, "string", tool.Get("parameters.properties.input.type").String())
+	}
+}
+
+func openAIClientToolsTestService(upstream *httpUpstreamRecorder) *OpenAIGatewayService {
+	return &OpenAIGatewayService{
+		httpUpstream: upstream,
+		cfg: &config.Config{Security: config.SecurityConfig{
+			URLAllowlist: config.URLAllowlistConfig{Enabled: false},
+		}},
+	}
+}
+
+func TestAdaptOpenAIResponsesClientToolsLeavesNamespaceOnlyBodyUnchanged(t *testing.T) {
+	body := []byte(`{
+		"model": "gpt-5.5",
+		"tools": [{"type": "namespace", "name": "code_tools", "tools": [{"type": "function", "name": "run"}]}],
+		"tool_choice": "auto"
+	}`)
+
+	adapted, mapping, err := adaptOpenAIResponsesClientTools(body)
+
+	require.NoError(t, err)
+	require.Equal(t, body, adapted)
+	require.Empty(t, mapping.CustomTools)
+	require.Empty(t, mapping.NamespaceTools)
+	require.False(t, mapping.ToolSearch)
+}
+
+func TestAdaptOpenAIResponsesClientToolsRejectsTrailingData(t *testing.T) {
+	tests := map[string][]byte{
+		"trailing garbage":     append(openAIClientToolsRequest(false), []byte(` garbage`)...),
+		"second JSON document": append(openAIClientToolsRequest(false), []byte(` {"model":"other"}`)...),
+	}
+
+	for name, body := range tests {
+		t.Run(name, func(t *testing.T) {
+			adapted, mapping, err := adaptOpenAIResponsesClientTools(body)
+
+			require.ErrorContains(t, err, "decode OpenAI Responses client tools trailing data")
+			require.Equal(t, body, adapted)
+			require.Empty(t, mapping)
+		})
+	}
+}
+
+func TestClearOpenAIResponsesClientToolMappingRemovesStaleContextState(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	c, _ := gin.CreateTestContext(httptest.NewRecorder())
+	c.Set(openAIResponsesClientToolMappingContextKey, apicompat.ResponsesClientToolMapping{CustomTools: map[string]bool{"exec": true}})
+
+	clearOpenAIResponsesClientToolMapping(c)
+
+	_, ok := openAIResponsesClientToolMapping(c)
+	require.False(t, ok)
+}
+
+func TestOpenAIPassthroughAPIKeyRestoresClientToolsNonStreaming(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	body := openAIClientToolsRequest(false)
+	recorder := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(recorder)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", bytes.NewReader(body))
+
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"application/json"}},
+		Body: io.NopCloser(strings.NewReader(`{"id":"resp_tools","status":"completed","output":[
+			{"type":"function_call","id":"i1","call_id":"c1","name":"exec","arguments":"{\"input\":\"pwd\"}"},
+			{"type":"function_call","id":"i2","call_id":"c2","name":"apply_patch","arguments":"{\"input\":\"*** Begin Patch\"}"}],"usage":{}}`)),
+	}}
+	svc := openAIClientToolsTestService(upstream)
+	account := &Account{ID: 5659, Platform: PlatformOpenAI, Type: AccountTypeAPIKey, Credentials: map[string]any{"api_key": "test-key"}}
+
+	result, err := svc.forwardOpenAIPassthrough(context.Background(), c, account, body, body, "gpt-5.4", false, nil, false, time.Now())
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	assertOpenAIClientToolsLowered(t, upstream.lastBody)
+	require.Equal(t, "custom_tool_call", gjson.Get(recorder.Body.String(), "output.0.type").String())
+	require.Equal(t, "pwd", gjson.Get(recorder.Body.String(), "output.0.input").String())
+	require.Equal(t, "custom_tool_call", gjson.Get(recorder.Body.String(), "output.1.type").String())
+	require.Equal(t, "*** Begin Patch", gjson.Get(recorder.Body.String(), "output.1.input").String())
+}
+
+func TestOpenAIPassthroughAPIKeyRestoresClientToolsStreaming(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	body := openAIClientToolsRequest(true)
+	recorder := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(recorder)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", bytes.NewReader(body))
+
+	sse := strings.Join([]string{
+		`data: {"type":"response.output_item.added","sequence_number":0,"output_index":0,"item":{"type":"function_call","id":"i1","call_id":"c1","name":"apply_patch","status":"in_progress"}}`,
+		`data: {"type":"response.function_call_arguments.done","sequence_number":1,"item_id":"i1","call_id":"c1","name":"apply_patch","arguments":"{\"input\":\"*** Begin Patch\"}"}`,
+		`data: {"type":"response.output_item.done","sequence_number":2,"output_index":0,"item":{"type":"function_call","id":"i1","call_id":"c1","name":"apply_patch","arguments":"{\"input\":\"*** Begin Patch\"}","status":"completed"}}`,
+		`data: {"type":"response.completed","sequence_number":3,"response":{"id":"resp_stream_tools","status":"completed","output":[{"type":"function_call","id":"i1","call_id":"c1","name":"apply_patch","arguments":"{\"input\":\"*** Begin Patch\"}"}],"usage":{"input_tokens":1,"output_tokens":1}}}`,
+	}, "\n\n") + "\n\n"
+	upstream := &httpUpstreamRecorder{resp: &http.Response{StatusCode: http.StatusOK, Header: http.Header{"Content-Type": []string{"text/event-stream"}}, Body: io.NopCloser(strings.NewReader(sse))}}
+	svc := openAIClientToolsTestService(upstream)
+	account := &Account{ID: 5660, Platform: PlatformOpenAI, Type: AccountTypeAPIKey, Credentials: map[string]any{"api_key": "test-key"}}
+
+	result, err := svc.forwardOpenAIPassthrough(context.Background(), c, account, body, body, "gpt-5.4", false, nil, true, time.Now())
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	assertOpenAIClientToolsLowered(t, upstream.lastBody)
+	output := recorder.Body.String()
+	require.Contains(t, output, `"type":"custom_tool_call"`)
+	require.Contains(t, output, `"type":"response.custom_tool_call_input.done"`)
+	require.Contains(t, output, `"input":"*** Begin Patch"`)
+	require.NotContains(t, output, `"input":{`)
+}

From d677d67dda8a961dff7671ce5c9bf24bff26b505 Mon Sep 17 00:00:00 2001
From: feeeei <g540203430@gmail.com>
Date: Sat, 15 Aug 2026 16:43:20 +0800
Subject: [PATCH 20/68] =?UTF-8?q?feat:=20OpenAI=20Team=20=E8=81=94?=
 =?UTF-8?q?=E5=8A=A8=E7=86=94=E6=96=AD?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

OpenAI OAuth 账户收到 402 deactivated_workspace（ChatGPT Team 工作区被停用）时，
将同一 Team（credentials.chatgpt_account_id 相同）的其余 active 账户一并置为
error 并进程内立即熔断，让调度瞬间切走，避免流量继续打到已死的兄弟账户。

- 触发条件仅限 402 + detail.code == deactivated_workspace；通用 402 与 401 行为不变
- 调用点两处：fastpath 各类早退之前 + HandleUpstreamError 顶部，进程内 60s 按 Team 去重
- 逐账户 SetError（错误信息标注触发账户），单账户失败不中断其余
- 默认生效，无配置开关
---
 .../openai_account_runtime_block_fastpath.go  |   4 +
 .../service/openai_team_linked_error.go       |  99 +++++++++
 backend/internal/service/ratelimit_service.go |   7 +
 ...telimit_service_openai_team_linked_test.go | 192 ++++++++++++++++++
 4 files changed, 302 insertions(+)
 create mode 100644 backend/internal/service/openai_team_linked_error.go
 create mode 100644 backend/internal/service/ratelimit_service_openai_team_linked_test.go

diff --git a/backend/internal/service/openai_account_runtime_block_fastpath.go b/backend/internal/service/openai_account_runtime_block_fastpath.go
index ace8366a0eb0..2822374257fc 100644
--- a/backend/internal/service/openai_account_runtime_block_fastpath.go
+++ b/backend/internal/service/openai_account_runtime_block_fastpath.go
@@ -72,6 +72,10 @@ func (s *OpenAIGatewayService) handleOpenAIAccountUpstreamError(ctx context.Cont
 	if s == nil || account == nil {
 		return false
 	}
+	// Team 联动熔断必须先于 model-not-found 与账户级临时不可调度规则的早退。
+	if s.rateLimitService != nil {
+		s.rateLimitService.maybeHandleOpenAITeamLinkedError(stateCtx, account, statusCode, responseBody)
+	}
 	stateCtx = withTempUnschedulableModel(stateCtx, canonicalModel)
 	if s.rateLimitService != nil && len(canonicalModel) > 0 && s.rateLimitService.HandleUpstreamModelNotFound(stateCtx, account, canonicalModel[0], statusCode, responseBody) {
 		return true
diff --git a/backend/internal/service/openai_team_linked_error.go b/backend/internal/service/openai_team_linked_error.go
new file mode 100644
index 000000000000..2225e59a5b5b
--- /dev/null
+++ b/backend/internal/service/openai_team_linked_error.go
@@ -0,0 +1,99 @@
+package service
+
+import (
+	"context"
+	"fmt"
+	"log/slog"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/tidwall/gjson"
+)
+
+const (
+	openAITeamLinkedErrorDedupTTL      = 60 * time.Second
+	openAITeamLinkedErrorFanoutTimeout = 30 * time.Second
+	openAITeamLinkedErrorBlockReason   = "team_linked_error"
+)
+
+// maybeHandleOpenAITeamLinkedError 在 OpenAI OAuth 账户收到 402 deactivated_workspace
+// （ChatGPT Team 工作区被停用）时，把同一 Team（credentials.chatgpt_account_id 相同）
+// 的其余 active 账户一并置为 error 并立即熔断。触发账户自身不在 fan-out 范围内，
+// 仍由常规 402 处理标记。
+func (s *RateLimitService) maybeHandleOpenAITeamLinkedError(ctx context.Context, account *Account, statusCode int, responseBody []byte) {
+	if s == nil || s.accountRepo == nil || statusCode != http.StatusPaymentRequired || !isOpenAIOAuthAccount(account) {
+		return
+	}
+	if gjson.GetBytes(responseBody, "detail.code").String() != "deactivated_workspace" {
+		return
+	}
+	teamID := strings.TrimSpace(account.GetChatGPTAccountID())
+	if teamID == "" {
+		return
+	}
+	if !s.markOpenAITeamLinkedFired(teamID) {
+		return
+	}
+	// 上游报错场景请求 ctx 往往已被取消，落库需要独立生命周期。
+	opCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), openAITeamLinkedErrorFanoutTimeout)
+	defer cancel()
+
+	accounts, err := s.accountRepo.ListByPlatform(opCtx, PlatformOpenAI)
+	if err != nil {
+		slog.Warn("openai_team_linked_error_list_failed", "trigger_account_id", account.ID, "error", err)
+		return
+	}
+	var targets []*Account
+	for i := range accounts {
+		acc := &accounts[i]
+		if acc.ID == account.ID || acc.IsShadow() || strings.TrimSpace(acc.GetChatGPTAccountID()) != teamID {
+			continue
+		}
+		targets = append(targets, acc)
+	}
+	if len(targets) == 0 {
+		return
+	}
+	// 先全部进程内熔断（微秒级生效），再逐个落库，避免后面的账户等待前面的 DB 写入。
+	for _, acc := range targets {
+		s.notifyAccountSchedulingBlocked(acc, time.Time{}, openAITeamLinkedErrorBlockReason)
+	}
+	errorMsg := fmt.Sprintf("Workspace deactivated (402): team-linked error triggered by account #%d", account.ID)
+	marked := 0
+	for _, acc := range targets {
+		// 单账户写入失败不中断其余账户；进程内熔断已先行，且该账户仍为 active，
+		// 下一个 402 在去重 TTL 过期后会重新触发 fan-out。
+		if err := s.accountRepo.SetError(opCtx, acc.ID, errorMsg); err != nil {
+			slog.Warn("openai_team_linked_error_set_error_failed", "account_id", acc.ID, "error", err)
+			continue
+		}
+		marked++
+	}
+	slog.Warn("openai_team_linked_error_fanout",
+		"trigger_account_id", account.ID,
+		"chatgpt_account_id", teamID,
+		"affected", marked,
+		"targets", len(targets),
+	)
+}
+
+// markOpenAITeamLinkedFired 以 teamID 为键做进程内去重：TTL 内同一 Team 只允许一次 fan-out。
+func (s *RateLimitService) markOpenAITeamLinkedFired(teamID string) bool {
+	now := time.Now()
+	s.openaiTeamLinkedMu.Lock()
+	defer s.openaiTeamLinkedMu.Unlock()
+	if expiry, ok := s.openaiTeamLinkedRecent[teamID]; ok && expiry.After(now) {
+		return false
+	}
+	if s.openaiTeamLinkedRecent == nil {
+		s.openaiTeamLinkedRecent = make(map[string]time.Time)
+	}
+	for k, v := range s.openaiTeamLinkedRecent {
+		if !v.After(now) {
+			delete(s.openaiTeamLinkedRecent, k)
+		}
+	}
+	s.openaiTeamLinkedRecent[teamID] = now.Add(openAITeamLinkedErrorDedupTTL)
+	return true
+}
diff --git a/backend/internal/service/ratelimit_service.go b/backend/internal/service/ratelimit_service.go
index ee5b09bbb4b1..ec0c6bf19245 100644
--- a/backend/internal/service/ratelimit_service.go
+++ b/backend/internal/service/ratelimit_service.go
@@ -32,6 +32,10 @@ type RateLimitService struct {
 	runtimeBlocker        AccountRuntimeBlocker
 	usageCacheMu          sync.RWMutex
 	usageCache            map[int64]*geminiUsageCacheEntry
+
+	// OpenAI Team 联动熔断的进程内去重：teamID → 去重窗口截止时间
+	openaiTeamLinkedMu     sync.Mutex
+	openaiTeamLinkedRecent map[string]time.Time
 }
 
 type AccountRuntimeBlocker interface {
@@ -268,6 +272,9 @@ func (s *RateLimitService) CheckErrorPolicy(ctx context.Context, account *Accoun
 // 返回是否应该停止该账号的调度
 func (s *RateLimitService) HandleUpstreamError(ctx context.Context, account *Account, statusCode int, headers http.Header, responseBody []byte, requestedModel ...string) (shouldDisable bool) {
 	ctx = withTempUnschedulableModel(ctx, requestedModel)
+	// Team 联动熔断必须先于池模式/自定义错误码/临时不可调度的各类早退；
+	// 同请求内与 fastpath 调用点的重复触发由方法内去重吸收。
+	s.maybeHandleOpenAITeamLinkedError(ctx, account, statusCode, responseBody)
 	customErrorCodesEnabled := account.IsCustomErrorCodesEnabled()
 
 	// 池模式默认不标记本地账号状态；但管理员显式配置的临时不可调度规则优先。
diff --git a/backend/internal/service/ratelimit_service_openai_team_linked_test.go b/backend/internal/service/ratelimit_service_openai_team_linked_test.go
new file mode 100644
index 000000000000..e72763e2638c
--- /dev/null
+++ b/backend/internal/service/ratelimit_service_openai_team_linked_test.go
@@ -0,0 +1,192 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/stretchr/testify/require"
+)
+
+const teamLinkedDeactivatedBody = `{"detail":{"code":"deactivated_workspace","message":"This workspace has been deactivated."}}`
+
+type teamLinkedAccountRepoStub struct {
+	mockAccountRepoForGemini
+	teamAccounts []Account
+	listErr      error
+	listCalls    int
+	setErrorIDs  []int64
+	setErrorMsgs map[int64]string
+	failSetError map[int64]error
+}
+
+// ListByPlatform 镜像真实仓库语义：仅返回该平台的 active 账户。
+func (r *teamLinkedAccountRepoStub) ListByPlatform(ctx context.Context, platform string) ([]Account, error) {
+	r.listCalls++
+	if r.listErr != nil {
+		return nil, r.listErr
+	}
+	out := make([]Account, 0, len(r.teamAccounts))
+	for _, acc := range r.teamAccounts {
+		if acc.Platform == platform && acc.Status == StatusActive {
+			out = append(out, acc)
+		}
+	}
+	return out, nil
+}
+
+func (r *teamLinkedAccountRepoStub) SetError(ctx context.Context, id int64, errorMsg string) error {
+	if err, ok := r.failSetError[id]; ok {
+		return err
+	}
+	r.setErrorIDs = append(r.setErrorIDs, id)
+	if r.setErrorMsgs == nil {
+		r.setErrorMsgs = make(map[int64]string)
+	}
+	r.setErrorMsgs[id] = errorMsg
+	return nil
+}
+
+func newTeamLinkedAccount(id int64, teamID string) Account {
+	return Account{
+		ID:          id,
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Status:      StatusActive,
+		Credentials: map[string]any{"chatgpt_account_id": teamID},
+	}
+}
+
+// newTeamLinkedFixture: #1 触发者(team-A) #2 同队 #3 异队 #4 apikey #5 影子 #6 同队 #7 同队但已 error
+func newTeamLinkedFixture() []Account {
+	parentID := int64(1)
+	shadow := Account{
+		ID:              5,
+		Platform:        PlatformOpenAI,
+		Type:            AccountTypeOAuth,
+		Status:          StatusActive,
+		ParentAccountID: &parentID,
+	}
+	apikey := newTeamLinkedAccount(4, "team-A")
+	apikey.Type = AccountTypeAPIKey
+	erroredSibling := newTeamLinkedAccount(7, "team-A")
+	erroredSibling.Status = StatusError
+	return []Account{
+		newTeamLinkedAccount(1, "team-A"),
+		newTeamLinkedAccount(2, "team-A"),
+		newTeamLinkedAccount(3, "team-B"),
+		apikey,
+		shadow,
+		newTeamLinkedAccount(6, "team-A"),
+		erroredSibling,
+	}
+}
+
+func newTeamLinkedTestService(repo *teamLinkedAccountRepoStub) (*RateLimitService, *runtimeBlockRecorder) {
+	rl := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	blocker := &runtimeBlockRecorder{}
+	rl.SetAccountRuntimeBlocker(blocker)
+	return rl, blocker
+}
+
+func TestTeamLinkedError_FanoutMarksSameTeamAccounts(t *testing.T) {
+	repo := &teamLinkedAccountRepoStub{teamAccounts: newTeamLinkedFixture()}
+	rl, blocker := newTeamLinkedTestService(repo)
+	trigger := newTeamLinkedAccount(1, "team-A")
+
+	shouldDisable := rl.HandleUpstreamError(context.Background(), &trigger, http.StatusPaymentRequired, http.Header{}, []byte(teamLinkedDeactivatedBody))
+
+	require.True(t, shouldDisable)
+	// fan-out 先标记同队兄弟（#2、#6），触发账户 #1 随后由常规 case 402 标记
+	require.Equal(t, []int64{2, 6, 1}, repo.setErrorIDs)
+	require.Contains(t, repo.setErrorMsgs[2], "team-linked error triggered by account #1")
+	require.Contains(t, repo.setErrorMsgs[6], "team-linked error triggered by account #1")
+	require.Contains(t, repo.setErrorMsgs[1], "Workspace deactivated (402)")
+	require.NotContains(t, repo.setErrorMsgs[1], "team-linked")
+	// 熔断顺序：兄弟账户先于落库全部进程内熔断，触发账户走 auth_error
+	require.Equal(t, []string{openAITeamLinkedErrorBlockReason, openAITeamLinkedErrorBlockReason, "auth_error"}, blocker.reasons)
+	require.Equal(t, int64(2), blocker.accounts[0].ID)
+	require.Equal(t, int64(6), blocker.accounts[1].ID)
+	require.Equal(t, int64(1), blocker.accounts[2].ID)
+}
+
+func TestTeamLinkedError_GenericPaymentErrorDoesNotFanout(t *testing.T) {
+	repo := &teamLinkedAccountRepoStub{teamAccounts: newTeamLinkedFixture()}
+	rl, _ := newTeamLinkedTestService(repo)
+	trigger := newTeamLinkedAccount(1, "team-A")
+
+	rl.HandleUpstreamError(context.Background(), &trigger, http.StatusPaymentRequired, http.Header{}, []byte(`{"error":{"message":"insufficient balance"}}`))
+
+	require.Equal(t, []int64{1}, repo.setErrorIDs)
+	require.Contains(t, repo.setErrorMsgs[1], "Payment required (402)")
+	require.Zero(t, repo.listCalls)
+}
+
+func TestTeamLinkedError_DedupWithinTTL(t *testing.T) {
+	repo := &teamLinkedAccountRepoStub{teamAccounts: newTeamLinkedFixture()}
+	rl, _ := newTeamLinkedTestService(repo)
+	first := newTeamLinkedAccount(1, "team-A")
+	second := newTeamLinkedAccount(2, "team-A")
+
+	rl.HandleUpstreamError(context.Background(), &first, http.StatusPaymentRequired, http.Header{}, []byte(teamLinkedDeactivatedBody))
+	rl.HandleUpstreamError(context.Background(), &second, http.StatusPaymentRequired, http.Header{}, []byte(teamLinkedDeactivatedBody))
+
+	// 第二次触发被去重：只有 #2 自身经 case 402 标记，未再次 fan-out
+	require.Equal(t, []int64{2, 6, 1, 2}, repo.setErrorIDs)
+	require.Equal(t, 1, repo.listCalls)
+}
+
+func TestTeamLinkedError_APIKeyTriggerDoesNotFanout(t *testing.T) {
+	repo := &teamLinkedAccountRepoStub{teamAccounts: newTeamLinkedFixture()}
+	rl, _ := newTeamLinkedTestService(repo)
+	trigger := newTeamLinkedAccount(4, "team-A")
+	trigger.Type = AccountTypeAPIKey
+
+	rl.HandleUpstreamError(context.Background(), &trigger, http.StatusPaymentRequired, http.Header{}, []byte(teamLinkedDeactivatedBody))
+
+	require.Equal(t, []int64{4}, repo.setErrorIDs)
+	require.Zero(t, repo.listCalls)
+}
+
+func TestTeamLinkedError_DirectCallSkipsTriggerAccount(t *testing.T) {
+	// 直调对应 fastpath 调用点：账户级临时不可调度规则短路时联动仍然生效
+	repo := &teamLinkedAccountRepoStub{teamAccounts: newTeamLinkedFixture()}
+	rl, blocker := newTeamLinkedTestService(repo)
+	trigger := newTeamLinkedAccount(1, "team-A")
+
+	rl.maybeHandleOpenAITeamLinkedError(context.Background(), &trigger, http.StatusPaymentRequired, []byte(teamLinkedDeactivatedBody))
+
+	require.Equal(t, []int64{2, 6}, repo.setErrorIDs)
+	require.Equal(t, []string{openAITeamLinkedErrorBlockReason, openAITeamLinkedErrorBlockReason}, blocker.reasons)
+}
+
+func TestTeamLinkedError_MissingTeamIDDoesNothing(t *testing.T) {
+	repo := &teamLinkedAccountRepoStub{teamAccounts: newTeamLinkedFixture()}
+	rl, blocker := newTeamLinkedTestService(repo)
+	trigger := Account{ID: 9, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive}
+
+	rl.maybeHandleOpenAITeamLinkedError(context.Background(), &trigger, http.StatusPaymentRequired, []byte(teamLinkedDeactivatedBody))
+
+	require.Empty(t, repo.setErrorIDs)
+	require.Empty(t, blocker.reasons)
+	require.Zero(t, repo.listCalls)
+}
+
+func TestTeamLinkedError_SetErrorFailureDoesNotAbortRemaining(t *testing.T) {
+	repo := &teamLinkedAccountRepoStub{
+		teamAccounts: newTeamLinkedFixture(),
+		failSetError: map[int64]error{2: errors.New("db down")},
+	}
+	rl, blocker := newTeamLinkedTestService(repo)
+	trigger := newTeamLinkedAccount(1, "team-A")
+
+	rl.maybeHandleOpenAITeamLinkedError(context.Background(), &trigger, http.StatusPaymentRequired, []byte(teamLinkedDeactivatedBody))
+
+	require.Equal(t, []int64{6}, repo.setErrorIDs)
+	// 进程内熔断先于落库执行，两个账户都已被熔断
+	require.Len(t, blocker.reasons, 2)
+}

From 901a0439f1575a45c29150e04d2ccc3ed87f4948 Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Sat, 15 Aug 2026 07:14:19 +0000
Subject: [PATCH 21/68] =?UTF-8?q?feat:=20=E5=9B=BD=E4=BA=A7=E4=BE=9B?=
 =?UTF-8?q?=E5=BA=94=E5=95=86=E4=B8=80=E7=AD=89=E6=94=AF=E6=8C=81=EF=BC=88?=
 =?UTF-8?q?Kimi/Zhipu/DeepSeek=20=E5=A4=9A=E5=8D=8F=E8=AE=AE=20+=20?=
 =?UTF-8?q?=E9=85=8D=E9=A2=9D/=E4=BD=99=E9=A2=9D=E7=9B=91=E6=8E=A7?=
 =?UTF-8?q?=EF=BC=89?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

后端：
- 协议凭证维度 credentials[api_protocol] ∈ chat_completions(默认)/anthropic/responses(deepseek)
- /v1/messages 零转换直通原生 Anthropic 端点（kimi/zhipu/deepseek），CC/Responses
  入站交叉组合走 apicompat 双向转换链（responses/chat_completions anthropic-native 转发器）
- count_tokens：anthropic 协议透传原生端点；其余 CN 协议本地 tiktoken 估算
- Coding Plan 额度探测（5h/weekly 滚动窗口）+ payg 余额探测（kimi/deepseek），
  deepseek 双币种 CNY+USD 明细，任一币种达标不停调
- 周期任务 [CNBalance] 并发探测 + 预算随工作量放大；响应式 429 冷却到最早窗口
  重置点；余额不足可恢复临时停调；智谱 CREDIT_LIMIT 不污染窗口解析
- CC→anthropic 流式客户端断开后继续排水上游保住 usage 计量

前端：
- 创建/编辑弹窗 account_mode + api_protocol + base_url 联动预设（含 watcher 竞态防护）
- 用量单元格：kimi/zhipu coding 显示 5h/weekly 窗口，kimi/deepseek payg 显示余额，
  多币种并列展示；探测失败保留快照；挂载自动探测 5min 去抖
- 调度阈值设置面板补 kimi/zhipu 平台（对齐后端 AllowedSchedulingThresholdPlatforms）
---
 backend/cmd/server/wire.go                    |   7 +
 backend/cmd/server/wire_gen.go                |  15 +-
 backend/cmd/server/wire_gen_test.go           |   1 +
 backend/ent/schema/user_platform_quota.go     |   3 +-
 backend/internal/config/config.go             |  20 +
 backend/internal/domain/constants.go          |  22 +-
 .../internal/handler/admin/account_handler.go |   3 +-
 .../handler/admin/cn_provider_handler.go      |  69 ++
 .../admin/user_platform_quota_admin_test.go   |   9 +-
 backend/internal/handler/handler.go           |   1 +
 .../handler/openai_gateway_handler.go         |  14 +-
 backend/internal/handler/wire.go              |   3 +
 backend/internal/server/api_contract_test.go  |   4 +-
 backend/internal/server/routes/admin.go       |  14 +
 backend/internal/server/routes/gateway.go     |   6 +-
 backend/internal/service/account.go           | 195 +++++-
 .../account_scheduling_threshold_eval.go      |  43 ++
 backend/internal/service/account_service.go   |   3 +
 .../internal/service/anthropic_apikey_auth.go |   9 +-
 backend/internal/service/bedrock_stream.go    |   2 +-
 .../cn_provider_balance_check_service.go      | 279 ++++++++
 .../cn_provider_balance_check_service_test.go | 101 +++
 .../service/cn_provider_balance_service.go    | 270 ++++++++
 .../service/cn_provider_quota_service.go      | 557 ++++++++++++++++
 backend/internal/service/cn_providers_test.go | 625 ++++++++++++++++++
 backend/internal/service/domain_constants.go  |  56 +-
 ...ropic_apikey_passthrough_benchmark_test.go |   6 +-
 ...teway_anthropic_apikey_passthrough_test.go |  18 +-
 .../service/gateway_anthropic_passthrough.go  |  20 +-
 .../service/gateway_upstream_response.go      |   2 +-
 .../service/openai_account_scheduler.go       |  10 +-
 .../service/openai_apikey_responses_probe.go  |  22 +-
 backend/internal/service/openai_embeddings.go |   6 +-
 .../service/openai_gateway_cc_pipeline.go     |   2 +-
 .../openai_gateway_chat_completions.go        |   8 +
 ...teway_chat_completions_anthropic_native.go | 427 ++++++++++++
 .../service/openai_gateway_count_tokens.go    |  33 +-
 .../service/openai_gateway_forward.go         |  13 +-
 .../service/openai_gateway_messages.go        |   9 +
 ...penai_gateway_messages_anthropic_native.go | 619 +++++++++++++++++
 .../openai_gateway_model_availability.go      |   2 +-
 .../service/openai_gateway_passthrough.go     |   5 +-
 .../service/openai_gateway_request_body.go    |  27 +
 ...enai_gateway_responses_anthropic_native.go | 414 ++++++++++++
 .../service/openai_gateway_scheduling.go      |  35 +-
 .../service/openai_gateway_service.go         |   2 +-
 .../service/openai_ws_forwarder_payload.go    |   2 +-
 .../service/ratelimit_cn_providers.go         | 154 +++++
 backend/internal/service/ratelimit_service.go |  14 +
 .../service/upstream_billing_probe.go         |  14 +-
 ...stream_billing_probe_multiplatform_test.go |  23 +-
 backend/internal/service/upstream_models.go   |   9 +-
 backend/internal/service/wire.go              |  42 ++
 docker-compose.yml                            | 302 +++++++++
 frontend/src/api/admin/cnProviders.ts         |  70 ++
 frontend/src/api/admin/index.ts               |   3 +
 frontend/src/api/admin/settings.ts            |   6 +
 .../components/account/AccountUsageCell.vue   |  36 +
 .../account/CNProviderBalanceCell.vue         | 168 +++++
 .../account/CNProviderQuotaCell.vue           | 219 ++++++
 .../components/account/CnBaseUrlPresets.vue   |  53 ++
 .../components/account/CreateAccountModal.vue | 321 ++++++++-
 .../components/account/EditAccountModal.vue   | 166 ++++-
 .../account/ModelWhitelistSelector.vue        |  11 +-
 .../components/account/credentialsBuilder.ts  |  85 +++
 .../src/components/common/PlatformIcon.vue    |  19 +
 .../components/common/PlatformTypeBadge.vue   |  21 +
 frontend/src/composables/useModelWhitelist.ts |   7 +-
 .../src/i18n/locales/en/admin/accounts.ts     |  27 +
 .../src/i18n/locales/zh/admin/accounts.ts     |  27 +
 frontend/src/types/index.ts                   |   4 +-
 frontend/src/utils/platformColors.ts          |  65 +-
 72 files changed, 5754 insertions(+), 125 deletions(-)
 create mode 100644 backend/internal/handler/admin/cn_provider_handler.go
 create mode 100644 backend/internal/service/cn_provider_balance_check_service.go
 create mode 100644 backend/internal/service/cn_provider_balance_check_service_test.go
 create mode 100644 backend/internal/service/cn_provider_balance_service.go
 create mode 100644 backend/internal/service/cn_provider_quota_service.go
 create mode 100644 backend/internal/service/cn_providers_test.go
 create mode 100644 backend/internal/service/openai_gateway_chat_completions_anthropic_native.go
 create mode 100644 backend/internal/service/openai_gateway_messages_anthropic_native.go
 create mode 100644 backend/internal/service/openai_gateway_responses_anthropic_native.go
 create mode 100644 backend/internal/service/ratelimit_cn_providers.go
 create mode 100644 docker-compose.yml
 create mode 100644 frontend/src/api/admin/cnProviders.ts
 create mode 100644 frontend/src/components/account/CNProviderBalanceCell.vue
 create mode 100644 frontend/src/components/account/CNProviderQuotaCell.vue
 create mode 100644 frontend/src/components/account/CnBaseUrlPresets.vue

diff --git a/backend/cmd/server/wire.go b/backend/cmd/server/wire.go
index 8bd9ac5b1685..cf5984372dc1 100644
--- a/backend/cmd/server/wire.go
+++ b/backend/cmd/server/wire.go
@@ -88,6 +88,7 @@ func provideCleanup(
 	schedulerSnapshot *service.SchedulerSnapshotService,
 	tokenRefresh *service.TokenRefreshService,
 	accountExpiry *service.AccountExpiryService,
+	cnProviderBalanceCheck *service.CNProviderBalanceCheckService,
 	codexVersionSync *service.OpenAICodexVersionSyncService,
 	proxyExpiry *service.ProxyExpiryService,
 	subscriptionExpiry *service.SubscriptionExpiryService,
@@ -238,6 +239,12 @@ func provideCleanup(
 				accountExpiry.Stop()
 				return nil
 			}},
+			{"CNProviderBalanceCheckService", func() error {
+				if cnProviderBalanceCheck != nil {
+					cnProviderBalanceCheck.Stop()
+				}
+				return nil
+			}},
 			{"OpenAICodexVersionSyncService", func() error {
 				codexVersionSync.Stop()
 				return nil
diff --git a/backend/cmd/server/wire_gen.go b/backend/cmd/server/wire_gen.go
index ab9f4530b866..2b0caf982c4a 100644
--- a/backend/cmd/server/wire_gen.go
+++ b/backend/cmd/server/wire_gen.go
@@ -218,6 +218,9 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	antigravityOAuthHandler := admin.NewAntigravityOAuthHandler(antigravityOAuthService)
 	tokenRefreshService := service.ProvideTokenRefreshService(accountRepository, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, grokOAuthService, compositeTokenCacheInvalidator, schedulerCache, configConfig, tempUnschedCache, privacyClientFactory, proxyRepository, oAuthRefreshAPI, openAIGatewayService)
 	grokOAuthHandler := admin.NewGrokOAuthHandler(grokOAuthService, adminService, grokQuotaService, tokenRefreshService)
+	cnProviderQuotaService := service.ProvideCNProviderQuotaService(accountRepository, proxyRepository, httpUpstream, configConfig)
+	cnProviderBalanceService := service.ProvideCNProviderBalanceService(accountRepository, proxyRepository, httpUpstream, configConfig)
+	cnProviderHandler := admin.NewCNProviderHandler(cnProviderQuotaService, cnProviderBalanceService)
 	proxyHandler := admin.NewProxyHandler(adminService)
 	adminRedeemHandler := admin.NewRedeemHandler(adminService, redeemService)
 	promoHandler := admin.NewPromoHandler(promoService)
@@ -277,7 +280,7 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	auditLogHandler := admin.NewAuditLogHandler(auditLogService, totpService)
 	upstreamBillingProbeService := service.ProvideUpstreamBillingProbeService(accountRepository, accountTestService, settingService, leaderLockCache, db)
 	ollamaCloudUsageService := service.ProvideOllamaCloudUsageService(accountRepository, httpUpstream, settingService, secretEncryptor, configConfig, leaderLockCache, db)
-	adminHandlers := handler.ProvideAdminHandlers(dashboardHandler, adminUserHandler, groupHandler, accountHandler, adminAnnouncementHandler, dataManagementHandler, backupHandler, oAuthHandler, openAIOAuthHandler, geminiOAuthHandler, antigravityOAuthHandler, grokOAuthHandler, proxyHandler, adminRedeemHandler, promoHandler, settingHandler, opsHandler, systemHandler, adminSubscriptionHandler, adminUsageHandler, userAttributeHandler, errorPassthroughHandler, tlsFingerprintProfileHandler, adminAPIKeyHandler, scheduledTestHandler, channelHandler, channelMonitorHandler, channelMonitorRequestTemplateHandler, contentModerationHandler, promptAdminHandler, paymentHandler, affiliateHandler, complianceHandler, auditLogHandler, upstreamBillingProbeService, ollamaCloudUsageService)
+	adminHandlers := handler.ProvideAdminHandlers(dashboardHandler, adminUserHandler, groupHandler, accountHandler, adminAnnouncementHandler, dataManagementHandler, backupHandler, oAuthHandler, openAIOAuthHandler, geminiOAuthHandler, antigravityOAuthHandler, grokOAuthHandler, cnProviderHandler, proxyHandler, adminRedeemHandler, promoHandler, settingHandler, opsHandler, systemHandler, adminSubscriptionHandler, adminUsageHandler, userAttributeHandler, errorPassthroughHandler, tlsFingerprintProfileHandler, adminAPIKeyHandler, scheduledTestHandler, channelHandler, channelMonitorHandler, channelMonitorRequestTemplateHandler, contentModerationHandler, promptAdminHandler, paymentHandler, affiliateHandler, complianceHandler, auditLogHandler, upstreamBillingProbeService, ollamaCloudUsageService)
 	usageRecordWorkerPool := service.NewUsageRecordWorkerPool(configConfig)
 	userMsgQueueCache := repository.NewUserMsgQueueCache(redisClient)
 	userMessageQueueService := service.ProvideUserMessageQueueService(userMsgQueueCache, rpmCache, configConfig)
@@ -327,6 +330,7 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	opsScheduledReportService := service.ProvideOpsScheduledReportService(opsService, userService, emailService, redisClient, configConfig)
 	opsIngressRejectAggregator := service.ProvideOpsIngressRejectAggregator(opsRepository, opsService)
 	accountExpiryService := service.ProvideAccountExpiryService(accountRepository)
+	cnProviderBalanceCheckService := service.ProvideCNProviderBalanceCheckService(accountRepository, cnProviderBalanceService, cnProviderQuotaService, configConfig)
 	openAICodexVersionSyncService := service.ProvideOpenAICodexVersionSyncService(settingRepository, settingService, gitHubReleaseClient)
 	proxyExpiryService := service.ProvideProxyExpiryService(proxyRepository)
 	subscriptionExpiryService := service.ProvideSubscriptionExpiryService(userSubscriptionRepository, settingRepository, notificationEmailService, leaderLockCache, db)
@@ -336,7 +340,7 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	channelMonitorRunner := service.ProvideChannelMonitorRunner(channelMonitorService, settingService)
 	channelMonitorV2Aggregator := service.ProvideChannelMonitorV2Aggregator(channelMonitorV2Repository, db, settingService)
 	userPlatformQuotaUsageFlusher := service.ProvideUserPlatformQuotaUsageFlusher(configConfig, billingCache, serviceUserPlatformQuotaRepository, timingWheelService)
-	v := provideCleanup(client, redisClient, opsMetricsCollector, opsAggregationService, opsAlertEvaluatorService, opsCleanupService, opsScheduledReportService, opsSystemLogSink, opsService, opsIngressRejectAggregator, apiKeyService, authCacheInvalidationWorker, schedulerSnapshotService, tokenRefreshService, accountExpiryService, openAICodexVersionSyncService, proxyExpiryService, subscriptionExpiryService, usageCleanupService, idempotencyCleanupService, batchImageCleanupService, batchImageWorkerRuntime, pricingService, emailQueueService, billingCacheService, usageRecordWorkerPool, subscriptionService, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, grokOAuthService, openAIGatewayService, scheduledTestRunnerService, backupService, paymentOrderExpiryService, channelMonitorRunner, channelMonitorV2Aggregator, userPlatformQuotaUsageFlusher, upstreamBillingProbeService, ollamaCloudUsageService, auditLogService, promptService)
+	v := provideCleanup(client, redisClient, opsMetricsCollector, opsAggregationService, opsAlertEvaluatorService, opsCleanupService, opsScheduledReportService, opsSystemLogSink, opsService, opsIngressRejectAggregator, apiKeyService, authCacheInvalidationWorker, schedulerSnapshotService, tokenRefreshService, accountExpiryService, cnProviderBalanceCheckService, openAICodexVersionSyncService, proxyExpiryService, subscriptionExpiryService, usageCleanupService, idempotencyCleanupService, batchImageCleanupService, batchImageWorkerRuntime, pricingService, emailQueueService, billingCacheService, usageRecordWorkerPool, subscriptionService, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, grokOAuthService, openAIGatewayService, scheduledTestRunnerService, backupService, paymentOrderExpiryService, channelMonitorRunner, channelMonitorV2Aggregator, userPlatformQuotaUsageFlusher, upstreamBillingProbeService, ollamaCloudUsageService, auditLogService, promptService)
 	application := &Application{
 		Server:      httpServer,
 		PromptAudit: promptService,
@@ -380,6 +384,7 @@ func provideCleanup(
 	schedulerSnapshot *service.SchedulerSnapshotService,
 	tokenRefresh *service.TokenRefreshService,
 	accountExpiry *service.AccountExpiryService,
+	cnProviderBalanceCheck *service.CNProviderBalanceCheckService,
 	codexVersionSync *service.OpenAICodexVersionSyncService,
 	proxyExpiry *service.ProxyExpiryService,
 	subscriptionExpiry *service.SubscriptionExpiryService,
@@ -529,6 +534,12 @@ func provideCleanup(
 				accountExpiry.Stop()
 				return nil
 			}},
+			{"CNProviderBalanceCheckService", func() error {
+				if cnProviderBalanceCheck != nil {
+					cnProviderBalanceCheck.Stop()
+				}
+				return nil
+			}},
 			{"OpenAICodexVersionSyncService", func() error {
 				codexVersionSync.Stop()
 				return nil
diff --git a/backend/cmd/server/wire_gen_test.go b/backend/cmd/server/wire_gen_test.go
index dff3a9129efb..41e41bce1319 100644
--- a/backend/cmd/server/wire_gen_test.go
+++ b/backend/cmd/server/wire_gen_test.go
@@ -66,6 +66,7 @@ func TestProvideCleanup_WithMinimalDependencies_NoPanic(t *testing.T) {
 		schedulerSnapshotSvc,
 		tokenRefreshSvc,
 		accountExpirySvc,
+		nil, // cnProviderBalanceCheck
 		codexVersionSyncSvc,
 		proxyExpirySvc,
 		subscriptionExpirySvc,
diff --git a/backend/ent/schema/user_platform_quota.go b/backend/ent/schema/user_platform_quota.go
index a0b5598600c3..123d7c0bfa3f 100644
--- a/backend/ent/schema/user_platform_quota.go
+++ b/backend/ent/schema/user_platform_quota.go
@@ -41,7 +41,8 @@ func (UserPlatformQuota) Fields() []ent.Field {
 				// 注意：平台列表的单一权威源为 service.AllowedQuotaPlatforms；
 				// 此处为 ent 构建期约束，需与 service.AllowedQuotaPlatforms 保持同步。
 				switch s {
-				case "anthropic", "openai", "gemini", "antigravity", "grok":
+				case "anthropic", "openai", "gemini", "antigravity", "grok",
+					"kimi", "zhipu", "deepseek":
 					return nil
 				default:
 					return fmt.Errorf("platform %q is not allowed", s)
diff --git a/backend/internal/config/config.go b/backend/internal/config/config.go
index 2c5d5cf545fe..e437b1115001 100644
--- a/backend/internal/config/config.go
+++ b/backend/internal/config/config.go
@@ -1025,6 +1025,10 @@ type GatewayConfig struct {
 
 	// Grok: Grok/xAI gateway scheduling and free-tier soft-gate settings.
 	Grok GatewayGrokConfig `mapstructure:"grok"`
+
+	// CNProviders: 国产 OpenAI 兼容供应商（kimi/zhipu/deepseek）的余额检测配置。
+	// 仅作用于 payg（按量付费）账号：周期探测余额，低于阈值则临时停调。
+	CNProviders GatewayCNProvidersConfig `mapstructure:"cn_providers"`
 }
 
 // GatewayGrokConfig holds Grok-specific gateway scheduling knobs.
@@ -1057,6 +1061,18 @@ type GatewayGrokConfig struct {
 	FreeQuotaStatsCacheSeconds int `mapstructure:"free_quota_stats_cache_seconds"`
 }
 
+// GatewayCNProvidersConfig 国产 OpenAI 兼容供应商（kimi/zhipu/deepseek）的余额检测配置。
+//
+// 仅作用于 payg（按量付费）账号（kimi/deepseek 有公开余额端点；zhipu 无，仅靠响应式 429/402）。
+//   - balance_check_enabled: 是否启用周期余额检测（默认 true）
+//   - balance_threshold: 余额低于此值（账户货币单位，默认 0.5）触发临时停调
+//   - balance_check_interval_minutes: 余额检测周期（分钟，默认 10）
+type GatewayCNProvidersConfig struct {
+	BalanceCheckEnabled         bool    `mapstructure:"balance_check_enabled"`
+	BalanceThreshold            float64 `mapstructure:"balance_threshold"`
+	BalanceCheckIntervalMinutes int     `mapstructure:"balance_check_interval_minutes"`
+}
+
 type GatewayLiveConfig struct {
 	// MaxSessionDurationSeconds 是 Live 会话的硬上限。
 	MaxSessionDurationSeconds int `mapstructure:"max_session_duration_seconds"`
@@ -2355,6 +2371,10 @@ func setDefaults() {
 	viper.SetDefault("gateway.grok.free_quota_soft_gate_percent", 95)
 	viper.SetDefault("gateway.grok.free_quota_window_hours", 24)
 	viper.SetDefault("gateway.grok.free_quota_stats_cache_seconds", 60)
+	// 国产供应商余额检测（kimi/deepseek payg；zhipu 无余额端点，仅靠响应式 429/402）。
+	viper.SetDefault("gateway.cn_providers.balance_check_enabled", true)
+	viper.SetDefault("gateway.cn_providers.balance_threshold", 0.5)
+	viper.SetDefault("gateway.cn_providers.balance_check_interval_minutes", 10)
 	viper.SetDefault("gateway.image_concurrency.enabled", false)
 	viper.SetDefault("gateway.image_concurrency.max_concurrent_requests", 0)
 	viper.SetDefault("gateway.image_concurrency.overflow_mode", ImageConcurrencyOverflowModeReject)
diff --git a/backend/internal/domain/constants.go b/backend/internal/domain/constants.go
index 148c1cd5d301..3640d1b82e03 100644
--- a/backend/internal/domain/constants.go
+++ b/backend/internal/domain/constants.go
@@ -23,7 +23,27 @@ const (
 	PlatformGemini      = "gemini"
 	PlatformAntigravity = "antigravity"
 	PlatformGrok        = "grok"
-	PlatformComposite   = "composite"
+	// 国产 OpenAI 兼容供应商（经 OpenAI 网关转发，按 Chat Completions 协议）。
+	PlatformKimi      = "kimi"     // Kimi (月之暗面 / Moonshot)
+	PlatformZhipu     = "zhipu"    // 智谱 GLM (bigmodel)
+	PlatformDeepseek  = "deepseek" // DeepSeek
+	PlatformComposite = "composite"
+)
+
+// Account mode constants 区分国产供应商的「按量付费（余额）」与「Coding Plan」两种接入方式。
+// 存储于 credentials["account_mode"]，决定 base_url 预设与额度监控方式。
+const (
+	AccountModePayG   = "payg"   // 按量付费：消耗余额，做余额检测冷却
+	AccountModeCoding = "coding" // Coding Plan：滚动用量窗口冷却（5h / weekly）
+)
+
+// API protocol constants 国产供应商的上游 API 协议维度。存储于
+// credentials["api_protocol"]，与 account_mode 正交：协议决定转发端点与格式，
+// 模式决定额度监控方式。同协议请求零转换直通；跨协议组合才走转换链。
+const (
+	APIProtocolChatCompletions = "chat_completions" // OpenAI Chat Completions（默认）
+	APIProtocolAnthropic       = "anthropic"        // 原生 Anthropic /v1/messages（适配 Claude Code）
+	APIProtocolResponses       = "responses"        // OpenAI Responses（仅 deepseek，适配 Codex）
 )
 
 // Account type constants
diff --git a/backend/internal/handler/admin/account_handler.go b/backend/internal/handler/admin/account_handler.go
index 30fc6b982882..465c6a7816ae 100644
--- a/backend/internal/handler/admin/account_handler.go
+++ b/backend/internal/handler/admin/account_handler.go
@@ -1025,7 +1025,8 @@ func (h *AccountHandler) Update(c *gin.Context) {
 // 当前请求。探测错误仅记录日志，不向上下文传播：探测失败时标记保持缺失，
 // 网关会按"现状即证据"默认走 Responses。
 func (h *AccountHandler) scheduleOpenAIResponsesProbe(account *service.Account) {
-	if account == nil || account.Platform != service.PlatformOpenAI || account.Type != service.AccountTypeAPIKey {
+	if account == nil || account.Type != service.AccountTypeAPIKey ||
+		(account.Platform != service.PlatformOpenAI && !service.IsCNProvider(account.Platform)) {
 		return
 	}
 	if h.accountTestService == nil {
diff --git a/backend/internal/handler/admin/cn_provider_handler.go b/backend/internal/handler/admin/cn_provider_handler.go
new file mode 100644
index 000000000000..7c8d67848aa6
--- /dev/null
+++ b/backend/internal/handler/admin/cn_provider_handler.go
@@ -0,0 +1,69 @@
+package admin
+
+import (
+	"strconv"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/response"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+)
+
+// CNProviderHandler 暴露国产供应商（kimi/zhipu/deepseek）的额度与余额查询端点。
+//
+//   - GET /admin/cn-providers/accounts/:id/quota   Coding Plan 滚动窗口用量（kimi/zhipu）
+//   - GET /admin/cn-providers/accounts/:id/balance  payg 账号余额（kimi/deepseek）
+//
+// 智谱（zhipu）无余额端点，故同一账号仅 quota 或 balance 其一可用：服务端按账号
+// platform + account_mode 校验并返回明确错误（见 CNProvider*Service 的 load*Account）。
+type CNProviderHandler struct {
+	quotaService   *service.CNProviderQuotaService
+	balanceService *service.CNProviderBalanceService
+}
+
+func NewCNProviderHandler(
+	quotaService *service.CNProviderQuotaService,
+	balanceService *service.CNProviderBalanceService,
+) *CNProviderHandler {
+	return &CNProviderHandler{
+		quotaService:   quotaService,
+		balanceService: balanceService,
+	}
+}
+
+// QueryQuota 查询 Coding Plan 滚动窗口用量（5h + weekly）。
+func (h *CNProviderHandler) QueryQuota(c *gin.Context) {
+	accountID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid account ID")
+		return
+	}
+	if h == nil || h.quotaService == nil {
+		response.BadRequest(c, "cn provider quota service is not enabled")
+		return
+	}
+	result, err := h.quotaService.QueryUsage(c.Request.Context(), accountID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, result)
+}
+
+// QueryBalance 查询 payg 账号余额。
+func (h *CNProviderHandler) QueryBalance(c *gin.Context) {
+	accountID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid account ID")
+		return
+	}
+	if h == nil || h.balanceService == nil {
+		response.BadRequest(c, "cn provider balance service is not enabled")
+		return
+	}
+	result, err := h.balanceService.QueryBalance(c.Request.Context(), accountID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, result)
+}
diff --git a/backend/internal/handler/admin/user_platform_quota_admin_test.go b/backend/internal/handler/admin/user_platform_quota_admin_test.go
index 0211480b8281..f1d1b5b605fb 100644
--- a/backend/internal/handler/admin/user_platform_quota_admin_test.go
+++ b/backend/internal/handler/admin/user_platform_quota_admin_test.go
@@ -112,12 +112,13 @@ func TestUpdateUserPlatformQuotas_Success(t *testing.T) {
 	if len(repo.upsertCalls) != 1 {
 		t.Fatalf("UpsertForUser should be called once, got %d", len(repo.upsertCalls))
 	}
-	if repo.upsertCalls[0].userID != 42 || len(repo.upsertCalls[0].records) != len(service.AllowedQuotaPlatforms) {
+	// upsert 记录数 = 请求体中给出的平台数（未给出的平台不落库）。
+	if repo.upsertCalls[0].userID != 42 || len(repo.upsertCalls[0].records) != 5 {
 		t.Errorf("unexpected upsert call: %+v", repo.upsertCalls[0])
 	}
-	// 缓存失效：按全部允许平台统一失效。
-	if len(cache.deleteCalls) != 5 {
-		t.Errorf("expected 5 cache delete calls, got %d: %+v", len(cache.deleteCalls), cache.deleteCalls)
+	// 缓存失效：按全部允许平台统一失效（含 kimi/zhipu/deepseek）。
+	if len(cache.deleteCalls) != len(service.AllowedQuotaPlatforms) {
+		t.Errorf("expected %d cache delete calls, got %d: %+v", len(service.AllowedQuotaPlatforms), len(cache.deleteCalls), cache.deleteCalls)
 	}
 }
 
diff --git a/backend/internal/handler/handler.go b/backend/internal/handler/handler.go
index 359aff007188..7c0d3921ebd7 100644
--- a/backend/internal/handler/handler.go
+++ b/backend/internal/handler/handler.go
@@ -19,6 +19,7 @@ type AdminHandlers struct {
 	GeminiOAuth            *admin.GeminiOAuthHandler
 	AntigravityOAuth       *admin.AntigravityOAuthHandler
 	GrokOAuth              *admin.GrokOAuthHandler
+	CNProvider             *admin.CNProviderHandler
 	Proxy                  *admin.ProxyHandler
 	Redeem                 *admin.RedeemHandler
 	Promo                  *admin.PromoHandler
diff --git a/backend/internal/handler/openai_gateway_handler.go b/backend/internal/handler/openai_gateway_handler.go
index 1695eccc1ba5..28515d98f0d7 100644
--- a/backend/internal/handler/openai_gateway_handler.go
+++ b/backend/internal/handler/openai_gateway_handler.go
@@ -164,13 +164,11 @@ func wrapUsageRecordTaskContext(parent context.Context, task service.UsageRecord
 
 func openAICompatibleRequestPlatform(ctx context.Context, apiKey *service.APIKey) string {
 	if platform, ok := service.ResolvedTargetPlatformFromContext(ctx); ok {
-		if platform == service.PlatformGrok {
-			return service.PlatformGrok
-		}
-		return service.PlatformOpenAI
+		// 保留 grok 与国产供应商原值，其他归一为 openai（与调度器精确匹配语义一致）。
+		return service.NormalizeOpenAICompatiblePlatform(platform)
 	}
-	if apiKey != nil && apiKey.Group != nil && apiKey.Group.Platform == service.PlatformGrok {
-		return service.PlatformGrok
+	if apiKey != nil && apiKey.Group != nil {
+		return service.NormalizeOpenAICompatiblePlatform(apiKey.Group.Platform)
 	}
 	return service.PlatformOpenAI
 }
@@ -203,7 +201,9 @@ func allowOpenAICompatibleMessagesDispatch(apiKey *service.APIKey) bool {
 }
 
 func openAICompatibleTextTargetAllowed(c *gin.Context, apiKey *service.APIKey, model string) bool {
-	return compositeTargetPlatformAllowed(c, apiKey, model, service.PlatformOpenAI, service.PlatformGrok)
+	return compositeTargetPlatformAllowed(c, apiKey, model,
+		service.PlatformOpenAI, service.PlatformGrok,
+		service.PlatformKimi, service.PlatformZhipu, service.PlatformDeepseek)
 }
 
 // NewOpenAIGatewayHandler creates a new OpenAIGatewayHandler
diff --git a/backend/internal/handler/wire.go b/backend/internal/handler/wire.go
index 36fdb8528014..1eef8c0019b3 100644
--- a/backend/internal/handler/wire.go
+++ b/backend/internal/handler/wire.go
@@ -23,6 +23,7 @@ func ProvideAdminHandlers(
 	geminiOAuthHandler *admin.GeminiOAuthHandler,
 	antigravityOAuthHandler *admin.AntigravityOAuthHandler,
 	grokOAuthHandler *admin.GrokOAuthHandler,
+	cnProviderHandler *admin.CNProviderHandler,
 	proxyHandler *admin.ProxyHandler,
 	redeemHandler *admin.RedeemHandler,
 	promoHandler *admin.PromoHandler,
@@ -63,6 +64,7 @@ func ProvideAdminHandlers(
 		GeminiOAuth:            geminiOAuthHandler,
 		AntigravityOAuth:       antigravityOAuthHandler,
 		GrokOAuth:              grokOAuthHandler,
+		CNProvider:             cnProviderHandler,
 		Proxy:                  proxyHandler,
 		Redeem:                 redeemHandler,
 		Promo:                  promoHandler,
@@ -253,6 +255,7 @@ var ProviderSet = wire.NewSet(
 	admin.NewGeminiOAuthHandler,
 	admin.NewAntigravityOAuthHandler,
 	admin.NewGrokOAuthHandler,
+	admin.NewCNProviderHandler,
 	admin.NewProxyHandler,
 	admin.NewRedeemHandler,
 	admin.NewPromoHandler,
diff --git a/backend/internal/server/api_contract_test.go b/backend/internal/server/api_contract_test.go
index 79dfb919a68b..b0b1df7bc0f4 100644
--- a/backend/internal/server/api_contract_test.go
+++ b/backend/internal/server/api_contract_test.go
@@ -861,7 +861,7 @@ func TestAPIContracts(t *testing.T) {
 					"force_email_on_third_party_signup": false,
 					"default_concurrency": 5,
 					"default_balance": 1.25,
-					"default_platform_quotas": {"anthropic":{"daily":null,"weekly":null,"monthly":null},"antigravity":{"daily":null,"weekly":null,"monthly":null},"gemini":{"daily":null,"weekly":null,"monthly":null},"grok":{"daily":null,"weekly":null,"monthly":null},"openai":{"daily":null,"weekly":null,"monthly":null}},
+					"default_platform_quotas": {"anthropic":{"daily":null,"weekly":null,"monthly":null},"antigravity":{"daily":null,"weekly":null,"monthly":null},"deepseek":{"daily":null,"weekly":null,"monthly":null},"gemini":{"daily":null,"weekly":null,"monthly":null},"grok":{"daily":null,"weekly":null,"monthly":null},"kimi":{"daily":null,"weekly":null,"monthly":null},"openai":{"daily":null,"weekly":null,"monthly":null},"zhipu":{"daily":null,"weekly":null,"monthly":null}},
 					"auth_source_default_email_platform_quotas": null,
 					"auth_source_default_github_platform_quotas": null,
 					"auth_source_default_google_platform_quotas": null,
@@ -1175,7 +1175,7 @@ func TestAPIContracts(t *testing.T) {
 					"purchase_subscription_url": "",
 					"table_default_page_size": 20,
 					"table_page_size_options": [10, 20, 50],
-					"default_platform_quotas": {"anthropic":{"daily":null,"weekly":null,"monthly":null},"antigravity":{"daily":null,"weekly":null,"monthly":null},"gemini":{"daily":null,"weekly":null,"monthly":null},"grok":{"daily":null,"weekly":null,"monthly":null},"openai":{"daily":null,"weekly":null,"monthly":null}},
+					"default_platform_quotas": {"anthropic":{"daily":null,"weekly":null,"monthly":null},"antigravity":{"daily":null,"weekly":null,"monthly":null},"deepseek":{"daily":null,"weekly":null,"monthly":null},"gemini":{"daily":null,"weekly":null,"monthly":null},"grok":{"daily":null,"weekly":null,"monthly":null},"kimi":{"daily":null,"weekly":null,"monthly":null},"openai":{"daily":null,"weekly":null,"monthly":null},"zhipu":{"daily":null,"weekly":null,"monthly":null}},
 					"auth_source_default_email_platform_quotas": null,
 					"auth_source_default_github_platform_quotas": null,
 					"auth_source_default_google_platform_quotas": null,
diff --git a/backend/internal/server/routes/admin.go b/backend/internal/server/routes/admin.go
index 149d4f95bcef..60d6516bd024 100644
--- a/backend/internal/server/routes/admin.go
+++ b/backend/internal/server/routes/admin.go
@@ -58,6 +58,9 @@ func RegisterAdminRoutes(
 		// Grok OAuth
 		registerGrokOAuthRoutes(admin, h)
 
+		// 国产供应商（kimi/zhipu/deepseek）额度与余额
+		registerCNProviderRoutes(admin, h)
+
 		// 代理管理
 		registerProxyRoutes(admin, h, stepUpAuth)
 
@@ -482,6 +485,17 @@ func registerGrokOAuthRoutes(admin *gin.RouterGroup, h *handler.Handlers) {
 	}
 }
 
+// registerCNProviderRoutes 注册国产供应商（kimi/zhipu/deepseek）的额度与余额查询端点。
+func registerCNProviderRoutes(admin *gin.RouterGroup, h *handler.Handlers) {
+	cn := admin.Group("/cn-providers")
+	{
+		// Coding Plan 滚动窗口用量（kimi/zhipu coding 账号）。
+		cn.GET("/accounts/:id/quota", h.Admin.CNProvider.QueryQuota)
+		// payg 账号余额（kimi/deepseek；zhipu 无余额端点）。
+		cn.GET("/accounts/:id/balance", h.Admin.CNProvider.QueryBalance)
+	}
+}
+
 func registerProxyRoutes(admin *gin.RouterGroup, h *handler.Handlers, stepUpAuth middleware.StepUpAuthMiddleware) {
 	proxies := admin.Group("/proxies")
 	{
diff --git a/backend/internal/server/routes/gateway.go b/backend/internal/server/routes/gateway.go
index 6d99d2c7ea39..3c4b33f519fc 100644
--- a/backend/internal/server/routes/gateway.go
+++ b/backend/internal/server/routes/gateway.go
@@ -47,7 +47,9 @@ func RegisterGatewayRoutes(
 
 	isOpenAIResponsesCompatibleGatewayPlatform := func(c *gin.Context) bool {
 		switch getGroupPlatform(c) {
-		case service.PlatformOpenAI, service.PlatformGrok:
+		case service.PlatformOpenAI, service.PlatformGrok,
+			service.PlatformKimi, service.PlatformZhipu, service.PlatformDeepseek:
+			// 国产 OpenAI 兼容供应商（kimi/zhipu/deepseek）与 openai/grok 一样经 OpenAI 网关转发。
 			return true
 		default:
 			return false
@@ -58,7 +60,7 @@ func RegisterGatewayRoutes(
 	}
 	countTokensHandler := func(c *gin.Context) {
 		switch getGroupPlatform(c) {
-		case service.PlatformOpenAI:
+		case service.PlatformOpenAI, service.PlatformKimi, service.PlatformZhipu, service.PlatformDeepseek:
 			h.OpenAIGateway.CountTokens(c)
 		case service.PlatformGrok:
 			h.OpenAIGateway.GrokCountTokens(c)
diff --git a/backend/internal/service/account.go b/backend/internal/service/account.go
index cb74aaf50f9e..cd66ef89ce3d 100644
--- a/backend/internal/service/account.go
+++ b/backend/internal/service/account.go
@@ -271,8 +271,30 @@ func (a *Account) IsGrokOAuth() bool {
 	return a.IsGrok() && a.Type == AccountTypeOAuth
 }
 
+// IsKimi / IsZhipu / IsDeepseek 标识国产 OpenAI 兼容供应商账号。
+func (a *Account) IsKimi() bool {
+	return a.Platform == PlatformKimi
+}
+
+func (a *Account) IsZhipu() bool {
+	return a.Platform == PlatformZhipu
+}
+
+func (a *Account) IsDeepseek() bool {
+	return a.Platform == PlatformDeepseek
+}
+
+// IsCNProvider 报告是否为国产 OpenAI 兼容供应商（kimi/zhipu/deepseek）。
+func (a *Account) IsCNProvider() bool {
+	return a != nil && IsCNProvider(a.Platform)
+}
+
+// IsOpenAICompatible 报告账号是否走 OpenAI 网关（OpenAI 协议族）。
+// openai/grok 原生走 OpenAI 网关；kimi/zhipu/deepseek 同为 OpenAI Chat Completions
+// 兼容上游，也经 OpenAI 网关转发。
 func (a *Account) IsOpenAICompatible() bool {
-	return a != nil && (a.Platform == PlatformOpenAI || a.Platform == PlatformGrok)
+	return a != nil && (a.Platform == PlatformOpenAI || a.Platform == PlatformGrok ||
+		a.Platform == PlatformKimi || a.Platform == PlatformZhipu || a.Platform == PlatformDeepseek)
 }
 
 func (a *Account) GeminiOAuthType() string {
@@ -1281,17 +1303,161 @@ func (a *Account) IsOpenAIApiKey() bool {
 	return a.IsOpenAI() && a.Type == AccountTypeAPIKey
 }
 
+// GetOpenAIBaseURL 解析 OpenAI 协议族账号的上游 base_url。
+// 适用 openai 与国产 OpenAI 兼容供应商（kimi/zhipu/deepseek）；grok 走 GetGrokBaseURL，
+// 此处对 grok 返回 "" 以保持原有行为。
 func (a *Account) GetOpenAIBaseURL() string {
-	if !a.IsOpenAI() {
+	if !a.IsOpenAI() && !a.IsCNProvider() {
+		return ""
+	}
+	if a.Type == AccountTypeAPIKey || a.Type == AccountTypeUpstream {
+		if baseURL := strings.TrimSpace(a.GetCredential("base_url")); baseURL != "" {
+			return baseURL
+		}
+	}
+	// 平台默认 base_url：CN 供应商按 account_mode 选择 payg / coding 默认值。
+	switch a.Platform {
+	case PlatformKimi:
+		if a.GetAccountMode() == AccountModeCoding {
+			return DefaultKimiCodingBaseURL
+		}
+		return DefaultKimiPayGBaseURL
+	case PlatformZhipu:
+		if a.GetAccountMode() == AccountModeCoding {
+			return DefaultZhipuCodingBaseURL
+		}
+		return DefaultZhipuPayGBaseURL
+	case PlatformDeepseek:
+		return DefaultDeepseekBaseURL
+	default:
+		return "https://api.openai.com"
+	}
+}
+
+// GetAccountMode 返回国产供应商账号的接入模式（payg / coding）；非国产供应商或未设置时
+// 返回空串。存储于 credentials["account_mode"]。
+func (a *Account) GetAccountMode() string {
+	if a == nil {
+		return ""
+	}
+	mode := strings.TrimSpace(a.GetCredential("account_mode"))
+	if mode == AccountModePayG || mode == AccountModeCoding {
+		return mode
+	}
+	return ""
+}
+
+// IsCodingPlan 报告账号是否为 Coding Plan 模式（用于滚动用量窗口冷却）。
+func (a *Account) IsCodingPlan() bool {
+	return a.GetAccountMode() == AccountModeCoding
+}
+
+// GetAPIProtocol 返回国产供应商账号的上游 API 协议。存储于
+// credentials["api_protocol"]；缺失或与平台不匹配时回退 chat_completions
+// （与既有行为完全一致）。responses 协议仅 deepseek 支持（官方原生 /responses
+// 端点，适配 Codex）；kimi/zhipu 无此端点。
+func (a *Account) GetAPIProtocol() string {
+	if a == nil || !a.IsCNProvider() {
+		return APIProtocolChatCompletions
+	}
+	switch strings.TrimSpace(a.GetCredential("api_protocol")) {
+	case APIProtocolAnthropic:
+		return APIProtocolAnthropic
+	case APIProtocolResponses:
+		if a.Platform == PlatformDeepseek {
+			return APIProtocolResponses
+		}
+	case APIProtocolChatCompletions:
+		return APIProtocolChatCompletions
+	}
+	return APIProtocolChatCompletions
+}
+
+// IsAnthropicProtocol 报告账号是否以原生 Anthropic 协议接入上游
+// （/v1/messages 直通，适配 Claude Code 等客户端）。
+func (a *Account) IsAnthropicProtocol() bool {
+	return a.GetAPIProtocol() == APIProtocolAnthropic
+}
+
+// GetAnthropicProtocolBaseURL 返回 Anthropic 协议账号的上游 base_url
+// （上游路径为 {base}/v1/messages）。优先取凭证 base_url，缺失时按
+// 供应商 × 接入模式返回默认端点。非 Anthropic 协议账号返回空串。
+func (a *Account) GetAnthropicProtocolBaseURL() string {
+	if a == nil || !a.IsAnthropicProtocol() {
 		return ""
 	}
-	if a.Type == AccountTypeAPIKey {
-		baseURL := a.GetCredential("base_url")
-		if baseURL != "" {
+	if a.Type == AccountTypeAPIKey || a.Type == AccountTypeUpstream {
+		if baseURL := strings.TrimSpace(a.GetCredential("base_url")); baseURL != "" {
 			return baseURL
 		}
 	}
-	return "https://api.openai.com"
+	switch a.Platform {
+	case PlatformKimi:
+		if a.GetAccountMode() == AccountModeCoding {
+			return DefaultKimiCodingAnthropicBaseURL
+		}
+		return DefaultKimiPayGAnthropicBaseURL
+	case PlatformZhipu:
+		return DefaultZhipuAnthropicBaseURL
+	case PlatformDeepseek:
+		return DefaultDeepseekAnthropicBaseURL
+	default:
+		return ""
+	}
+}
+
+// GetOpenAIFormatBaseURL 返回供 OpenAI 格式端点（/v1/models、/v1/chat/completions
+// 等）使用的 base。chat_completions / responses 协议下与 GetOpenAIBaseURL
+// 一致（凭证 base_url 或平台默认）；anthropic 协议下凭证 base_url 指向 Anthropic
+// 端点，不能拿来拼 OpenAI 路径，此时返回该供应商 × 模式的 Chat Completions
+// 默认 base（模型同步等协议族共用路径仍可用）。
+func (a *Account) GetOpenAIFormatBaseURL() string {
+	if a == nil || !a.IsAnthropicProtocol() {
+		return a.GetOpenAIBaseURL()
+	}
+	switch a.Platform {
+	case PlatformKimi:
+		if a.GetAccountMode() == AccountModeCoding {
+			return DefaultKimiCodingBaseURL
+		}
+		return DefaultKimiPayGBaseURL
+	case PlatformZhipu:
+		if a.GetAccountMode() == AccountModeCoding {
+			return DefaultZhipuCodingBaseURL
+		}
+		return DefaultZhipuPayGBaseURL
+	case PlatformDeepseek:
+		return DefaultDeepseekBaseURL
+	default:
+		return a.GetOpenAIBaseURL()
+	}
+}
+
+// GetCNAPIKey 返回国产 OpenAI 兼容供应商账号的 api_key 凭据（kimi/zhipu/deepseek）。
+// 与 openai 的 GetOpenAIApiKey 区分：后者仅对 openai 平台返回。
+func (a *Account) GetCNAPIKey() string {
+	if a == nil || !a.IsCNProvider() {
+		return ""
+	}
+	return a.GetCredential("api_key")
+}
+
+// GetCodingPlanProvider 根据 base_url 识别 Coding Plan 供应商（kimi / zhipu），
+// 用于路由到对应的额度查询端点。非 coding 模式或无法识别时返回空串。
+// 判定规则与 cc-switch coding_plan.rs::detect_provider 保持一致。
+func (a *Account) GetCodingPlanProvider() string {
+	if a == nil || a.GetAccountMode() != AccountModeCoding {
+		return ""
+	}
+	baseURL := strings.ToLower(a.GetOpenAIBaseURL())
+	switch {
+	case strings.Contains(baseURL, "api.kimi.com/coding"):
+		return PlatformKimi
+	case strings.Contains(baseURL, "bigmodel.cn"), strings.Contains(baseURL, "api.z.ai"):
+		return PlatformZhipu
+	default:
+		return ""
+	}
 }
 
 func (a *Account) GetOpenAIAccessToken() string {
@@ -1404,6 +1570,23 @@ func (a *Account) GetOpenAIApiKey() string {
 	return a.GetCredential("api_key")
 }
 
+// GetOpenAIProtocolAPIKey 返回 OpenAI 协议族 APIKey 账号的密钥。
+// 覆盖 openai 原生账号与国产 OpenAI 兼容供应商（kimi/zhipu/deepseek）账号，
+// 供转发鉴权、模型列表同步等协议族共用路径使用。注意 IsOpenAIApiKey 语义上
+// 仅指 openai 平台账号，调度倍率/WS 能力门控继续以其为准，不受本方法影响。
+func (a *Account) GetOpenAIProtocolAPIKey() string {
+	if a == nil {
+		return ""
+	}
+	if a.IsCNProvider() {
+		if a.Type != AccountTypeAPIKey {
+			return ""
+		}
+		return a.GetCredential("api_key")
+	}
+	return a.GetOpenAIApiKey()
+}
+
 func (a *Account) GetOpenAIUserAgent() string {
 	if !a.IsOpenAI() {
 		return ""
diff --git a/backend/internal/service/account_scheduling_threshold_eval.go b/backend/internal/service/account_scheduling_threshold_eval.go
index b8a9dae80d10..7dde1fd163b2 100644
--- a/backend/internal/service/account_scheduling_threshold_eval.go
+++ b/backend/internal/service/account_scheduling_threshold_eval.go
@@ -58,6 +58,10 @@ func EvaluateAccountSchedulingThreshold(account *Account, thresholds map[string]
 		winner = pickLatestResetSchedulingCandidate(anthropicThresholdCandidates(account), threshold, now)
 	case PlatformGrok:
 		winner = pickLatestResetSchedulingCandidate(grokThresholdCandidates(account), threshold, now)
+	case PlatformKimi:
+		winner = pickLatestResetSchedulingCandidate(cnProviderThresholdCandidates(account, PlatformKimi), threshold, now)
+	case PlatformZhipu:
+		winner = pickLatestResetSchedulingCandidate(cnProviderThresholdCandidates(account, PlatformZhipu), threshold, now)
 	default:
 		return decision
 	}
@@ -303,6 +307,45 @@ func grokThresholdCandidates(account *Account) []*accountSchedulingThresholdCand
 	}
 }
 
+// cnProviderThresholdCandidates 读取国产供应商 Coding Plan 账号的 5h / weekly 滚动窗口
+// 用量快照（由 CNProviderQuotaService 写入 account.Extra，键形如
+// <provider>_5h_used_percent / <provider>_weekly_reset_at）。payg 账号无此快照，
+// 候选为空 → 不触发阈值停调（余额型走余额检测）。与 openai 的快照驱动停调一致：
+// 仅当用量超阈值且窗口尚未重置时才停调。
+func cnProviderThresholdCandidates(account *Account, provider string) []*accountSchedulingThresholdCandidate {
+	if account == nil || len(account.Extra) == 0 {
+		return nil
+	}
+	return []*accountSchedulingThresholdCandidate{
+		cnThresholdCandidate(account.Extra, provider, "5h"),
+		cnThresholdCandidate(account.Extra, provider, "weekly"),
+	}
+}
+
+func cnThresholdCandidate(extra map[string]any, provider, window string) *accountSchedulingThresholdCandidate {
+	var usedKey, resetKey string
+	switch window {
+	case "5h":
+		usedKey = cnExtraKey(provider, cnExtraSuffix5hUsed)
+		resetKey = cnExtraKey(provider, cnExtraSuffix5hReset)
+	case "weekly":
+		usedKey = cnExtraKey(provider, cnExtraSuffixWeeklyUsed)
+		resetKey = cnExtraKey(provider, cnExtraSuffixWeeklyReset)
+	default:
+		return nil
+	}
+	usedPercent, ok := extra[usedKey]
+	if !ok {
+		return nil
+	}
+	return &accountSchedulingThresholdCandidate{
+		window:      window,
+		scope:       provider,
+		usedPercent: schedulingPercentValue(usedPercent),
+		until:       parseSchedulingResetAt(extra[resetKey]),
+	}
+}
+
 func pickLatestResetSchedulingCandidate(candidates []*accountSchedulingThresholdCandidate, threshold int, now time.Time) *accountSchedulingThresholdCandidate {
 	var winner *accountSchedulingThresholdCandidate
 	for _, candidate := range candidates {
diff --git a/backend/internal/service/account_service.go b/backend/internal/service/account_service.go
index 65d4d3f26846..be199cb2561c 100644
--- a/backend/internal/service/account_service.go
+++ b/backend/internal/service/account_service.go
@@ -509,6 +509,9 @@ func (s *AccountService) TestCredentials(ctx context.Context, id int64) error {
 	case PlatformGrok:
 		// Grok OAuth credentials are validated via token exchange/refresh and request-path probes.
 		return nil
+	case PlatformKimi, PlatformZhipu, PlatformDeepseek:
+		// 国产 OpenAI 兼容供应商：凭证为 API Key，实际可用性经余额/额度探测与转发路径验证。
+		return nil
 	default:
 		return fmt.Errorf("unsupported platform: %s", account.Platform)
 	}
diff --git a/backend/internal/service/anthropic_apikey_auth.go b/backend/internal/service/anthropic_apikey_auth.go
index 7752d6fe3a02..044d5be96ca7 100644
--- a/backend/internal/service/anthropic_apikey_auth.go
+++ b/backend/internal/service/anthropic_apikey_auth.go
@@ -14,9 +14,14 @@ const (
 
 // GetAnthropicAPIKeyAuthScheme returns the upstream authentication scheme for
 // Anthropic API-key accounts. Missing or invalid values keep the historical
-// x-api-key behavior.
+// x-api-key behavior. CN providers using their native Anthropic endpoints
+// (api_protocol=anthropic) share the same override knob — Kimi/DeepSeek default
+// to x-api-key, Zhipu can opt into Authorization: Bearer.
 func (a *Account) GetAnthropicAPIKeyAuthScheme() string {
-	if a == nil || a.Platform != PlatformAnthropic || a.Type != AccountTypeAPIKey {
+	if a == nil || a.Type != AccountTypeAPIKey {
+		return AnthropicAPIKeyAuthSchemeXAPIKey
+	}
+	if a.Platform != PlatformAnthropic && !a.IsCNProvider() {
 		return AnthropicAPIKeyAuthSchemeXAPIKey
 	}
 
diff --git a/backend/internal/service/bedrock_stream.go b/backend/internal/service/bedrock_stream.go
index 98196d27ec86..9ce6b87e242d 100644
--- a/backend/internal/service/bedrock_stream.go
+++ b/backend/internal/service/bedrock_stream.go
@@ -139,7 +139,7 @@ func (s *GatewayService) handleBedrockStreamingResponse(
 			sseData = transformBedrockInvocationMetrics(sseData)
 
 			// 解析 SSE 事件数据提取 usage
-			s.parseSSEUsagePassthrough(string(sseData), usage)
+			parseSSEUsagePassthrough(string(sseData), usage)
 
 			// 确定 SSE event type
 			eventType := gjson.GetBytes(sseData, "type").String()
diff --git a/backend/internal/service/cn_provider_balance_check_service.go b/backend/internal/service/cn_provider_balance_check_service.go
new file mode 100644
index 000000000000..d5ddd6712e89
--- /dev/null
+++ b/backend/internal/service/cn_provider_balance_check_service.go
@@ -0,0 +1,279 @@
+package service
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+)
+
+// cnQuotaProber 抽象额度探测（*CNProviderQuotaService 实现，测试可替换）。
+type cnQuotaProber interface {
+	QueryUsage(ctx context.Context, accountID int64) (*CNProviderQuotaProbeResult, error)
+}
+
+// cnQuotaProbeConcurrency 周期任务并发探测额度账号的并发度。
+const cnQuotaProbeConcurrency = 4
+
+// CNProviderBalanceCheckService 周期性探测国产供应商账号：
+//   - payg（按量付费）：余额低于阈值则临时停调，恢复则清除（仅清除本服务写入的停调）；
+//   - coding plan：调用 CNProviderQuotaService 探测 5h/weekly 滚动窗口并落 extra 快照，
+//     调度阈值评估（cnProviderThresholdCandidates）据此自动停调/恢复。
+//
+// 克隆自 AccountExpiryService 的 Start/Stop/runOnce + ticker 骨架。
+// 余额探测仅覆盖有公开余额端点的 kimi / deepseek；智谱无余额端点，仅靠响应式 429/402。
+// 额度探测覆盖 kimi / zhipu 的 coding plan 账号（deepseek 无 coding 套餐）。
+type CNProviderBalanceCheckService struct {
+	accountRepo    AccountRepository
+	balanceService *CNProviderBalanceService
+	quotaService   cnQuotaProber
+	cfg            *config.Config
+	interval       time.Duration
+	stopCh         chan struct{}
+	stopOnce       sync.Once
+	wg             sync.WaitGroup
+}
+
+// NewCNProviderBalanceCheckService 构造周期余额/额度检测服务。
+// interval <= 0 时 Start() 直接返回（不启动），便于通过配置关闭。
+func NewCNProviderBalanceCheckService(
+	accountRepo AccountRepository,
+	balanceService *CNProviderBalanceService,
+	quotaService *CNProviderQuotaService,
+	cfg *config.Config,
+	interval time.Duration,
+) *CNProviderBalanceCheckService {
+	return &CNProviderBalanceCheckService{
+		accountRepo:    accountRepo,
+		balanceService: balanceService,
+		quotaService:   quotaService,
+		cfg:            cfg,
+		interval:       interval,
+		stopCh:         make(chan struct{}),
+	}
+}
+
+func (s *CNProviderBalanceCheckService) Start() {
+	if s == nil || s.accountRepo == nil || s.balanceService == nil || s.cfg == nil {
+		return
+	}
+	if !s.cfg.Gateway.CNProviders.BalanceCheckEnabled {
+		return
+	}
+	if s.interval <= 0 {
+		return
+	}
+	log.Printf("[CNBalance] started (interval=%s threshold=%.2f)", s.interval, s.cfg.Gateway.CNProviders.BalanceThreshold)
+	s.wg.Add(1)
+	go func() {
+		defer s.wg.Done()
+		ticker := time.NewTicker(s.interval)
+		defer ticker.Stop()
+
+		// 启动后先等待一个周期再首次探测，避免与进程启动峰重叠。
+		for {
+			select {
+			case <-ticker.C:
+				s.runOnce()
+			case <-s.stopCh:
+				return
+			}
+		}
+	}()
+}
+
+func (s *CNProviderBalanceCheckService) Stop() {
+	if s == nil {
+		return
+	}
+	s.stopOnce.Do(func() {
+		close(s.stopCh)
+	})
+	s.wg.Wait()
+}
+
+func (s *CNProviderBalanceCheckService) runOnce() {
+	// 收集 coding 探测目标（kimi/deepseek + 智谱）与 payg 检查队列。
+	// coding 探测统一在收集完成后按 4 并发执行：单账号探测 15-20s，串行 ×
+	// 多账号会耗尽整体预算（120s 上限），排在后面的账号快照会饥饿，
+	// 连锁影响阈值停调的新鲜度判定。
+	type quotaTarget struct {
+		id       int64
+		platform string
+	}
+	var quotaTargets []quotaTarget
+	var paygTargets []*Account
+	collect := func(platform string, accounts []Account) {
+		for i := range accounts {
+			account := &accounts[i]
+			if !account.IsActive() {
+				continue
+			}
+			// coding 账号：探测滚动窗口并落快照（不要求 Schedulable——已被
+			// 阈值停调的账号也需要新鲜快照决定是否续停）。
+			if account.IsCodingPlan() {
+				quotaTargets = append(quotaTargets, quotaTarget{id: account.ID, platform: account.Platform})
+				continue
+			}
+			// payg 余额探测仅 kimi/deepseek（智谱无公开余额端点，payg 账号
+			// 依赖响应式 402/429 处理）。
+			if platform != PlatformZhipu && account.Schedulable {
+				paygTargets = append(paygTargets, account)
+			}
+		}
+	}
+	for _, platform := range s.platforms() {
+		accounts, err := s.accountRepo.ListByPlatform(context.Background(), platform)
+		if err != nil {
+			log.Printf("[CNBalance] list %s accounts failed: %v", platform, err)
+			continue
+		}
+		collect(platform, accounts)
+	}
+	// 智谱无余额端点，仅进额度探测。
+	if s.quotaService != nil {
+		accounts, err := s.accountRepo.ListByPlatform(context.Background(), PlatformZhipu)
+		if err != nil {
+			log.Printf("[CNBalance] list %s accounts failed: %v", PlatformZhipu, err)
+		} else {
+			collect(PlatformZhipu, accounts)
+		}
+	}
+
+	// 预算按工作量放大：4 并发 × 15s/批 + payg 每账号 5s，下限 30s 上限 300s。
+	batches := (len(quotaTargets) + cnQuotaProbeConcurrency - 1) / cnQuotaProbeConcurrency
+	timeout := 30*time.Second + time.Duration(batches)*15*time.Second + time.Duration(len(paygTargets))*5*time.Second
+	if timeout > 300*time.Second {
+		timeout = 300 * time.Second
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)
+	defer cancel()
+
+	threshold := s.cfg.Gateway.CNProviders.BalanceThreshold
+	paused, cleared := 0, 0
+	for _, account := range paygTargets {
+		switch s.checkOne(ctx, account, threshold) {
+		case cnBalancePaused:
+			paused++
+		case cnBalanceCleared:
+			cleared++
+		}
+	}
+
+	if len(quotaTargets) > 0 && s.quotaService != nil {
+		sem := make(chan struct{}, cnQuotaProbeConcurrency)
+		var wg sync.WaitGroup
+		for _, target := range quotaTargets {
+			wg.Add(1)
+			go func(t quotaTarget) {
+				defer wg.Done()
+				sem <- struct{}{}
+				defer func() { <-sem }()
+				s.probeQuota(ctx, t.id, t.platform)
+			}(target)
+		}
+		wg.Wait()
+	}
+
+	if paused > 0 || cleared > 0 {
+		log.Printf("[CNBalance] paused=%d cleared=%d (threshold=%.2f)", paused, cleared, threshold)
+	}
+}
+
+// probeQuota 探测单个 coding plan 账号的滚动窗口用量并落 extra 快照。
+// 不在此处做停调/恢复决策：调度阈值评估读取快照统一判定（含暂停账号的续停）。
+func (s *CNProviderBalanceCheckService) probeQuota(ctx context.Context, accountID int64, platform string) {
+	if s.quotaService == nil {
+		return
+	}
+	result, err := s.quotaService.QueryUsage(ctx, accountID)
+	if err != nil {
+		log.Printf("[CNBalance] quota probe account %d (%s) failed: %v", accountID, platform, err)
+		return
+	}
+	if result != nil && !result.Success && result.Error != "" {
+		log.Printf("[CNBalance] quota probe account %d (%s) error: %s", accountID, platform, result.Error)
+	}
+}
+
+type cnBalanceCheckOutcome int
+
+const (
+	cnBalanceNoChange cnBalanceCheckOutcome = iota
+	cnBalancePaused
+	cnBalanceCleared
+)
+
+// checkOne 探测单账号余额并决定停调/恢复。探测失败时不动现状（避免瞬时网络抖动
+// 误解除或误停调）。
+func (s *CNProviderBalanceCheckService) checkOne(ctx context.Context, account *Account, threshold float64) cnBalanceCheckOutcome {
+	result, err := s.balanceService.QueryBalance(ctx, account.ID)
+	if err != nil || result == nil || !result.Success {
+		return cnBalanceNoChange
+	}
+
+	// 双币种（deepseek CNY+USD）任一币种余额达标即可继续调度；仅当全部低于
+	// 阈值（或不可用）才停调。
+	low := !result.Available || allCNBalancesBelowThreshold(result, threshold)
+	if low {
+		// 已被（任何来源）停调时不覆盖其 reason。
+		if !account.IsSchedulable() {
+			return cnBalanceNoChange
+		}
+		reason := cnBalanceLowReason(fmt.Sprintf("余额 %.4g %s 低于阈值 %.2f", result.Balance, result.Currency, threshold))
+		if err := s.accountRepo.SetTempUnschedulable(ctx, account.ID, time.Now().Add(s.cooldown()), reason); err != nil {
+			log.Printf("[CNBalance] pause account %d failed: %v", account.ID, err)
+			return cnBalanceNoChange
+		}
+		log.Printf("[CNBalance] paused account %d (%s): balance=%.4g %s", account.ID, account.Platform, result.Balance, result.Currency)
+		return cnBalancePaused
+	}
+
+	// 余额健康：仅清除「本服务写入」的临时停调（reason 前缀匹配），不触碰其他子系统。
+	if account.TempUnschedulableUntil != nil && strings.HasPrefix(account.TempUnschedulableReason, cnBalanceLowReasonPrefix) {
+		if err := s.accountRepo.ClearTempUnschedulable(ctx, account.ID); err != nil {
+			log.Printf("[CNBalance] clear account %d failed: %v", account.ID, err)
+			return cnBalanceNoChange
+		}
+		log.Printf("[CNBalance] reactivated account %d (%s): balance=%.4g %s", account.ID, account.Platform, result.Balance, result.Currency)
+		return cnBalanceCleared
+	}
+	return cnBalanceNoChange
+}
+
+func (s *CNProviderBalanceCheckService) platforms() []string {
+	return []string{PlatformKimi, PlatformDeepseek}
+}
+
+// allCNBalancesBelowThreshold 判断全部币种余额是否均低于阈值。
+// 无明细时退回主币种判定（与旧行为一致）。
+func allCNBalancesBelowThreshold(result *CNProviderBalanceResult, threshold float64) bool {
+	if len(result.Balances) == 0 {
+		return result.Balance < threshold
+	}
+	for _, entry := range result.Balances {
+		if entry.Balance >= threshold {
+			return false
+		}
+	}
+	return true
+}
+
+// cooldown 返回临时停调持续时长（= 2× 检测周期），与响应式 402/429 路径一致。
+func (s *CNProviderBalanceCheckService) cooldown() time.Duration {
+	minutes := 10
+	if s.cfg != nil {
+		if cfgMin := s.cfg.Gateway.CNProviders.BalanceCheckIntervalMinutes; cfgMin > 0 {
+			minutes = cfgMin
+		}
+	}
+	cooldown := time.Duration(minutes) * time.Minute * 2
+	if cooldown < time.Minute {
+		cooldown = 10 * time.Minute
+	}
+	return cooldown
+}
diff --git a/backend/internal/service/cn_provider_balance_check_service_test.go b/backend/internal/service/cn_provider_balance_check_service_test.go
new file mode 100644
index 000000000000..26f7b69ea296
--- /dev/null
+++ b/backend/internal/service/cn_provider_balance_check_service_test.go
@@ -0,0 +1,101 @@
+package service
+
+import (
+	"context"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/stretchr/testify/require"
+)
+
+// 周期任务对 coding plan 账号的额度探测行为（runOnce 集成路径）：
+//   - kimi coding 账号（含已被阈值停调的）→ 额度探测被调用；
+//   - 智谱 coding 账号 → 额度探测被调用（智谱不进 kimi/deepseek 余额循环）；
+//   - payg 账号不经过额度探测（走余额路径，本测试不放 payg 账号避免真实网络）；
+//   - 非激活账号完全跳过。
+
+type fakeCNQuotaProber struct {
+	probed []int64
+}
+
+func (f *fakeCNQuotaProber) QueryUsage(ctx context.Context, accountID int64) (*CNProviderQuotaProbeResult, error) {
+	f.probed = append(f.probed, accountID)
+	return &CNProviderQuotaProbeResult{Success: true, Persisted: true}, nil
+}
+
+type fakeCNCheckRepo struct {
+	AccountRepository
+	byPlatform map[string][]Account
+}
+
+func (r *fakeCNCheckRepo) ListByPlatform(ctx context.Context, platform string) ([]Account, error) {
+	return r.byPlatform[platform], nil
+}
+
+func TestCNProviderBalanceCheckRunOnceProbesCodingPlanQuota(t *testing.T) {
+	kimiActive := Account{ID: 1, Platform: PlatformKimi, Type: AccountTypeAPIKey, Status: StatusActive,
+		Credentials: map[string]any{"account_mode": "coding"}}
+	// 已被阈值停调的 coding 账号也要刷新快照（决定是否续停）。
+	kimiPaused := Account{ID: 2, Platform: PlatformKimi, Type: AccountTypeAPIKey, Status: StatusActive, Schedulable: false,
+		Credentials: map[string]any{"account_mode": "coding"}}
+	// 非激活账号跳过。
+	kimiInactive := Account{ID: 3, Platform: PlatformKimi, Type: AccountTypeAPIKey, Status: StatusDisabled,
+		Credentials: map[string]any{"account_mode": "coding"}}
+	zhipuCoding := Account{ID: 4, Platform: PlatformZhipu, Type: AccountTypeAPIKey, Status: StatusActive,
+		Credentials: map[string]any{"account_mode": "coding"}}
+
+	repo := &fakeCNCheckRepo{byPlatform: map[string][]Account{
+		PlatformKimi:  {kimiActive, kimiPaused, kimiInactive},
+		PlatformZhipu: {zhipuCoding},
+	}}
+	prober := &fakeCNQuotaProber{}
+	svc := &CNProviderBalanceCheckService{
+		accountRepo:  repo,
+		quotaService: prober,
+		cfg:          &config.Config{},
+	}
+
+	svc.runOnce()
+
+	require.ElementsMatch(t, []int64{1, 2, 4}, prober.probed)
+}
+
+// runOnceZhipuQuota 在 quotaService 缺失时安全跳过（Start 门控不启动的老部署路径）。
+func TestCNProviderBalanceCheckRunOnceWithoutQuotaService(t *testing.T) {
+	repo := &fakeCNCheckRepo{byPlatform: map[string][]Account{
+		PlatformZhipu: {{ID: 4, Platform: PlatformZhipu, Type: AccountTypeAPIKey, Status: StatusActive,
+			Credentials: map[string]any{"account_mode": "coding"}}},
+	}}
+	svc := &CNProviderBalanceCheckService{accountRepo: repo, cfg: &config.Config{}}
+	require.NotPanics(t, func() { svc.runOnce() })
+}
+
+// 双币种（deepseek CNY+USD）停调判定：任一币种达标即不停调，全部低于阈值才停；
+// 无明细时退回主币种（兼容旧结果）。
+func TestAllCNBalancesBelowThreshold(t *testing.T) {
+	dualLow := &CNProviderBalanceResult{
+		Balance:  1.0,
+		Currency: "CNY",
+		Balances: []CNProviderBalanceEntry{
+			{Currency: "CNY", Balance: 1.0},
+			{Currency: "USD", Balance: 0.5},
+		},
+	}
+	require.True(t, allCNBalancesBelowThreshold(dualLow, 5.0))
+
+	dualMixed := &CNProviderBalanceResult{
+		Balance:  1.0,
+		Currency: "CNY",
+		Balances: []CNProviderBalanceEntry{
+			{Currency: "CNY", Balance: 1.0},
+			{Currency: "USD", Balance: 20.0},
+		},
+	}
+	require.False(t, allCNBalancesBelowThreshold(dualMixed, 5.0))
+
+	// 无明细：按主币种判定（旧行为）。
+	singleLow := &CNProviderBalanceResult{Balance: 1.0, Currency: "CNY"}
+	require.True(t, allCNBalancesBelowThreshold(singleLow, 5.0))
+	singleOK := &CNProviderBalanceResult{Balance: 10.0, Currency: "CNY"}
+	require.False(t, allCNBalancesBelowThreshold(singleOK, 5.0))
+}
diff --git a/backend/internal/service/cn_provider_balance_service.go b/backend/internal/service/cn_provider_balance_service.go
new file mode 100644
index 000000000000..2f306e8b6235
--- /dev/null
+++ b/backend/internal/service/cn_provider_balance_service.go
@@ -0,0 +1,270 @@
+package service
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"log/slog"
+	"net/http"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
+	"github.com/tidwall/gjson"
+	"golang.org/x/sync/singleflight"
+)
+
+// 国产供应商 payg（按量付费）账号余额探测服务。
+//
+// 仅覆盖有公开余额端点的供应商：
+//   - Kimi/Moonshot：GET https://api.moonshot.cn/v1/users/me/balance (Bearer) → data.available_balance
+//   - DeepSeek：     GET https://api.deepseek.com/user/balance (Bearer) → balance_infos[].total_balance + is_available
+//
+// 智谱（zhipu）无公开余额端点（OpenAPI 规格验证），仅靠响应式 429/402（见
+// ratelimit_cn_providers.go）。解析逻辑对齐 cc-switch services/balance.rs::query_deepseek。
+const (
+	cnBalanceUpstreamTimeout = 15 * time.Second
+	cnBalanceMaxBodyBytes    = 256 * 1024
+
+	// Extra 余额快照键后缀（加 provider 前缀）。
+	cnBalanceExtraSuffixBalance   = "balance"
+	cnBalanceExtraSuffixCurrency  = "balance_currency"
+	cnBalanceExtraSuffixAvailable = "balance_available" // deepseek is_available 健康标记
+	cnBalanceExtraSuffixUpdated   = "balance_updated_at"
+	cnBalanceExtraSuffixBalances  = "balances" // 多币种明细（deepseek USD+CNY）
+)
+
+// CNProviderBalanceEntry 是单一币种的余额明细。
+type CNProviderBalanceEntry struct {
+	Currency string  `json:"currency"`
+	Balance  float64 `json:"balance"`
+}
+
+// CNProviderBalanceResult 是余额探测的返回结构（管理端 + UI 消费）。
+type CNProviderBalanceResult struct {
+	Provider string `json:"provider"`
+	Success  bool   `json:"success"`
+	// Balance/Currency 为主币种（balance_infos 首条，兼容单币种消费方）；
+	// 完整明细见 Balances（deepseek 双币种账号含 CNY + USD 两条）。
+	Balance    float64                  `json:"balance"`
+	Currency   string                   `json:"currency,omitempty"`
+	Balances   []CNProviderBalanceEntry `json:"balances,omitempty"`
+	Available  bool                     `json:"available"` // 健康标记（deepseek is_available；kimi 无此概念恒 true）
+	StatusCode int                      `json:"status_code,omitempty"`
+	FetchedAt  int64                    `json:"fetched_at"`
+	Persisted  bool                     `json:"persisted"`
+	Error      string                   `json:"error,omitempty"`
+}
+
+// CNProviderBalanceService 探测 Kimi / DeepSeek payg 账号的账户余额。
+type CNProviderBalanceService struct {
+	accountRepo  AccountRepository
+	proxyRepo    ProxyRepository
+	httpUpstream HTTPUpstream
+	cfg          *config.Config
+	flight       singleflight.Group
+}
+
+// NewCNProviderBalanceService 构造余额探测服务。
+func NewCNProviderBalanceService(
+	accountRepo AccountRepository,
+	proxyRepo ProxyRepository,
+	httpUpstream HTTPUpstream,
+	cfg *config.Config,
+) *CNProviderBalanceService {
+	return &CNProviderBalanceService{
+		accountRepo:  accountRepo,
+		proxyRepo:    proxyRepo,
+		httpUpstream: httpUpstream,
+		cfg:          cfg,
+	}
+}
+
+// QueryBalance 探测指定 payg 账号的余额并落 Extra 快照。
+func (s *CNProviderBalanceService) QueryBalance(ctx context.Context, accountID int64) (*CNProviderBalanceResult, error) {
+	if s == nil || s.accountRepo == nil || s.httpUpstream == nil {
+		return nil, infraerrors.New(http.StatusInternalServerError, "CN_BALANCE_NOT_CONFIGURED", "cn provider balance service is not configured")
+	}
+	key := "cn_balance:" + strconv.FormatInt(accountID, 10)
+	resultCh := s.flight.DoChan(key, func() (any, error) {
+		probeCtx, cancel := context.WithTimeout(context.Background(), cnBalanceUpstreamTimeout+5*time.Second)
+		defer cancel()
+		return s.queryBalance(probeCtx, accountID)
+	})
+	select {
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	case flightResult := <-resultCh:
+		if flightResult.Err != nil {
+			return nil, flightResult.Err
+		}
+		result, ok := flightResult.Val.(*CNProviderBalanceResult)
+		if !ok || result == nil {
+			return nil, infraerrors.New(http.StatusInternalServerError, "CN_BALANCE_PROBE_RESULT_INVALID", "invalid cn provider balance probe result")
+		}
+		cloned := *result
+		return &cloned, nil
+	}
+}
+
+func (s *CNProviderBalanceService) queryBalance(ctx context.Context, accountID int64) (*CNProviderBalanceResult, error) {
+	account, err := s.loadPayGAccount(ctx, accountID)
+	if err != nil {
+		return nil, err
+	}
+	provider := account.Platform
+	if provider != PlatformKimi && provider != PlatformDeepseek {
+		return nil, infraerrors.New(http.StatusBadRequest, "CN_BALANCE_NO_ENDPOINT", "account provider has no balance endpoint")
+	}
+
+	apiKey := strings.TrimSpace(account.GetCNAPIKey())
+	if apiKey == "" {
+		return nil, infraerrors.New(http.StatusBadRequest, "CN_BALANCE_NO_APIKEY", "account api_key is empty")
+	}
+
+	targetURL := cnBalanceURL(account)
+	proxyURL := s.resolveProxyURL(ctx, account)
+	callCtx, cancel := context.WithTimeout(ctx, cnBalanceUpstreamTimeout)
+	defer cancel()
+	req, err := http.NewRequestWithContext(callCtx, http.MethodGet, targetURL, nil)
+	if err != nil {
+		return nil, infraerrors.Newf(http.StatusInternalServerError, "CN_BALANCE_REQUEST_BUILD_FAILED", "build request: %v", err)
+	}
+	req.Header.Set("Authorization", "Bearer "+apiKey)
+	req.Header.Set("Accept", "application/json")
+	account.ApplyHeaderOverrides(req.Header)
+
+	resp, err := s.httpUpstream.Do(req, proxyURL, account.ID, maxInt(account.Concurrency, 1))
+	if err != nil {
+		return nil, infraerrors.Newf(http.StatusBadGateway, "CN_BALANCE_REQUEST_FAILED", "upstream request failed: %v", err)
+	}
+	defer func() { _ = resp.Body.Close() }()
+	bodyBytes, _ := io.ReadAll(io.LimitReader(resp.Body, cnBalanceMaxBodyBytes))
+
+	now := time.Now().UTC()
+	result := &CNProviderBalanceResult{
+		Provider:   provider,
+		FetchedAt:  now.Unix(),
+		StatusCode: resp.StatusCode,
+		Available:  true,
+	}
+
+	if resp.StatusCode == http.StatusUnauthorized || resp.StatusCode == http.StatusForbidden {
+		result.Error = fmt.Sprintf("Authentication failed (HTTP %d)", resp.StatusCode)
+		return result, nil
+	}
+	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		result.Error = fmt.Sprintf("API error (HTTP %d): %s", resp.StatusCode, truncate(strings.TrimSpace(string(bodyBytes)), 240))
+		return result, nil
+	}
+
+	var entries []CNProviderBalanceEntry
+	available := true
+	switch provider {
+	case PlatformKimi:
+		// Moonshot：code==0 成功；data.available_balance（number），单币种 CNY。
+		balance, _ := cnParseF64(gjson.GetBytes(bodyBytes, "data.available_balance").Value())
+		entries = append(entries, CNProviderBalanceEntry{Currency: "CNY", Balance: balance})
+	case PlatformDeepseek:
+		// is_available 缺省视为 true（健康）；显式存在时取其值。
+		if v := gjson.GetBytes(bodyBytes, "is_available"); v.Exists() {
+			available = v.Bool()
+		}
+		// balance_infos 逐条解析：双币种账号同时返回 CNY + USD（数组顺序即
+		// 主次序，首条为主币种）。
+		gjson.GetBytes(bodyBytes, "balance_infos").ForEach(func(_, info gjson.Result) bool {
+			currency := strings.ToUpper(strings.TrimSpace(info.Get("currency").String()))
+			balance, _ := cnParseF64(info.Get("total_balance").Value())
+			if currency == "" {
+				currency = "CNY"
+			}
+			entries = append(entries, CNProviderBalanceEntry{Currency: currency, Balance: balance})
+			return true
+		})
+		if len(entries) == 0 {
+			entries = append(entries, CNProviderBalanceEntry{Currency: "CNY"})
+		}
+	}
+	result.Balances = entries
+	result.Balance = entries[0].Balance
+	result.Currency = entries[0].Currency
+	result.Available = available
+	result.Success = true
+
+	balanceUpdates := make([]any, 0, len(entries))
+	for _, entry := range entries {
+		balanceUpdates = append(balanceUpdates, map[string]any{
+			"currency": entry.Currency,
+			"balance":  entry.Balance,
+		})
+	}
+	updates := map[string]any{
+		cnExtraKey(provider, cnBalanceExtraSuffixBalance):   result.Balance,
+		cnExtraKey(provider, cnBalanceExtraSuffixCurrency):  result.Currency,
+		cnExtraKey(provider, cnBalanceExtraSuffixAvailable): available,
+		cnExtraKey(provider, cnBalanceExtraSuffixUpdated):   now.Format(time.RFC3339),
+		cnExtraKey(provider, cnBalanceExtraSuffixBalances):  balanceUpdates,
+		// 余额探测成功即清除响应式 402/429 写下的 balance_low 标记。
+		cnExtraKey(provider, cnBalanceExtraSuffixLow): false,
+	}
+	if err := s.accountRepo.UpdateExtra(ctx, account.ID, updates); err != nil {
+		slog.Warn("cn_balance_persist_failed", "account_id", account.ID, "provider", provider, "error", err)
+	} else {
+		result.Persisted = true
+	}
+	return result, nil
+}
+
+// loadPayGAccount 加载 payg 模式的国产供应商账号（余额仅对 payg 有意义；coding 走额度）。
+func (s *CNProviderBalanceService) loadPayGAccount(ctx context.Context, accountID int64) (*Account, error) {
+	account, err := s.accountRepo.GetByID(ctx, accountID)
+	if err != nil {
+		return nil, infraerrors.Newf(http.StatusNotFound, "CN_BALANCE_ACCOUNT_NOT_FOUND", "account not found: %v", err)
+	}
+	if account == nil {
+		return nil, infraerrors.New(http.StatusNotFound, "CN_BALANCE_ACCOUNT_NOT_FOUND", "account not found")
+	}
+	if !account.IsCNProvider() {
+		return nil, infraerrors.New(http.StatusBadRequest, "CN_BALANCE_INVALID_PLATFORM", "account is not a CN provider account")
+	}
+	// coding 账号走额度探测，余额端点不适用。
+	if account.IsCodingPlan() {
+		return nil, infraerrors.New(http.StatusBadRequest, "CN_BALANCE_CODING_PLAN", "coding plan account has no balance endpoint; use quota probe")
+	}
+	return account, nil
+}
+
+func (s *CNProviderBalanceService) resolveProxyURL(ctx context.Context, account *Account) string {
+	if account == nil || account.ProxyID == nil {
+		return ""
+	}
+	if account.Proxy != nil {
+		return account.Proxy.URL()
+	}
+	if s != nil && s.proxyRepo != nil {
+		if proxy, err := s.proxyRepo.GetByID(ctx, *account.ProxyID); err == nil && proxy != nil {
+			account.Proxy = proxy
+			return proxy.URL()
+		}
+	}
+	return ""
+}
+
+// cnBalanceURL 解析账号的余额端点。
+//
+//   - Kimi：固定 https://api.moonshot.cn/v1/users/me/balance（与 base_url 无关，Moonshot 仅此一处）
+//   - DeepSeek：基于 base_url 拼接 /user/balance（支持自定义域名）
+func cnBalanceURL(account *Account) string {
+	switch account.Platform {
+	case PlatformKimi:
+		return "https://api.moonshot.cn/v1/users/me/balance"
+	case PlatformDeepseek:
+		// Anthropic 协议账号的凭证 base_url 指向 /anthropic 端点，余额探测需回退
+		// 到 OpenAI 格式 base（协议感知）再拼接 /user/balance。
+		return strings.TrimRight(account.GetOpenAIFormatBaseURL(), "/") + "/user/balance"
+	default:
+		return ""
+	}
+}
diff --git a/backend/internal/service/cn_provider_quota_service.go b/backend/internal/service/cn_provider_quota_service.go
new file mode 100644
index 000000000000..e8fdba9c5609
--- /dev/null
+++ b/backend/internal/service/cn_provider_quota_service.go
@@ -0,0 +1,557 @@
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"log/slog"
+	"net/http"
+	"sort"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
+	"github.com/tidwall/gjson"
+	"golang.org/x/sync/singleflight"
+)
+
+// 国产供应商 Coding Plan 滚动窗口额度探测服务（Kimi For Coding / 智谱 GLM Coding Plan）。
+//
+// 与 grok_quota_service 不同：CN 供应商走数据面 API Key（无 OAuth token provider），
+// 额度端点为只读 GET，解析 5h + weekly 两档滚动窗口并落 account.Extra 快照，
+// 供账号调度阈值评估（account_scheduling_threshold_eval.go）做主动停调。
+//
+// 解析逻辑对齐 cc-switch（farion1231/cc-switch）services/coding_plan.rs 的
+// query_kimi / query_zhipu，包括智谱 unit 字段优先分类与 reset 兜底启发式。
+const (
+	cnQuotaUpstreamTimeout = 15 * time.Second
+	cnQuotaMaxBodyBytes    = 256 * 1024
+
+	// Extra 快照键后缀（加 provider 前缀，如 kimi_5h_used_percent）。
+	cnExtraSuffix5hUsed       = "5h_used_percent"
+	cnExtraSuffix5hReset      = "5h_reset_at"
+	cnExtraSuffixWeeklyUsed   = "weekly_used_percent"
+	cnExtraSuffixWeeklyReset  = "weekly_reset_at"
+	cnExtraSuffixUsageUpdated = "usage_updated_at"
+)
+
+// cnExtraKey 拼接 provider 维度的 extra 键。
+func cnExtraKey(provider, suffix string) string { return provider + "_" + suffix }
+
+// CNQuotaTier 表示一个滚动用量窗口档位（5h / weekly）。
+type CNQuotaTier struct {
+	Window      string  `json:"window"`             // "5h" | "weekly"
+	UsedPercent float64 `json:"used_percent"`       // 已用百分比（0-100+，不做裁剪）
+	ResetAt     string  `json:"reset_at,omitempty"` // RFC3339，空表示无重置时间
+}
+
+// CNProviderQuotaProbeResult 是 Coding Plan 额度探测的返回结构（管理端 + UI 消费）。
+type CNProviderQuotaProbeResult struct {
+	Provider        string        `json:"provider"`
+	Source          string        `json:"source"`
+	Success         bool          `json:"success"`
+	CredentialValid bool          `json:"credential_valid"` // false = 401/403 鉴权失败
+	Tiers           []CNQuotaTier `json:"tiers,omitempty"`
+	PlanLevel       string        `json:"plan_level,omitempty"` // 智谱套餐等级
+	StatusCode      int           `json:"status_code,omitempty"`
+	FetchedAt       int64         `json:"fetched_at"`
+	Persisted       bool          `json:"persisted"`
+	Error           string        `json:"error,omitempty"`
+}
+
+// CNProviderQuotaService 探测 Kimi / Zhipu Coding Plan 的滚动窗口用量。
+type CNProviderQuotaService struct {
+	accountRepo  AccountRepository
+	proxyRepo    ProxyRepository
+	httpUpstream HTTPUpstream
+	cfg          *config.Config
+	flight       singleflight.Group
+}
+
+// NewCNProviderQuotaService 构造 Coding Plan 额度探测服务。
+func NewCNProviderQuotaService(
+	accountRepo AccountRepository,
+	proxyRepo ProxyRepository,
+	httpUpstream HTTPUpstream,
+	cfg *config.Config,
+) *CNProviderQuotaService {
+	return &CNProviderQuotaService{
+		accountRepo:  accountRepo,
+		proxyRepo:    proxyRepo,
+		httpUpstream: httpUpstream,
+		cfg:          cfg,
+	}
+}
+
+// QueryUsage 探测指定账号的 Coding Plan 滚动窗口用量并落 Extra 快照。
+// 同一账号的并发探测会被 singleflight 合并。
+func (s *CNProviderQuotaService) QueryUsage(ctx context.Context, accountID int64) (*CNProviderQuotaProbeResult, error) {
+	if s == nil || s.accountRepo == nil || s.httpUpstream == nil {
+		return nil, infraerrors.New(http.StatusInternalServerError, "CN_QUOTA_NOT_CONFIGURED", "cn provider quota service is not configured")
+	}
+	key := "cn_quota:" + strconv.FormatInt(accountID, 10)
+	resultCh := s.flight.DoChan(key, func() (any, error) {
+		probeCtx, cancel := context.WithTimeout(context.Background(), cnQuotaUpstreamTimeout+5*time.Second)
+		defer cancel()
+		return s.queryUsage(probeCtx, accountID)
+	})
+	select {
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	case flightResult := <-resultCh:
+		if flightResult.Err != nil {
+			return nil, flightResult.Err
+		}
+		result, ok := flightResult.Val.(*CNProviderQuotaProbeResult)
+		if !ok || result == nil {
+			return nil, infraerrors.New(http.StatusInternalServerError, "CN_QUOTA_PROBE_RESULT_INVALID", "invalid cn provider quota probe result")
+		}
+		cloned := *result
+		return &cloned, nil
+	}
+}
+
+func (s *CNProviderQuotaService) queryUsage(ctx context.Context, accountID int64) (*CNProviderQuotaProbeResult, error) {
+	account, err := s.loadCodingPlanAccount(ctx, accountID)
+	if err != nil {
+		return nil, err
+	}
+
+	provider := account.GetCodingPlanProvider()
+	if provider != PlatformKimi && provider != PlatformZhipu {
+		return nil, infraerrors.New(http.StatusBadRequest, "CN_QUOTA_NOT_CODING_PLAN", "account is not a kimi/zhipu coding plan account")
+	}
+
+	apiKey := strings.TrimSpace(account.GetCNAPIKey())
+	if apiKey == "" {
+		return nil, infraerrors.New(http.StatusBadRequest, "CN_QUOTA_NO_APIKEY", "account api_key is empty")
+	}
+
+	baseURL := account.GetOpenAIBaseURL()
+	var (
+		targetURL  string
+		authHeader string
+	)
+	switch provider {
+	case PlatformKimi:
+		targetURL = kimiQuotaURL(baseURL)
+		authHeader = "Bearer " + apiKey
+	case PlatformZhipu:
+		targetURL = zhipuQuotaURL(baseURL)
+		authHeader = apiKey // 智谱额度端点鉴权不加 Bearer 前缀
+	}
+
+	proxyURL := s.resolveProxyURL(ctx, account)
+	callCtx, cancel := context.WithTimeout(ctx, cnQuotaUpstreamTimeout)
+	defer cancel()
+	req, err := http.NewRequestWithContext(callCtx, http.MethodGet, targetURL, nil)
+	if err != nil {
+		return nil, infraerrors.Newf(http.StatusInternalServerError, "CN_QUOTA_REQUEST_BUILD_FAILED", "build request: %v", err)
+	}
+	req.Header.Set("Authorization", authHeader)
+	req.Header.Set("Accept", "application/json")
+	if provider == PlatformZhipu {
+		req.Header.Set("Content-Type", "application/json")
+		req.Header.Set("Accept-Language", "en-US,en")
+	}
+	// 探测与真实转发保持同一套账号级请求头覆写，避免探测通过但转发失败。
+	account.ApplyHeaderOverrides(req.Header)
+
+	resp, err := s.httpUpstream.Do(req, proxyURL, account.ID, maxInt(account.Concurrency, 1))
+	if err != nil {
+		return nil, infraerrors.Newf(http.StatusBadGateway, "CN_QUOTA_REQUEST_FAILED", "upstream request failed: %v", err)
+	}
+	defer func() { _ = resp.Body.Close() }()
+	bodyBytes, _ := io.ReadAll(io.LimitReader(resp.Body, cnQuotaMaxBodyBytes))
+
+	now := time.Now().UTC()
+	result := &CNProviderQuotaProbeResult{
+		Provider:   provider,
+		Source:     "coding_plan",
+		FetchedAt:  now.Unix(),
+		StatusCode: resp.StatusCode,
+	}
+
+	if resp.StatusCode == http.StatusUnauthorized || resp.StatusCode == http.StatusForbidden {
+		// 鉴权失败：不落快照（不覆盖之前的有效值），仅返回失败结果供前端提示。
+		result.Error = fmt.Sprintf("Authentication failed (HTTP %d)", resp.StatusCode)
+		return result, nil
+	}
+	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		result.Error = fmt.Sprintf("API error (HTTP %d): %s", resp.StatusCode, truncate(strings.TrimSpace(string(bodyBytes)), 240))
+		return result, nil
+	}
+
+	// 智谱业务级错误（HTTP 2xx 但 success=false）。
+	if provider == PlatformZhipu {
+		if success := gjson.GetBytes(bodyBytes, "success"); success.Exists() && !success.Bool() {
+			msg := strings.TrimSpace(gjson.GetBytes(bodyBytes, "msg").String())
+			if msg == "" {
+				msg = "unknown zhipu quota error"
+			}
+			result.Error = "API error: " + msg
+			return result, nil
+		}
+	}
+
+	var tiers []CNQuotaTier
+	switch provider {
+	case PlatformKimi:
+		tiers = parseKimiUsageTiers(bodyBytes)
+	case PlatformZhipu:
+		tiers = parseZhipuTokenTiers(gjson.GetBytes(bodyBytes, "data"))
+		result.PlanLevel = strings.TrimSpace(gjson.GetBytes(bodyBytes, "data.level").String())
+	}
+	result.Tiers = tiers
+	result.Success = true
+	result.CredentialValid = true
+
+	updates := cnQuotaExtraUpdates(provider, tiers, now)
+	if err := s.accountRepo.UpdateExtra(ctx, account.ID, updates); err != nil {
+		slog.Warn("cn_quota_persist_failed", "account_id", account.ID, "provider", provider, "error", err)
+	} else {
+		result.Persisted = true
+	}
+	return result, nil
+}
+
+func (s *CNProviderQuotaService) loadCodingPlanAccount(ctx context.Context, accountID int64) (*Account, error) {
+	account, err := s.accountRepo.GetByID(ctx, accountID)
+	if err != nil {
+		return nil, infraerrors.Newf(http.StatusNotFound, "CN_QUOTA_ACCOUNT_NOT_FOUND", "account not found: %v", err)
+	}
+	if account == nil {
+		return nil, infraerrors.New(http.StatusNotFound, "CN_QUOTA_ACCOUNT_NOT_FOUND", "account not found")
+	}
+	if !account.IsCNProvider() {
+		return nil, infraerrors.New(http.StatusBadRequest, "CN_QUOTA_INVALID_PLATFORM", "account is not a CN provider account")
+	}
+	if !account.IsCodingPlan() {
+		return nil, infraerrors.New(http.StatusBadRequest, "CN_QUOTA_NOT_CODING_PLAN", "account is not a coding plan account")
+	}
+	return account, nil
+}
+
+func (s *CNProviderQuotaService) resolveProxyURL(ctx context.Context, account *Account) string {
+	if account == nil || account.ProxyID == nil {
+		return ""
+	}
+	if account.Proxy != nil {
+		return account.Proxy.URL()
+	}
+	if s != nil && s.proxyRepo != nil {
+		if proxy, err := s.proxyRepo.GetByID(ctx, *account.ProxyID); err == nil && proxy != nil {
+			account.Proxy = proxy
+			return proxy.URL()
+		}
+	}
+	return ""
+}
+
+// zhipuQuotaURL 根据 base_url 解析智谱额度端点（与数据面推理域名同主机）。
+func zhipuQuotaURL(baseURL string) string {
+	return zhipuQuotaHost(baseURL) + "/api/monitor/usage/quota/limit"
+}
+
+// kimiQuotaURL 根据 base_url 解析 Kimi For Coding 额度端点。
+// cc-switch query_kimi 固定探测 https://api.kimi.com/coding/v1/usages
+// （实测 /coding/usages 无 /v1 → 404）。coding/v1（CC 协议默认）与
+// coding（Anthropic 协议默认）两种 base 统一剥掉尾部后拼回 /v1/usages，
+// 协议切换不影响额度探测端点。
+func kimiQuotaURL(baseURL string) string {
+	base := strings.TrimSuffix(strings.TrimRight(baseURL, "/"), "/v1")
+	return base + "/v1/usages"
+}
+
+func zhipuQuotaHost(baseURL string) string {
+	switch u := strings.ToLower(baseURL); {
+	case strings.Contains(u, "bigmodel.cn"):
+		return "https://open.bigmodel.cn"
+	case strings.Contains(u, "z.ai"):
+		return "https://api.z.ai"
+	default:
+		// 国产优先：未知域名回落国内站（与前端 zhipu 预设一致）。
+		return "https://open.bigmodel.cn"
+	}
+}
+
+// parseKimiUsageTiers 解析 Kimi For Coding 的 /usages 响应。
+//
+//   - limits[].detail.{limit,remaining,resetTime} → 5h 窗口（取首个 detail）
+//   - usage.{limit,remaining,resetTime} → 周窗口
+//
+// utilization = (limit-remaining)/limit*100。
+func parseKimiUsageTiers(body []byte) []CNQuotaTier {
+	var tiers []CNQuotaTier
+
+	if limits := gjson.GetBytes(body, "limits"); limits.IsArray() {
+		limits.ForEach(func(_, item gjson.Result) bool {
+			detail := item.Get("detail")
+			if !detail.Exists() {
+				return true
+			}
+			limit, _ := cnParseF64(detail.Get("limit").Value())
+			remaining, _ := cnParseF64(detail.Get("remaining").Value())
+			used := limit - remaining
+			if used < 0 {
+				used = 0
+			}
+			var util float64
+			if limit > 0 {
+				util = used / limit * 100
+			}
+			tiers = append(tiers, CNQuotaTier{
+				Window:      "5h",
+				UsedPercent: util,
+				ResetAt:     cnNormalizeResetTime(detail.Get("resetTime").Value()),
+			})
+			return false // 取首个 detail 作为 5h 窗口
+		})
+	}
+
+	if usage := gjson.GetBytes(body, "usage"); usage.Exists() {
+		limit, _ := cnParseF64(usage.Get("limit").Value())
+		remaining, _ := cnParseF64(usage.Get("remaining").Value())
+		used := limit - remaining
+		if used < 0 {
+			used = 0
+		}
+		var util float64
+		if limit > 0 {
+			util = used / limit * 100
+		}
+		tiers = append(tiers, CNQuotaTier{
+			Window:      "weekly",
+			UsedPercent: util,
+			ResetAt:     cnNormalizeResetTime(usage.Get("resetTime").Value()),
+		})
+	}
+
+	return tiers
+}
+
+// cnZhipuWindow 标识智谱 TOKENS_LIMIT 条目所属窗口。
+type cnZhipuWindow int
+
+const (
+	cnZhipuWindowUnknown cnZhipuWindow = iota
+	cnZhipuWindow5h
+	cnZhipuWindowWeekly
+)
+
+// classifyZhipuWindowUnit 按 unit 字段判定窗口类型（3=5h，6=weekly）。
+// unit 缺失或未识别时返回 Unknown，由调用方走 reset 时间启发式兜底。
+func classifyZhipuWindowUnit(unit int64) cnZhipuWindow {
+	switch unit {
+	case 3:
+		return cnZhipuWindow5h
+	case 6:
+		return cnZhipuWindowWeekly
+	default:
+		return cnZhipuWindowUnknown
+	}
+}
+
+// parseZhipuTokenTiers 解析智谱额度响应 data.limits 为 5h + weekly 两档。
+//
+// 分类优先级（对齐 cc-switch parse_zhipu_token_tiers，issue #3036）：
+//  1. 显式 unit 字段（3=5h / 6=weekly）——不能用 reset 排序代替，周期末尾
+//     周窗口会比 5h 更早重置，时间排序必然标反。
+//  2. unit 缺失/未识别：无 nextResetTime 的条目优先归 5h（0% 状态下 5h 桶可能
+//     没有 reset），其余按 reset 升序依次填入仍空缺的槽位。
+//
+// CREDIT_LIMIT（信用额度）与 TOKENS_LIMIT（token 窗口）度量不同：两者同时返回时
+// 只让 TOKENS_LIMIT 参与 5h/weekly 槽位竞争，避免信用额度百分比污染阈值停调
+// 快照；仅当无任何 TOKENS_LIMIT 条目时才降级用 CREDIT_LIMIT 展示。
+// 老套餐只回 1 条 TOKENS_LIMIT，自然降级为仅 5h；新套餐回 2 条。
+func parseZhipuTokenTiers(data gjson.Result) []CNQuotaTier {
+	type entry struct {
+		resetMs    int64
+		hasReset   bool
+		percentage float64
+		resetISO   string
+	}
+	var (
+		fiveHour     entry
+		fiveHourSet  bool
+		weekly       entry
+		weeklySet    bool
+		unclassified []entry
+	)
+
+	classify := func(item gjson.Result, e entry) {
+		switch classifyZhipuWindowUnit(item.Get("unit").Int()) {
+		case cnZhipuWindow5h:
+			if !fiveHourSet {
+				fiveHour, fiveHourSet = e, true
+			} else {
+				unclassified = append(unclassified, e)
+			}
+		case cnZhipuWindowWeekly:
+			if !weeklySet {
+				weekly, weeklySet = e, true
+			} else {
+				unclassified = append(unclassified, e)
+			}
+		default:
+			unclassified = append(unclassified, e)
+		}
+	}
+	var creditFallback []entry
+	hasTokensLimit := false
+
+	data.Get("limits").ForEach(func(_, item gjson.Result) bool {
+		limitType := strings.ToUpper(strings.TrimSpace(item.Get("type").String()))
+		if limitType != "TOKENS_LIMIT" && limitType != "CREDIT_LIMIT" {
+			return true
+		}
+		percentage := 0.0
+		if p, ok := cnParseF64(item.Get("percentage").Value()); ok {
+			percentage = p
+		}
+		var (
+			resetMs  int64
+			hasReset bool
+			resetISO string
+		)
+		if nr := item.Get("nextResetTime"); nr.Exists() {
+			switch nr.Type {
+			case gjson.Number:
+				resetMs = nr.Int()
+				hasReset = resetMs > 0
+				resetISO = cnMillisToRFC3339(resetMs)
+			case gjson.String:
+				resetISO = cnNormalizeResetTime(nr.String())
+				hasReset = resetISO != ""
+			}
+		}
+		e := entry{resetMs: resetMs, hasReset: hasReset, percentage: percentage, resetISO: resetISO}
+		if limitType == "TOKENS_LIMIT" {
+			hasTokensLimit = true
+			classify(item, e)
+		} else {
+			creditFallback = append(creditFallback, e)
+		}
+		return true
+	})
+
+	// 无任何 TOKENS_LIMIT 条目（部分套餐只报信用额度）：降级用 CREDIT_LIMIT 展示。
+	if !hasTokensLimit {
+		unclassified = append(unclassified, creditFallback...)
+	}
+
+	// 无 reset 的条目排前，再按 reset 升序，依次填入仍空缺的槽位。
+	sort.SliceStable(unclassified, func(i, j int) bool {
+		if unclassified[i].hasReset != unclassified[j].hasReset {
+			return !unclassified[i].hasReset
+		}
+		return unclassified[i].resetMs < unclassified[j].resetMs
+	})
+	for _, e := range unclassified {
+		switch {
+		case !fiveHourSet:
+			fiveHour, fiveHourSet = e, true
+		case !weeklySet:
+			weekly, weeklySet = e, true
+		}
+	}
+
+	var tiers []CNQuotaTier
+	if fiveHourSet {
+		tiers = append(tiers, CNQuotaTier{Window: "5h", UsedPercent: fiveHour.percentage, ResetAt: fiveHour.resetISO})
+	}
+	if weeklySet {
+		tiers = append(tiers, CNQuotaTier{Window: "weekly", UsedPercent: weekly.percentage, ResetAt: weekly.resetISO})
+	}
+	return tiers
+}
+
+// cnQuotaExtraUpdates 根据 tier 列表构造 provider 维度的 Extra 快照更新。
+func cnQuotaExtraUpdates(provider string, tiers []CNQuotaTier, now time.Time) map[string]any {
+	updates := map[string]any{
+		cnExtraKey(provider, cnExtraSuffixUsageUpdated): now.Format(time.RFC3339),
+	}
+	for _, t := range tiers {
+		switch t.Window {
+		case "5h":
+			updates[cnExtraKey(provider, cnExtraSuffix5hUsed)] = t.UsedPercent
+			if t.ResetAt != "" {
+				updates[cnExtraKey(provider, cnExtraSuffix5hReset)] = t.ResetAt
+			}
+		case "weekly":
+			updates[cnExtraKey(provider, cnExtraSuffixWeeklyUsed)] = t.UsedPercent
+			if t.ResetAt != "" {
+				updates[cnExtraKey(provider, cnExtraSuffixWeeklyReset)] = t.ResetAt
+			}
+		}
+	}
+	return updates
+}
+
+// cnParseF64 把 JSON 数值或字符串解析为 float64（兼容 "100" 与 100）。
+func cnParseF64(raw any) (float64, bool) {
+	switch v := raw.(type) {
+	case float64:
+		return v, true
+	case float32:
+		return float64(v), true
+	case int:
+		return float64(v), true
+	case int64:
+		return float64(v), true
+	case json.Number:
+		f, err := v.Float64()
+		return f, err == nil
+	case string:
+		f, err := strconv.ParseFloat(strings.TrimSpace(v), 64)
+		return f, err == nil
+	default:
+		return 0, false
+	}
+}
+
+// cnNormalizeResetTime 把上游重置时间（ISO8601 字符串 / 秒级 / 毫秒级数字）归一化为
+// RFC3339 字符串；无法识别或非正时间戳返回空串。
+func cnNormalizeResetTime(raw any) string {
+	switch v := raw.(type) {
+	case string:
+		s := strings.TrimSpace(v)
+		if s == "" {
+			return ""
+		}
+		if ts, err := parseSchedulingTime(s); err == nil {
+			return ts.UTC().Format(time.RFC3339)
+		}
+		return ""
+	case float64:
+		return cnMillisToRFC3339(int64(v))
+	case int:
+		return cnMillisToRFC3339(int64(v))
+	case int64:
+		return cnMillisToRFC3339(v)
+	case json.Number:
+		if n, err := v.Int64(); err == nil {
+			return cnMillisToRFC3339(n)
+		}
+		return ""
+	default:
+		return ""
+	}
+}
+
+// cnMillisToRFC3339 把秒级（<1e12）或毫秒级时间戳转为 RFC3339 字符串；非正返回空串。
+func cnMillisToRFC3339(n int64) string {
+	if n <= 0 {
+		return ""
+	}
+	var ms int64
+	if n < 1_000_000_000_000 {
+		ms = n * 1000
+	} else {
+		ms = n
+	}
+	return time.UnixMilli(ms).UTC().Format(time.RFC3339)
+}
diff --git a/backend/internal/service/cn_providers_test.go b/backend/internal/service/cn_providers_test.go
new file mode 100644
index 000000000000..24b8bd440541
--- /dev/null
+++ b/backend/internal/service/cn_providers_test.go
@@ -0,0 +1,625 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+)
+
+// TestCNExtraKey 验证 provider 维度的 Extra 快照键由前缀 + 后缀拼接。
+func TestCNExtraKey(t *testing.T) {
+	t.Parallel()
+	require.Equal(t, "kimi_5h_used_percent", cnExtraKey(PlatformKimi, cnExtraSuffix5hUsed))
+	require.Equal(t, "zhipu_weekly_reset_at", cnExtraKey(PlatformZhipu, cnExtraSuffixWeeklyReset))
+	require.Equal(t, "deepseek_balance", cnExtraKey(PlatformDeepseek, cnBalanceExtraSuffixBalance))
+}
+
+// TestCNParseF64 兼容 JSON 数值与字符串（cc-switch 与上游字段类型不一致）。
+func TestCNParseF64(t *testing.T) {
+	t.Parallel()
+	cases := []struct {
+		name string
+		raw  any
+		want float64
+		ok   bool
+	}{
+		{"float64", float64(12.5), 12.5, true},
+		{"int", 100, 100, true},
+		{"numeric string", "33.3", 33.3, true},
+		{"trim string", "  7  ", 7, true},
+		{"non-numeric string", "abc", 0, false},
+		{"nil", nil, 0, false},
+	}
+	for _, tc := range cases {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			got, ok := cnParseF64(tc.raw)
+			require.Equal(t, tc.ok, ok)
+			if ok {
+				require.InDelta(t, tc.want, got, 1e-9)
+			}
+		})
+	}
+}
+
+// TestCNMillisToRFC3339 秒级（<1e12）按秒、毫秒级按毫秒处理；非正返回空串。
+func TestCNMillisToRFC3339(t *testing.T) {
+	t.Parallel()
+	// 1700000000 秒 = 1700000000000 毫秒
+	want := time.UnixMilli(1700000000000).UTC().Format(time.RFC3339)
+	require.Equal(t, want, cnMillisToRFC3339(1700000000))      // 秒级
+	require.Equal(t, want, cnMillisToRFC3339(1700000000000))   // 毫秒级
+	require.Equal(t, "", cnMillisToRFC3339(0))                 // 非正
+	require.Equal(t, "", cnMillisToRFC3339(-1))
+}
+
+// TestCnnormalizeResetTime 覆盖 ISO8601 字符串 / 数字（秒、毫秒）/ 非法输入。
+func TestCnnormalizeResetTime(t *testing.T) {
+	t.Parallel()
+	// ISO8601 字符串归一化为 RFC3339（UTC）。
+	require.Equal(t, "2026-08-14T10:00:00Z", cnNormalizeResetTime("2026-08-14T10:00:00Z"))
+	// 毫秒级 float64。
+	require.Equal(t,
+		time.UnixMilli(1700000000000).UTC().Format(time.RFC3339),
+		cnNormalizeResetTime(float64(1700000000000)))
+	// 非法字符串。
+	require.Equal(t, "", cnNormalizeResetTime("not-a-time"))
+	require.Equal(t, "", cnNormalizeResetTime(""))
+}
+
+// TestParseKimiUsageTiers 验证 Kimi For Coding /usages 解析：
+//   - 首个 limits[].detail → 5h 桶，utilization=(limit-remaining)/limit*100
+//   - usage → weekly 桶
+//   - 仅取首个 detail（多个 detail 时不应重复产出 5h）
+func TestParseKimiUsageTiers(t *testing.T) {
+	t.Parallel()
+	body := []byte(`{
+		"limits": [
+			{"name": "5h", "detail": {"limit": 1000, "remaining": 600, "resetTime": "2026-08-14T15:00:00Z"}},
+			{"name": "ignored-second-detail", "detail": {"limit": 999, "remaining": 0, "resetTime": "2026-08-14T20:00:00Z"}}
+		],
+		"usage": {"limit": 10000, "remaining": 4000, "resetTime": "2026-08-18T00:00:00Z"}
+	}`)
+	tiers := parseKimiUsageTiers(body)
+	require.Len(t, tiers, 2)
+	require.Equal(t, "5h", tiers[0].Window)
+	require.InDelta(t, 40.0, tiers[0].UsedPercent, 1e-9) // (1000-600)/1000*100
+	require.Equal(t, "2026-08-14T15:00:00Z", tiers[0].ResetAt)
+	require.Equal(t, "weekly", tiers[1].Window)
+	require.InDelta(t, 60.0, tiers[1].UsedPercent, 1e-9) // (10000-4000)/10000*100
+	require.Equal(t, "2026-08-18T00:00:00Z", tiers[1].ResetAt)
+}
+
+// TestParseKimiUsageTiers_LimitZero 不应除零：limit=0 → utilization=0。
+func TestParseKimiUsageTiers_LimitZero(t *testing.T) {
+	t.Parallel()
+	body := []byte(`{"limits":[{"detail":{"limit":0,"remaining":0,"resetTime":"2026-08-14T15:00:00Z"}}]}`)
+	tiers := parseKimiUsageTiers(body)
+	require.Len(t, tiers, 1)
+	require.InDelta(t, 0.0, tiers[0].UsedPercent, 1e-9)
+}
+
+// TestParseZhipuTokenTiers_UnitClassification 显式 unit（3=5h / 6=weekly）优先分类，
+// 不能被 reset 时间排序覆盖（周期末尾周窗口会更早重置）。
+func TestParseZhipuTokenTiers_UnitClassification(t *testing.T) {
+	t.Parallel()
+	// weekly 的 nextResetTime 早于 5h（模拟周期末尾），但 unit 必须胜出。
+	data := gjson.Parse(`{
+		"limits": [
+			{"type":"TOKENS_LIMIT","unit":6,"percentage":70,"nextResetTime":1700000000000},
+			{"type":"TOKENS_LIMIT","unit":3,"percentage":20,"nextResetTime":1700000099999}
+		]
+	}`)
+	tiers := parseZhipuTokenTiers(data)
+	require.Len(t, tiers, 2)
+	require.Equal(t, "5h", tiers[0].Window)
+	require.InDelta(t, 20.0, tiers[0].UsedPercent, 1e-9)
+	require.Equal(t, "weekly", tiers[1].Window)
+	require.InDelta(t, 70.0, tiers[1].UsedPercent, 1e-9)
+}
+
+// TestParseZhipuTokenTiers_SingleTierOldPlan 老套餐仅回 1 条 → 降级为仅 5h。
+func TestParseZhipuTokenTiers_SingleTierOldPlan(t *testing.T) {
+	t.Parallel()
+	data := gjson.Parse(`{"limits":[{"type":"TOKENS_LIMIT","unit":3,"percentage":15,"nextResetTime":1700000000000}]}`)
+	tiers := parseZhipuTokenTiers(data)
+	require.Len(t, tiers, 1)
+	require.Equal(t, "5h", tiers[0].Window)
+}
+
+// TestParseZhipuTokenTiers_FallbackHeuristic unit 缺失时：无 reset 的条目优先归 5h，
+// 其余按 reset 升序填入剩余槽位。
+func TestParseZhipuTokenTiers_FallbackHeuristic(t *testing.T) {
+	t.Parallel()
+	// 无 unit：A 无 reset、B 有 reset。A 先填 5h，B 填 weekly。
+	data := gjson.Parse(`{
+		"limits": [
+			{"type":"TOKENS_LIMIT","percentage":50,"nextResetTime":1700000000000},
+			{"type":"TOKENS_LIMIT","percentage":10}
+		]
+	}`)
+	tiers := parseZhipuTokenTiers(data)
+	require.Len(t, tiers, 2)
+	require.Equal(t, "5h", tiers[0].Window)
+	require.InDelta(t, 10.0, tiers[0].UsedPercent, 1e-9) // 无 reset 优先 5h
+	require.Equal(t, "weekly", tiers[1].Window)
+	require.InDelta(t, 50.0, tiers[1].UsedPercent, 1e-9)
+}
+
+// TestParseZhipuTokenTiers_IgnoresNonTokenEntries 非 TOKENS_LIMIT/CREDIT_LIMIT 条目跳过。
+func TestParseZhipuTokenTiers_IgnoresNonTokenEntries(t *testing.T) {
+	t.Parallel()
+	data := gjson.Parse(`{"limits":[{"type":"OTHER_LIMIT","unit":3,"percentage":99}]}`)
+	require.Empty(t, parseZhipuTokenTiers(data))
+}
+
+// TestCNQuotaExtraUpdates 验证 tier 列表落 Extra 快照键的 provider 前缀与窗口映射。
+func TestCNQuotaExtraUpdates(t *testing.T) {
+	t.Parallel()
+	now := time.Date(2026, 8, 14, 0, 0, 0, 0, time.UTC)
+	tiers := []CNQuotaTier{
+		{Window: "5h", UsedPercent: 40, ResetAt: "2026-08-14T15:00:00Z"},
+		{Window: "weekly", UsedPercent: 60, ResetAt: "2026-08-18T00:00:00Z"},
+	}
+	updates := cnQuotaExtraUpdates(PlatformKimi, tiers, now)
+	require.Equal(t, 40.0, updates["kimi_5h_used_percent"])
+	require.Equal(t, "2026-08-14T15:00:00Z", updates["kimi_5h_reset_at"])
+	require.Equal(t, 60.0, updates["kimi_weekly_used_percent"])
+	require.Equal(t, "2026-08-18T00:00:00Z", updates["kimi_weekly_reset_at"])
+	require.Equal(t, now.Format(time.RFC3339), updates["kimi_usage_updated_at"])
+}
+
+// TestCNProviderResponseIndicatesInsufficientBalance 覆盖中英文余额不足文案与否定用例。
+func TestCNProviderResponseIndicatesInsufficientBalance(t *testing.T) {
+	t.Parallel()
+	positive := []string{
+		`{"error":{"message":"余额不足"}}`,
+		`{"error":{"message":"Insufficient balance"}}`,
+		`{"code":"insufficient_credit"}`,
+		`"balance is not enough"`,
+		`"no enough balance"`,
+	}
+	for _, body := range positive {
+		require.True(t, cnProviderResponseIndicatesInsufficientBalance([]byte(body)), body)
+	}
+	negative := []string{
+		`{"error":{"message":"rate limit exceeded"}}`,
+		`{"error":{"message":"quota exhausted"}}`,
+		``,
+	}
+	for _, body := range negative {
+		require.False(t, cnProviderResponseIndicatesInsufficientBalance([]byte(body)), body)
+	}
+}
+
+// TestCNBalanceLowReason 验证稳定前缀（供周期检测任务识别并清除）。
+func TestCNBalanceLowReason(t *testing.T) {
+	t.Parallel()
+	require.Equal(t, "cn_balance_low: upstream said x",
+		cnBalanceLowReason("upstream said x"))
+	require.Equal(t, "cn_balance_low: 余额不足，账号临时停调",
+		cnBalanceLowReason("   "))
+	require.True(t, len(cnBalanceLowReason("")) > len(cnBalanceLowReasonPrefix))
+}
+
+// TestZhipuQuotaHost 按域名路由智谱额度端点主机（bigmodel.cn / z.ai / 默认国内站）。
+func TestZhipuQuotaHost(t *testing.T) {
+	t.Parallel()
+	require.Equal(t, "https://open.bigmodel.cn", zhipuQuotaHost("https://open.bigmodel.cn/api/paas/v4"))
+	require.Equal(t, "https://api.z.ai", zhipuQuotaHost("https://api.z.ai/api/paas/v4"))
+	require.Equal(t, "https://open.bigmodel.cn", zhipuQuotaHost("https://custom.example.com")) // 默认国内站
+	require.Equal(t, "https://open.bigmodel.cn/api/monitor/usage/quota/limit", zhipuQuotaURL("https://open.bigmodel.cn/api/paas/v4"))
+}
+
+// TestKimiQuotaURL 两种协议默认 base（coding/v1 与 coding）都归一到 /coding/v1/usages
+// （cc-switch 固定端点；无 /v1 的 /coding/usages 实测 404）。
+func TestKimiQuotaURL(t *testing.T) {
+	t.Parallel()
+	require.Equal(t, "https://api.kimi.com/coding/v1/usages", kimiQuotaURL("https://api.kimi.com/coding/v1"))
+	require.Equal(t, "https://api.kimi.com/coding/v1/usages", kimiQuotaURL("https://api.kimi.com/coding"))
+	require.Equal(t, "https://api.kimi.com/coding/v1/usages", kimiQuotaURL("https://api.kimi.com/coding/"))
+	require.Equal(t, "https://api.kimi.com/coding/v1/usages", kimiQuotaURL("https://api.kimi.com/coding/v1/"))
+}
+
+// TestCNBalanceURL Kimi 固定端点；DeepSeek 基于 base_url 拼接。
+func TestCNBalanceURL(t *testing.T) {
+	t.Parallel()
+	kimi := &Account{Platform: PlatformKimi}
+	require.Equal(t, "https://api.moonshot.cn/v1/users/me/balance", cnBalanceURL(kimi))
+
+	deepseek := &Account{
+		Platform:    PlatformDeepseek,
+		Credentials: map[string]any{"base_url": "https://api.deepseek.com"},
+	}
+	require.Equal(t, "https://api.deepseek.com/user/balance", cnBalanceURL(deepseek))
+}
+
+// TestCNProviderThresholdCandidates 从 Extra 快照读取 5h / weekly 候选。
+func TestCNProviderThresholdCandidates(t *testing.T) {
+	t.Parallel()
+	account := &Account{
+		Platform: PlatformKimi,
+		Extra: map[string]any{
+			"kimi_5h_used_percent":     90.0,
+			"kimi_5h_reset_at":         "2026-08-14T15:00:00Z",
+			"kimi_weekly_used_percent": 50.0,
+			"kimi_weekly_reset_at":     "2026-08-18T00:00:00Z",
+		},
+	}
+	cands := cnProviderThresholdCandidates(account, PlatformKimi)
+	// 仅返回非 nil 候选（两窗口均存在 → 2 条）。
+	var present []*accountSchedulingThresholdCandidate
+	for _, c := range cands {
+		if c != nil {
+			present = append(present, c)
+		}
+	}
+	require.Len(t, present, 2)
+
+	// 缺少 used 键的窗口不产生候选。
+	partial := &Account{
+		Platform: PlatformKimi,
+		Extra:    map[string]any{"kimi_5h_reset_at": "2026-08-14T15:00:00Z"}, // 无 used_percent
+	}
+	require.Empty(t, filterNil(cnProviderThresholdCandidates(partial, PlatformKimi)))
+
+	// 空 Extra / nil account 安全返回。
+	require.Empty(t, filterNil(cnProviderThresholdCandidates(&Account{Platform: PlatformKimi}, PlatformKimi)))
+}
+
+func filterNil(cands []*accountSchedulingThresholdCandidate) []*accountSchedulingThresholdCandidate {
+	var out []*accountSchedulingThresholdCandidate
+	for _, c := range cands {
+		if c != nil {
+			out = append(out, c)
+		}
+	}
+	return out
+}
+
+// TestEvaluateAccountSchedulingThreshold_KimiCodingPlan 集成验证：kimi coding 账号
+// 5h 用量超阈值且窗口未重置 → 主动停调至 5h 重置点。
+func TestEvaluateAccountSchedulingThreshold_KimiCodingPlan(t *testing.T) {
+	t.Parallel()
+	now := time.Date(2026, 8, 14, 12, 0, 0, 0, time.UTC)
+	reset := now.Add(3 * time.Hour)
+	account := &Account{
+		Platform: PlatformKimi,
+		Extra: map[string]any{
+			"kimi_5h_used_percent": 90.0,
+			"kimi_5h_reset_at":     reset.Format(time.RFC3339),
+			"kimi_weekly_used_percent": 30.0,
+			"kimi_weekly_reset_at":     now.Add(7 * 24 * time.Hour).Format(time.RFC3339),
+		},
+	}
+	decision := EvaluateAccountSchedulingThreshold(account, map[string]int{PlatformKimi: 80}, now)
+	require.True(t, decision.ShouldPause)
+	require.Equal(t, PlatformKimi, decision.Platform)
+	require.Equal(t, "5h", decision.Window)
+	require.InDelta(t, 90.0, decision.UsedPercent, 1e-9)
+	require.NotNil(t, decision.Until)
+	require.True(t, reset.Equal(*decision.Until))
+}
+
+// TestEvaluateAccountSchedulingThreshold_CNWindowResetSkipped 窗口已重置（reset<=now）
+// 或用量低于阈值 → 不停调（candidateMatchesThreshold 要求 until.After(now)）。
+func TestEvaluateAccountSchedulingThreshold_CNWindowResetSkipped(t *testing.T) {
+	t.Parallel()
+	now := time.Date(2026, 8, 14, 12, 0, 0, 0, time.UTC)
+	// 重置时间已过。
+	expired := &Account{
+		Platform: PlatformZhipu,
+		Extra: map[string]any{
+			"zhipu_5h_used_percent": 99.0,
+			"zhipu_5h_reset_at":     now.Add(-1 * time.Hour).Format(time.RFC3339),
+		},
+	}
+	require.False(t, EvaluateAccountSchedulingThreshold(expired, map[string]int{PlatformZhipu: 80}, now).ShouldPause)
+
+	// 用量低于阈值。
+	low := &Account{
+		Platform: PlatformZhipu,
+		Extra: map[string]any{
+			"zhipu_5h_used_percent": 20.0,
+			"zhipu_5h_reset_at":     now.Add(3 * time.Hour).Format(time.RFC3339),
+		},
+	}
+	require.False(t, EvaluateAccountSchedulingThreshold(low, map[string]int{PlatformZhipu: 80}, now).ShouldPause)
+}
+
+// TestCNProviderQuotaSnapshotReset Coding Plan 429 冷却：取快照中最早的「仍在未来」窗口重置点。
+func TestCNProviderQuotaSnapshotReset(t *testing.T) {
+	t.Parallel()
+	now := time.Date(2026, 8, 14, 12, 0, 0, 0, time.UTC)
+	future5h := now.Add(2 * time.Hour)
+	futureWeekly := now.Add(3 * 24 * time.Hour)
+	pastWeekly := now.Add(-24 * time.Hour)
+
+	// 5h 在未来、weekly 已过期 → 返回 5h。
+	account := &Account{
+		Platform:    PlatformKimi,
+		Credentials: map[string]any{"account_mode": AccountModeCoding},
+		Extra: map[string]any{
+			"kimi_5h_reset_at":     future5h.Format(time.RFC3339),
+			"kimi_weekly_reset_at": pastWeekly.Format(time.RFC3339),
+		},
+	}
+	got := cnProviderQuotaSnapshotReset(account, now)
+	require.NotNil(t, got)
+	require.True(t, future5h.Equal(*got))
+
+	// 两窗口均在未来 → 取较早者（429 多由 5h 窗口触发，避免冷却到 weekly 重置）。
+	both := &Account{
+		Platform:    PlatformKimi,
+		Credentials: map[string]any{"account_mode": AccountModeCoding},
+		Extra: map[string]any{
+			"kimi_5h_reset_at":     future5h.Format(time.RFC3339),
+			"kimi_weekly_reset_at": futureWeekly.Format(time.RFC3339),
+		},
+	}
+	gotBoth := cnProviderQuotaSnapshotReset(both, now)
+	require.NotNil(t, gotBoth)
+	require.True(t, future5h.Equal(*gotBoth))
+
+	// 两窗口均过期 → nil。
+	expired := &Account{
+		Platform:    PlatformKimi,
+		Credentials: map[string]any{"account_mode": AccountModeCoding},
+		Extra: map[string]any{
+			"kimi_5h_reset_at":     pastWeekly.Format(time.RFC3339),
+			"kimi_weekly_reset_at": pastWeekly.Format(time.RFC3339),
+		},
+	}
+	require.Nil(t, cnProviderQuotaSnapshotReset(expired, now))
+
+	// payg 账号（非 coding）→ nil（余额型走余额检测）。
+	payg := &Account{
+		Platform:    PlatformKimi,
+		Credentials: map[string]any{"account_mode": AccountModePayG},
+		Extra:       map[string]any{"kimi_5h_reset_at": future5h.Format(time.RFC3339)},
+	}
+	require.Nil(t, cnProviderQuotaSnapshotReset(payg, now))
+}
+
+// TestNormalizeOpenAICompatiblePlatform_SchedulerExactMatch 回归保护：
+// grok 与国产供应商原样保留，其余归一为 openai —— 保证 kimi/zhipu/deepseek 分组请求
+// 精确匹配同名账号（与 openai/grok 当前行为一致），不会错误并入 openai 池。
+func TestNormalizeOpenAICompatiblePlatform_SchedulerExactMatch(t *testing.T) {
+	t.Parallel()
+	require.Equal(t, PlatformGrok, NormalizeOpenAICompatiblePlatform(PlatformGrok))
+	require.Equal(t, PlatformKimi, NormalizeOpenAICompatiblePlatform(PlatformKimi))
+	require.Equal(t, PlatformZhipu, NormalizeOpenAICompatiblePlatform(PlatformZhipu))
+	require.Equal(t, PlatformDeepseek, NormalizeOpenAICompatiblePlatform(PlatformDeepseek))
+	// 其他平台（含空、anthropic、未知）一律归一为 openai。
+	require.Equal(t, PlatformOpenAI, NormalizeOpenAICompatiblePlatform(""))
+	require.Equal(t, PlatformOpenAI, NormalizeOpenAICompatiblePlatform(PlatformAnthropic))
+	require.Equal(t, PlatformOpenAI, NormalizeOpenAICompatiblePlatform("something-else"))
+}
+
+// TestGetOpenAIProtocolAPIKey_CNProviders 验证 OpenAI 协议族密钥读取覆盖国产供应商，
+// 同时保持 IsOpenAIApiKey 的 openai-only 语义（调度倍率/WS 门控不受影响）。
+func TestGetOpenAIProtocolAPIKey_CNProviders(t *testing.T) {
+	t.Parallel()
+
+	kimi := &Account{
+		Platform:     PlatformKimi,
+		Type:         AccountTypeAPIKey,
+		Credentials:  map[string]any{"api_key": "sk-kimi"},
+	}
+	require.Equal(t, "sk-kimi", kimi.GetOpenAIProtocolAPIKey())
+	require.False(t, kimi.IsOpenAIApiKey(), "IsOpenAIApiKey stays openai-only for scheduling gates")
+
+	// 非 APIKey 类型的 CN 账号不返回密钥
+	notAPIKey := &Account{
+		Platform:    PlatformDeepseek,
+		Type:        AccountTypeOAuth,
+		Credentials: map[string]any{"api_key": "sk-leak"},
+	}
+	require.Equal(t, "", notAPIKey.GetOpenAIProtocolAPIKey())
+
+	// openai 原生账号行为不变
+	openai := &Account{
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeAPIKey,
+		Credentials: map[string]any{"api_key": "sk-openai"},
+	}
+	require.Equal(t, "sk-openai", openai.GetOpenAIProtocolAPIKey())
+}
+
+// TestBuildUpstreamModelsRequest_CNProviders 验证“同步上游支持的模型”对国产供应商可用：
+// 密钥经 GetOpenAIProtocolAPIKey 读取，/models 端点拼接到账号 base_url（含默认值）。
+func TestBuildUpstreamModelsRequest_CNProviders(t *testing.T) {
+	t.Parallel()
+
+	svc := &AccountTestService{cfg: &config.Config{}}
+	cases := []struct {
+		name     string
+		platform string
+		mode     string
+		wantURL  string
+	}{
+		{"kimi default", PlatformKimi, "", "https://api.moonshot.cn/v1/models"},
+		{"kimi coding", PlatformKimi, AccountModeCoding, "https://api.kimi.com/coding/v1/models"},
+		{"zhipu default", PlatformZhipu, "", "https://open.bigmodel.cn/api/paas/v4/models"},
+		{"zhipu coding", PlatformZhipu, AccountModeCoding, "https://open.bigmodel.cn/api/coding/paas/v4/models"},
+		{"deepseek", PlatformDeepseek, "", "https://api.deepseek.com/v1/models"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			creds := map[string]any{"api_key": "sk-test"}
+			if tc.mode != "" {
+				creds["account_mode"] = tc.mode
+			}
+			account := &Account{ID: 1, Platform: tc.platform, Type: AccountTypeAPIKey, Credentials: creds}
+			req, err := svc.buildUpstreamModelsRequest(context.Background(), account)
+			require.NoError(t, err)
+			require.Equal(t, tc.wantURL, req.URL.String())
+			require.Equal(t, "Bearer sk-test", req.Header.Get("Authorization"))
+		})
+	}
+}
+
+// TestGetAPIProtocol 验证协议凭证维度的平台校验矩阵：
+// responses 仅 deepseek；缺失/非法值回退 chat_completions（与旧行为一致）。
+func TestGetAPIProtocol(t *testing.T) {
+	t.Parallel()
+
+	mk := func(platform, protocol string) *Account {
+		creds := map[string]any{"api_key": "sk-test"}
+		if protocol != "" {
+			creds["api_protocol"] = protocol
+		}
+		return &Account{Platform: platform, Type: AccountTypeAPIKey, Credentials: creds}
+	}
+
+	require.Equal(t, APIProtocolChatCompletions, mk(PlatformKimi, "").GetAPIProtocol(), "缺失回退默认")
+	require.Equal(t, APIProtocolAnthropic, mk(PlatformZhipu, APIProtocolAnthropic).GetAPIProtocol())
+	require.Equal(t, APIProtocolAnthropic, mk(PlatformKimi, APIProtocolAnthropic).GetAPIProtocol())
+	require.Equal(t, APIProtocolAnthropic, mk(PlatformDeepseek, APIProtocolAnthropic).GetAPIProtocol())
+	require.Equal(t, APIProtocolResponses, mk(PlatformDeepseek, APIProtocolResponses).GetAPIProtocol())
+	require.Equal(t, APIProtocolChatCompletions, mk(PlatformKimi, APIProtocolResponses).GetAPIProtocol(), "kimi 无 responses 端点")
+	require.Equal(t, APIProtocolChatCompletions, mk(PlatformZhipu, APIProtocolResponses).GetAPIProtocol(), "zhipu 无 responses 端点")
+	require.Equal(t, APIProtocolChatCompletions, mk(PlatformKimi, "bogus").GetAPIProtocol(), "非法值回退默认")
+	require.Equal(t, APIProtocolChatCompletions, (&Account{Platform: PlatformOpenAI, Type: AccountTypeAPIKey}).GetAPIProtocol(), "非 CN 供应商恒为默认")
+}
+
+// TestAnthropicProtocolBaseURL 验证 Anthropic 协议默认端点与协议感知的
+// OpenAI 格式 base 回退。
+func TestAnthropicProtocolBaseURL(t *testing.T) {
+	t.Parallel()
+
+	// 默认端点（按供应商 × 模式）
+	require.Equal(t, "https://api.moonshot.cn/anthropic", (&Account{
+		Platform: PlatformKimi, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"api_protocol": APIProtocolAnthropic},
+	}).GetAnthropicProtocolBaseURL())
+	require.Equal(t, "https://api.kimi.com/coding", (&Account{
+		Platform: PlatformKimi, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"api_protocol": APIProtocolAnthropic, "account_mode": AccountModeCoding},
+	}).GetAnthropicProtocolBaseURL())
+	require.Equal(t, "https://open.bigmodel.cn/api/anthropic", (&Account{
+		Platform: PlatformZhipu, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"api_protocol": APIProtocolAnthropic},
+	}).GetAnthropicProtocolBaseURL())
+	require.Equal(t, "https://api.deepseek.com/anthropic", (&Account{
+		Platform: PlatformDeepseek, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"api_protocol": APIProtocolAnthropic},
+	}).GetAnthropicProtocolBaseURL())
+
+	// 凭证 base_url 覆盖默认值
+	require.Equal(t, "https://custom.example.com/anthropic", (&Account{
+		Platform: PlatformZhipu, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"api_protocol": APIProtocolAnthropic, "base_url": "https://custom.example.com/anthropic"},
+	}).GetAnthropicProtocolBaseURL())
+
+	// 非 Anthropic 协议返回空串
+	require.Empty(t, (&Account{
+		Platform: PlatformZhipu, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"base_url": "https://open.bigmodel.cn/api/paas/v4"},
+	}).GetAnthropicProtocolBaseURL())
+}
+
+// TestGetOpenAIFormatBaseURL_ProtocolAware anthropic 协议账号的凭证 base_url
+// 指向 Anthropic 端点，OpenAI 格式路径（模型同步等）必须回退到 CC 默认 base。
+func TestGetOpenAIFormatBaseURL_ProtocolAware(t *testing.T) {
+	t.Parallel()
+
+	zhipuAnthropic := &Account{
+		Platform: PlatformZhipu, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{
+			"api_protocol": APIProtocolAnthropic,
+			"base_url":     "https://open.bigmodel.cn/api/anthropic",
+		},
+	}
+	require.Equal(t, "https://open.bigmodel.cn/api/paas/v4", zhipuAnthropic.GetOpenAIFormatBaseURL())
+
+	kimiCodingAnthropic := &Account{
+		Platform: PlatformKimi, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{
+			"api_protocol": APIProtocolAnthropic,
+			"account_mode": AccountModeCoding,
+			"base_url":     "https://api.kimi.com/coding",
+		},
+	}
+	require.Equal(t, "https://api.kimi.com/coding/v1", kimiCodingAnthropic.GetOpenAIFormatBaseURL())
+
+	// chat_completions 协议下行为不变（凭证 base_url 原样返回）
+	ccAccount := &Account{
+		Platform: PlatformDeepseek, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"base_url": "https://ds-relay.example.com"},
+	}
+	require.Equal(t, "https://ds-relay.example.com", ccAccount.GetOpenAIFormatBaseURL())
+}
+
+// TestBuildUpstreamModelsRequest_AnthropicProtocol 模型同步使用协议感知 base。
+func TestBuildUpstreamModelsRequest_AnthropicProtocol(t *testing.T) {
+	t.Parallel()
+	svc := &AccountTestService{cfg: &config.Config{}}
+	account := &Account{
+		ID: 1, Platform: PlatformZhipu, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{
+			"api_key":      "sk-test",
+			"api_protocol": APIProtocolAnthropic,
+			"base_url":     "https://open.bigmodel.cn/api/anthropic",
+		},
+	}
+	req, err := svc.buildUpstreamModelsRequest(context.Background(), account)
+	require.NoError(t, err)
+	require.Equal(t, "https://open.bigmodel.cn/api/paas/v4/models", req.URL.String())
+}
+
+// TestBuildOpenAIResponsesURLForPlatform deepseek 官方端点为 /responses（无 /v1）。
+func TestBuildOpenAIResponsesURLForPlatform(t *testing.T) {
+	t.Parallel()
+	require.Equal(t, "https://api.deepseek.com/responses", buildOpenAIResponsesURLForPlatform(PlatformDeepseek, "https://api.deepseek.com"))
+	require.Equal(t, "https://api.openai.com/v1/responses", buildOpenAIResponsesURLForPlatform(PlatformOpenAI, "https://api.openai.com"))
+	require.Equal(t, "https://open.bigmodel.cn/api/paas/v4/responses", buildOpenAIResponsesURLForPlatform(PlatformZhipu, "https://open.bigmodel.cn/api/paas/v4"))
+}
+
+// TestNormalizeDeepSeekResponsesRequestBody 无状态适配：强制 store=false、
+// 清除 previous_response_id；非 deepseek responses 协议原样返回。
+func TestNormalizeDeepSeekResponsesRequestBody(t *testing.T) {
+	t.Parallel()
+
+	deepseekResponses := &Account{
+		Platform: PlatformDeepseek, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"api_protocol": APIProtocolResponses},
+	}
+	body := []byte(`{"model":"deepseek-v4-pro","store":true,"previous_response_id":"resp_123","input":"hi"}`)
+	normalized := normalizeDeepSeekResponsesRequestBody(deepseekResponses, body)
+	require.False(t, gjson.GetBytes(normalized, "store").Bool())
+	require.False(t, gjson.GetBytes(normalized, "previous_response_id").Exists())
+	require.Equal(t, "deepseek-v4-pro", gjson.GetBytes(normalized, "model").String())
+
+	// 非 responses 协议（deepseek CC 账号）原样返回
+	deepseekCC := &Account{Platform: PlatformDeepseek, Type: AccountTypeAPIKey}
+	require.Equal(t, string(body), string(normalizeDeepSeekResponsesRequestBody(deepseekCC, body)))
+
+	// openai 账号原样返回
+	openai := &Account{Platform: PlatformOpenAI, Type: AccountTypeAPIKey}
+	require.Equal(t, string(body), string(normalizeDeepSeekResponsesRequestBody(openai, body)))
+}
+
+// TestGetAnthropicAPIKeyAuthScheme_CNProvider CN 账号可经 extra 覆写鉴权方案，
+// 默认保持 x-api-key。
+func TestGetAnthropicAPIKeyAuthScheme_CNProvider(t *testing.T) {
+	t.Parallel()
+
+	zhipu := &Account{
+		Platform: PlatformZhipu, Type: AccountTypeAPIKey,
+		Credentials: map[string]any{"api_protocol": APIProtocolAnthropic},
+	}
+	require.Equal(t, AnthropicAPIKeyAuthSchemeXAPIKey, zhipu.GetAnthropicAPIKeyAuthScheme())
+
+	zhipu.Extra = map[string]any{"anthropic_apikey_auth_scheme": "authorization_bearer"}
+	require.Equal(t, AnthropicAPIKeyAuthSchemeAuthorizationBearer, zhipu.GetAnthropicAPIKeyAuthScheme())
+}
diff --git a/backend/internal/service/domain_constants.go b/backend/internal/service/domain_constants.go
index 2d5311bb2442..c0efa2b527b9 100644
--- a/backend/internal/service/domain_constants.go
+++ b/backend/internal/service/domain_constants.go
@@ -43,12 +43,58 @@ const (
 	PlatformGemini      = domain.PlatformGemini
 	PlatformAntigravity = domain.PlatformAntigravity
 	PlatformGrok        = domain.PlatformGrok
-	PlatformComposite   = domain.PlatformComposite
+	// 国产 OpenAI 兼容供应商（与 grok 一样经 OpenAI 网关转发）。
+	PlatformKimi      = domain.PlatformKimi
+	PlatformZhipu     = domain.PlatformZhipu
+	PlatformDeepseek  = domain.PlatformDeepseek
+	PlatformComposite = domain.PlatformComposite
 	// PlatformKiro is retained for unsupported-platform threshold tests and legacy
 	// account rows. Scheduling-threshold evaluation never pauses kiro accounts.
 	PlatformKiro = "kiro"
 )
 
+// 账号接入模式（国产供应商）：按量付费 vs Coding Plan。
+const (
+	AccountModePayG   = domain.AccountModePayG
+	AccountModeCoding = domain.AccountModeCoding
+)
+
+// 上游 API 协议（国产供应商）：决定转发端点与格式，与接入模式正交。
+const (
+	APIProtocolChatCompletions = domain.APIProtocolChatCompletions
+	APIProtocolAnthropic       = domain.APIProtocolAnthropic
+	APIProtocolResponses       = domain.APIProtocolResponses
+)
+
+// 国产 OpenAI 兼容供应商各模式的默认 base_url。
+// 与前端 credentialsBuilder.ts 中的预设保持一致。
+const (
+	DefaultKimiPayGBaseURL    = "https://api.moonshot.cn/v1"
+	DefaultKimiCodingBaseURL  = "https://api.kimi.com/coding/v1"
+	DefaultZhipuPayGBaseURL   = "https://open.bigmodel.cn/api/paas/v4"
+	DefaultZhipuCodingBaseURL = "https://open.bigmodel.cn/api/coding/paas/v4"
+	DefaultDeepseekBaseURL    = "https://api.deepseek.com"
+)
+
+// 国产供应商 Anthropic 协议端点的默认 base_url（上游路径为 {base}/v1/messages）。
+// 与前端 credentialsBuilder.ts 中的预设保持一致。
+const (
+	DefaultKimiPayGAnthropicBaseURL   = "https://api.moonshot.cn/anthropic"
+	DefaultKimiCodingAnthropicBaseURL = "https://api.kimi.com/coding"
+	DefaultZhipuAnthropicBaseURL      = "https://open.bigmodel.cn/api/anthropic"
+	DefaultDeepseekAnthropicBaseURL   = "https://api.deepseek.com/anthropic"
+)
+
+// IsCNProvider 报告 platform 是否为国产 OpenAI 兼容供应商（kimi/zhipu/deepseek）。
+func IsCNProvider(platform string) bool {
+	switch platform {
+	case PlatformKimi, PlatformZhipu, PlatformDeepseek:
+		return true
+	default:
+		return false
+	}
+}
+
 // AllowedQuotaPlatforms 是允许设置 user × platform quota 的平台列表（单一权威来源）。
 // ent/schema/user_platform_quota.go 的 Validate 函数独立维护（构建期约束），
 // 若新增平台需同步修改该 schema。
@@ -58,14 +104,20 @@ var AllowedQuotaPlatforms = []string{
 	PlatformGemini,
 	PlatformAntigravity,
 	PlatformGrok,
+	PlatformKimi,
+	PlatformZhipu,
+	PlatformDeepseek,
 }
 
 // AllowedSchedulingThresholdPlatforms 是允许设置账号自动停调阈值的平台列表。
-// 仅 openai / anthropic / grok 有原生用量窗口可供评估；其他平台写入阈值无效果。
+// openai/anthropic/grok 有原生用量窗口；kimi/zhipu 的 Coding Plan 同样暴露 5h/weekly
+// 滚动窗口，纳入阈值评估。deepseek 为余额型，走余额检测而非阈值。
 var AllowedSchedulingThresholdPlatforms = []string{
 	PlatformOpenAI,
 	PlatformAnthropic,
 	PlatformGrok,
+	PlatformKimi,
+	PlatformZhipu,
 }
 
 // IsAllowedQuotaPlatform 报告 s 是否为合法的 quota platform 标识。
diff --git a/backend/internal/service/gateway_anthropic_apikey_passthrough_benchmark_test.go b/backend/internal/service/gateway_anthropic_apikey_passthrough_benchmark_test.go
index 37fd709f84cb..a7b05f175c90 100644
--- a/backend/internal/service/gateway_anthropic_apikey_passthrough_benchmark_test.go
+++ b/backend/internal/service/gateway_anthropic_apikey_passthrough_benchmark_test.go
@@ -14,13 +14,12 @@ func BenchmarkGatewayService_ParseSSEUsage_MessageStart(b *testing.B) {
 }
 
 func BenchmarkGatewayService_ParseSSEUsagePassthrough_MessageStart(b *testing.B) {
-	svc := &GatewayService{}
 	data := `{"type":"message_start","message":{"usage":{"input_tokens":123,"cache_creation_input_tokens":45,"cache_read_input_tokens":6,"cached_tokens":6,"cache_creation":{"ephemeral_5m_input_tokens":20,"ephemeral_1h_input_tokens":25}}}}`
 	b.ReportAllocs()
 	b.ResetTimer()
 	for i := 0; i < b.N; i++ {
 		usage := &ClaudeUsage{}
-		svc.parseSSEUsagePassthrough(data, usage)
+		parseSSEUsagePassthrough(data, usage)
 	}
 }
 
@@ -36,13 +35,12 @@ func BenchmarkGatewayService_ParseSSEUsage_MessageDelta(b *testing.B) {
 }
 
 func BenchmarkGatewayService_ParseSSEUsagePassthrough_MessageDelta(b *testing.B) {
-	svc := &GatewayService{}
 	data := `{"type":"message_delta","usage":{"output_tokens":456,"cache_creation_input_tokens":30,"cache_read_input_tokens":7,"cached_tokens":7,"cache_creation":{"ephemeral_5m_input_tokens":10,"ephemeral_1h_input_tokens":20}}}`
 	b.ReportAllocs()
 	b.ResetTimer()
 	for i := 0; i < b.N; i++ {
 		usage := &ClaudeUsage{}
-		svc.parseSSEUsagePassthrough(data, usage)
+		parseSSEUsagePassthrough(data, usage)
 	}
 }
 
diff --git a/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go b/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
index 4fa52cc9ce80..eba08a113be1 100644
--- a/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
+++ b/backend/internal/service/gateway_anthropic_apikey_passthrough_test.go
@@ -1256,11 +1256,10 @@ func TestExtractAnthropicSSEDataLine(t *testing.T) {
 }
 
 func TestGatewayService_ParseSSEUsagePassthrough_MessageStartFallbacks(t *testing.T) {
-	svc := &GatewayService{}
 	usage := &ClaudeUsage{}
 	data := `{"type":"message_start","message":{"usage":{"input_tokens":12,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"cached_tokens":9,"cache_creation":{"ephemeral_5m_input_tokens":3,"ephemeral_1h_input_tokens":4}}}}`
 
-	svc.parseSSEUsagePassthrough(data, usage)
+	parseSSEUsagePassthrough(data, usage)
 
 	require.Equal(t, 12, usage.InputTokens)
 	require.Equal(t, 9, usage.CacheReadInputTokens, "应兼容 cached_tokens 字段")
@@ -1270,7 +1269,6 @@ func TestGatewayService_ParseSSEUsagePassthrough_MessageStartFallbacks(t *testin
 }
 
 func TestGatewayService_ParseSSEUsagePassthrough_MessageDeltaSelectiveOverwrite(t *testing.T) {
-	svc := &GatewayService{}
 	usage := &ClaudeUsage{
 		InputTokens:           10,
 		CacheCreation5mTokens: 2,
@@ -1278,7 +1276,7 @@ func TestGatewayService_ParseSSEUsagePassthrough_MessageDeltaSelectiveOverwrite(
 	}
 	data := `{"type":"message_delta","usage":{"input_tokens":0,"output_tokens":5,"cache_creation_input_tokens":8,"cache_read_input_tokens":0,"cached_tokens":11,"cache_creation":{"ephemeral_5m_input_tokens":1,"ephemeral_1h_input_tokens":0}}}`
 
-	svc.parseSSEUsagePassthrough(data, usage)
+	parseSSEUsagePassthrough(data, usage)
 
 	require.Equal(t, 10, usage.InputTokens, "message_delta 中 0 值不应覆盖已有 input_tokens")
 	require.Equal(t, 5, usage.OutputTokens)
@@ -1289,28 +1287,26 @@ func TestGatewayService_ParseSSEUsagePassthrough_MessageDeltaSelectiveOverwrite(
 }
 
 func TestGatewayService_ParseSSEUsagePassthrough_NoopCases(t *testing.T) {
-	svc := &GatewayService{}
 
 	usage := &ClaudeUsage{InputTokens: 3}
-	svc.parseSSEUsagePassthrough("", usage)
+	parseSSEUsagePassthrough("", usage)
 	require.Equal(t, 3, usage.InputTokens)
 
-	svc.parseSSEUsagePassthrough("[DONE]", usage)
+	parseSSEUsagePassthrough("[DONE]", usage)
 	require.Equal(t, 3, usage.InputTokens)
 
-	svc.parseSSEUsagePassthrough("not-json", usage)
+	parseSSEUsagePassthrough("not-json", usage)
 	require.Equal(t, 3, usage.InputTokens)
 
 	// nil usage 不应 panic
-	svc.parseSSEUsagePassthrough(`{"type":"message_start"}`, nil)
+	parseSSEUsagePassthrough(`{"type":"message_start"}`, nil)
 }
 
 func TestGatewayService_ParseSSEUsagePassthrough_FallbackFromUsageNode(t *testing.T) {
-	svc := &GatewayService{}
 	usage := &ClaudeUsage{}
 	data := `{"type":"content_block_delta","usage":{"cached_tokens":6,"cache_creation":{"ephemeral_5m_input_tokens":2,"ephemeral_1h_input_tokens":1}}}`
 
-	svc.parseSSEUsagePassthrough(data, usage)
+	parseSSEUsagePassthrough(data, usage)
 
 	require.Equal(t, 6, usage.CacheReadInputTokens)
 	require.Equal(t, 3, usage.CacheCreationInputTokens)
diff --git a/backend/internal/service/gateway_anthropic_passthrough.go b/backend/internal/service/gateway_anthropic_passthrough.go
index c18d5bf047cd..def4d3464873 100644
--- a/backend/internal/service/gateway_anthropic_passthrough.go
+++ b/backend/internal/service/gateway_anthropic_passthrough.go
@@ -546,7 +546,7 @@ func (s *GatewayService) handleStreamingResponseAnthropicAPIKeyPassthrough(
 					ms := int(time.Since(startTime).Milliseconds())
 					firstTokenMs = &ms
 				}
-				s.parseSSEUsagePassthrough(data, usage)
+				parseSSEUsagePassthrough(data, usage)
 			} else {
 				trimmed := strings.TrimSpace(line)
 				if strings.HasPrefix(trimmed, "event:") && anthropicStreamEventIsTerminal(strings.TrimSpace(strings.TrimPrefix(trimmed, "event:")), "") {
@@ -625,7 +625,9 @@ func extractAnthropicSSEDataLine(line string) (string, bool) {
 	return line[start:], true
 }
 
-func (s *GatewayService) parseSSEUsagePassthrough(data string, usage *ClaudeUsage) {
+// parseSSEUsagePassthrough 从 Anthropic SSE data 行提取 usage（包级函数：
+// Anthropic 平台 passthrough 与国产供应商原生 Anthropic 直通共用）。
+func parseSSEUsagePassthrough(data string, usage *ClaudeUsage) {
 	if usage == nil || data == "" || data == "[DONE]" {
 		return
 	}
@@ -730,8 +732,12 @@ func parseClaudeUsageFromResponseBody(body []byte) *ClaudeUsage {
 	return usage
 }
 
-func (s *GatewayService) invalidNonStreamingJSONFailoverError(
+// invalidNonStreamingJSONFailoverError 把"上游 2xx 返回非 JSON body"归一为
+// failover 错误（包级函数：Anthropic 平台 passthrough 与国产供应商原生
+// Anthropic 直通共用）。
+func invalidNonStreamingJSONFailoverError(
 	ctx context.Context,
+	rateLimitService *RateLimitService,
 	resp *http.Response,
 	account *Account,
 	body []byte,
@@ -759,11 +765,11 @@ func (s *GatewayService) invalidNonStreamingJSONFailoverError(
 		parseErr,
 	)
 
-	if s.rateLimitService != nil && account != nil {
+	if rateLimitService != nil && account != nil {
 		if len(requestedModel) > 0 {
-			s.rateLimitService.HandleUpstreamError(ctx, account, statusCode, resp.Header, body, requestedModel[0])
+			rateLimitService.HandleUpstreamError(ctx, account, statusCode, resp.Header, body, requestedModel[0])
 		} else {
-			s.rateLimitService.HandleUpstreamError(ctx, account, statusCode, resp.Header, body)
+			rateLimitService.HandleUpstreamError(ctx, account, statusCode, resp.Header, body)
 		}
 	}
 
@@ -798,7 +804,7 @@ func (s *GatewayService) handleNonStreamingResponseAnthropicAPIKeyPassthrough(
 	if resp.StatusCode >= http.StatusOK && resp.StatusCode < http.StatusMultipleChoices {
 		var raw json.RawMessage
 		if err := json.Unmarshal(body, &raw); err != nil {
-			return nil, s.invalidNonStreamingJSONFailoverError(ctx, resp, account, body, err)
+			return nil, invalidNonStreamingJSONFailoverError(ctx, s.rateLimitService, resp, account, body, err)
 		}
 	}
 
diff --git a/backend/internal/service/gateway_upstream_response.go b/backend/internal/service/gateway_upstream_response.go
index 5a9fedf96433..20af32314f82 100644
--- a/backend/internal/service/gateway_upstream_response.go
+++ b/backend/internal/service/gateway_upstream_response.go
@@ -1395,7 +1395,7 @@ func (s *GatewayService) handleNonStreamingResponse(ctx context.Context, resp *h
 	}
 	if err := json.Unmarshal(body, &response); err != nil {
 		if resp.StatusCode >= http.StatusOK && resp.StatusCode < http.StatusMultipleChoices {
-			return nil, s.invalidNonStreamingJSONFailoverError(ctx, resp, account, body, err, mappedModel)
+			return nil, invalidNonStreamingJSONFailoverError(ctx, s.rateLimitService, resp, account, body, err, mappedModel)
 		}
 		return nil, fmt.Errorf("parse response: %w", err)
 	}
diff --git a/backend/internal/service/openai_account_scheduler.go b/backend/internal/service/openai_account_scheduler.go
index 6db0db766d1c..1daf81988cba 100644
--- a/backend/internal/service/openai_account_scheduler.go
+++ b/backend/internal/service/openai_account_scheduler.go
@@ -381,7 +381,7 @@ func (s *defaultOpenAIAccountScheduler) Select(
 	}()
 
 	previousResponseID := strings.TrimSpace(req.PreviousResponseID)
-	if previousResponseID != "" && normalizeOpenAICompatiblePlatform(req.Platform) == PlatformOpenAI &&
+	if previousResponseID != "" && NormalizeOpenAICompatiblePlatform(req.Platform) == PlatformOpenAI &&
 		(!req.StickyWeighted || !req.PreviousResponseCanMove) {
 		selection, err := s.service.selectAccountByPreviousResponseIDForCapability(
 			ctx,
@@ -486,7 +486,7 @@ func (s *defaultOpenAIAccountScheduler) selectBySessionHash(
 		_ = s.service.deleteStickySessionAccountID(ctx, req.GroupID, sessionHash)
 		return nil, false, nil
 	}
-	if shouldClearStickySession(account, req.RequestedModel) || account.Platform != normalizeOpenAICompatiblePlatform(req.Platform) || !account.IsOpenAICompatible() || !account.IsSchedulable() {
+	if shouldClearStickySession(account, req.RequestedModel) || account.Platform != NormalizeOpenAICompatiblePlatform(req.Platform) || !account.IsOpenAICompatible() || !account.IsSchedulable() {
 		_ = s.service.deleteStickySessionAccountID(ctx, req.GroupID, sessionHash)
 		return nil, false, nil
 	}
@@ -1406,7 +1406,7 @@ func (s *defaultOpenAIAccountScheduler) selectByLoadBalance(
 			filterStats.exclude("not_schedulable")
 			continue
 		}
-		if account.Platform != normalizeOpenAICompatiblePlatform(req.Platform) || !account.IsOpenAICompatible() {
+		if account.Platform != NormalizeOpenAICompatiblePlatform(req.Platform) || !account.IsOpenAICompatible() {
 			filterStats.exclude("platform_mismatch")
 			continue
 		}
@@ -2125,7 +2125,7 @@ func (s *OpenAIGatewayService) selectAccountWithScheduler(
 		return selection, decision, err
 	}
 	// The circuit only ever quarantines PlatformOpenAI accounts.
-	if normalizeOpenAICompatiblePlatform(platform) != PlatformOpenAI {
+	if NormalizeOpenAICompatiblePlatform(platform) != PlatformOpenAI {
 		return selection, decision, err
 	}
 	blocked := s.getOpenAIProxyStreamCircuit().activeBlockCount(time.Now())
@@ -2161,7 +2161,7 @@ func (s *OpenAIGatewayService) selectAccountWithSchedulerOnce(
 	if requiredImageCapability == "" {
 		ctx = s.withOpenAIProfitControlGate(ctx, groupID)
 	}
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 	decision := OpenAIAccountScheduleDecision{}
 	scheduler := s.getOpenAIAccountScheduler(ctx)
 	if scheduler == nil {
diff --git a/backend/internal/service/openai_apikey_responses_probe.go b/backend/internal/service/openai_apikey_responses_probe.go
index 79db519e6f6f..645a581694a4 100644
--- a/backend/internal/service/openai_apikey_responses_probe.go
+++ b/backend/internal/service/openai_apikey_responses_probe.go
@@ -119,7 +119,27 @@ func (s *AccountTestService) ProbeOpenAIAPIKeyResponsesSupport(ctx context.Conte
 		logger.LegacyPrintf("service.openai_probe", "probe_load_account_failed: account_id=%d err=%v", accountID, err)
 		return
 	}
-	if account.Platform != PlatformOpenAI || account.Type != AccountTypeAPIKey {
+	if account.Type != AccountTypeAPIKey {
+		return
+	}
+	if account.IsCNProvider() {
+		// 国产 OpenAI 兼容上游（kimi/zhipu/deepseek）普遍仅支持 /v1/chat/completions，
+		// 不存在 /v1/responses 端点。直接落标 false 走 Chat Completions 直转，跳过网络探测。
+		// 例外：deepseek 的 responses 协议账号（api_protocol=responses）使用官方原生
+		// /responses 端点，落标 force_responses 强制走 Responses 路径。
+		if account.GetAPIProtocol() == APIProtocolResponses {
+			_ = s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{
+				openai_compat.ExtraKeyResponsesMode:      string(openai_compat.ResponsesSupportModeForceResponses),
+				openai_compat.ExtraKeyResponsesSupported: true,
+			})
+			return
+		}
+		_ = s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{
+			openai_compat.ExtraKeyResponsesSupported: false,
+		})
+		return
+	}
+	if account.Platform != PlatformOpenAI {
 		// 仅 OpenAI APIKey 账号需要探测；其他账号类型无能力差异。
 		return
 	}
diff --git a/backend/internal/service/openai_embeddings.go b/backend/internal/service/openai_embeddings.go
index c0c59002977d..e68017a45373 100644
--- a/backend/internal/service/openai_embeddings.go
+++ b/backend/internal/service/openai_embeddings.go
@@ -46,11 +46,13 @@ func (s *OpenAIGatewayService) ForwardEmbeddings(
 		zap.String("upstream_model", upstreamModel),
 	)
 
-	apiKey := account.GetOpenAIApiKey()
+	apiKey := strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
 	if apiKey == "" {
 		return nil, fmt.Errorf("account %d missing api_key", account.ID)
 	}
-	baseURL := account.GetOpenAIBaseURL()
+	// 协议感知：Anthropic 协议账号的凭证 base_url 指向 /anthropic 端点，
+	// embeddings 需使用 OpenAI 格式 base。
+	baseURL := account.GetOpenAIFormatBaseURL()
 	if baseURL == "" {
 		baseURL = "https://api.openai.com"
 	}
diff --git a/backend/internal/service/openai_gateway_cc_pipeline.go b/backend/internal/service/openai_gateway_cc_pipeline.go
index 8080d49e5c35..ce9dd54d370c 100644
--- a/backend/internal/service/openai_gateway_cc_pipeline.go
+++ b/backend/internal/service/openai_gateway_cc_pipeline.go
@@ -150,7 +150,7 @@ func (s *OpenAIGatewayService) openAIChatCompletionsTargetURL(account *Account)
 // resolveCCFallbackTarget 解析两条 CC 回退路径共用的账号凭证与上游端点
 // （回退路径仅面向 APIKey 账号，凭证恒为 openai api_key）。
 func (s *OpenAIGatewayService) resolveCCFallbackTarget(account *Account) (apiKey string, targetURL string, err error) {
-	apiKey = account.GetOpenAIApiKey()
+	apiKey = strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
 	if apiKey == "" {
 		return "", "", fmt.Errorf("account %d missing api_key", account.ID)
 	}
diff --git a/backend/internal/service/openai_gateway_chat_completions.go b/backend/internal/service/openai_gateway_chat_completions.go
index 87973304c0a6..fed651bd7016 100644
--- a/backend/internal/service/openai_gateway_chat_completions.go
+++ b/backend/internal/service/openai_gateway_chat_completions.go
@@ -87,6 +87,14 @@ func (s *OpenAIGatewayService) ForwardAsChatCompletions(
 		return s.forwardAsRawChatCompletions(ctx, c, account, body, defaultMappedModel)
 	}
 
+	// 入口分流（国产供应商 Anthropic 协议）：上游为供应商原生 Anthropic 端点，
+	// CC 入站请求经 CC→Responses→Anthropic 转换链直通该端点。必须先于
+	// ShouldUseResponsesAPI 分流：该类账号经 probe 落标
+	// openai_responses_supported=false，会先命中下方的 CC 直转分支。
+	if account.IsAnthropicProtocol() {
+		return s.forwardChatCompletionsViaNativeAnthropic(ctx, c, account, body, defaultMappedModel)
+	}
+
 	// 入口分流：APIKey 账号 + 强制或已探测确认上游不支持 Responses，走 CC 直转。
 	// 自动模式下标记缺失（未探测）按"现状即证据"原则继续走下方原 Responses 转换路径。
 	if account.Type == AccountTypeAPIKey && !openai_compat.ShouldUseResponsesAPI(account.Extra) {
diff --git a/backend/internal/service/openai_gateway_chat_completions_anthropic_native.go b/backend/internal/service/openai_gateway_chat_completions_anthropic_native.go
new file mode 100644
index 000000000000..0ef89a7ed02e
--- /dev/null
+++ b/backend/internal/service/openai_gateway_chat_completions_anthropic_native.go
@@ -0,0 +1,427 @@
+package service
+
+// 国产供应商 Anthropic 协议账号的 CC 入站反向路径。
+//
+// 客户端说 OpenAI Chat Completions、上游是供应商原生 Anthropic 端点
+// （api_protocol=anthropic）时的交叉组合：请求 CC→Responses→Anthropic 转换，
+// 响应 Anthropic→Responses→CC 转换。转换链与 Anthropic 平台的
+// gateway_forward_as_chat_completions.go 完全一致（复用同一组 apicompat
+// 状态机），仅上游发送/错误处理对齐 OpenAI 网关语义。
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
+	"github.com/gin-gonic/gin"
+	"go.uber.org/zap"
+)
+
+// forwardChatCompletionsViaNativeAnthropic serves OpenAI /v1/chat/completions
+// clients through a CN provider's native Anthropic endpoint.
+//
+// Conversion chain:
+//
+//	Request:  Chat Completions → Responses → Anthropic (chained)
+//	Response: Anthropic events → Responses events → CC chunks (chained state machines)
+func (s *OpenAIGatewayService) forwardChatCompletionsViaNativeAnthropic(
+	ctx context.Context,
+	c *gin.Context,
+	account *Account,
+	body []byte,
+	defaultMappedModel string,
+) (*OpenAIForwardResult, error) {
+	startTime := time.Now()
+
+	// 1. Parse Chat Completions request
+	var ccReq apicompat.ChatCompletionsRequest
+	if err := json.Unmarshal(body, &ccReq); err != nil {
+		writeChatCompletionsError(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
+		return nil, fmt.Errorf("parse chat completions request: %w", err)
+	}
+	originalModel := ccReq.Model
+	if strings.TrimSpace(originalModel) == "" {
+		writeChatCompletionsError(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return nil, fmt.Errorf("missing model in request")
+	}
+	clientStream := ccReq.Stream
+	includeUsage := ccReq.StreamOptions != nil && ccReq.StreamOptions.IncludeUsage
+
+	// 2. Convert CC → Responses → Anthropic (chained conversion)
+	responsesReq, err := apicompat.ChatCompletionsToResponses(&ccReq)
+	if err != nil {
+		writeChatCompletionsError(c, http.StatusBadRequest, "invalid_request_error", "Failed to convert request")
+		return nil, fmt.Errorf("convert chat completions to responses: %w", err)
+	}
+	anthropicReq, err := apicompat.ResponsesToAnthropicRequest(responsesReq)
+	if err != nil {
+		writeChatCompletionsError(c, http.StatusBadRequest, "invalid_request_error", "Failed to convert request")
+		return nil, fmt.Errorf("convert responses to anthropic: %w", err)
+	}
+
+	// 3. Model mapping（OpenAI 网关统一入口的映射语义）
+	billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel)
+	upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
+	anthropicReq.Model = upstreamModel
+
+	// 4. Force upstream streaming（客户端原始终决定响应格式；
+	// 上游恒为流式，非流式由缓冲路径组装）。
+	anthropicReq.Stream = true
+	reqStream := true
+
+	logger.L().Debug("openai chat_completions: forwarding via native anthropic endpoint",
+		zap.Int64("account_id", account.ID),
+		zap.String("original_model", originalModel),
+		zap.String("billing_model", billingModel),
+		zap.String("upstream_model", upstreamModel),
+		zap.Bool("client_stream", clientStream),
+	)
+
+	anthropicBody, err := json.Marshal(anthropicReq)
+	if err != nil {
+		return nil, fmt.Errorf("marshal anthropic request: %w", err)
+	}
+
+	// 与 /v1/messages 直通路径相同的 pre-filter。
+	anthropicBody = StripEmptyTextBlocks(anthropicBody)
+	anthropicBody = FilterWebSearchHistoryBlocks(anthropicBody, upstreamModel)
+	anthropicBody = enforceCacheControlLimit(anthropicBody)
+
+	apiKey := strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
+	if apiKey == "" {
+		return nil, fmt.Errorf("account %d missing api_key", account.ID)
+	}
+	targetURL, err := s.nativeAnthropicTargetURL(account)
+	if err != nil {
+		return nil, err
+	}
+
+	proxyURL := ""
+	if account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+
+	upstreamCtx, releaseUpstreamCtx := detachStreamUpstreamContext(ctx, reqStream)
+	upstreamReq, _, err := s.buildNativeAnthropicUpstreamRequest(upstreamCtx, c, account, anthropicBody, apiKey, targetURL)
+	releaseUpstreamCtx()
+	if err != nil {
+		return nil, fmt.Errorf("build upstream request: %w", err)
+	}
+
+	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+	if err != nil {
+		return nil, s.handleOpenAIUpstreamTransportError(ctx, c, account, err, true)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	if resp.StatusCode >= 400 {
+		respBody, upstreamMsg := s.readOpenAIUpstreamError(resp)
+		if foErr := s.failoverOpenAIUpstreamHTTPError(ctx, c, account, resp, respBody, upstreamMsg, upstreamModel); foErr != nil {
+			return nil, foErr
+		}
+		writeChatCompletionsError(c, mapUpstreamStatusCode(resp.StatusCode), "server_error", upstreamMsg)
+		return nil, fmt.Errorf("upstream error: %d %s", resp.StatusCode, upstreamMsg)
+	}
+
+	reasoningEffort := extractCCReasoningEffortFromBody(body)
+	reasoningEffort = ApplyThinkingEnabledFallback(reasoningEffort, body, billingModel)
+
+	if clientStream {
+		return s.handleCCStreamingFromNativeAnthropic(resp, c, originalModel, billingModel, upstreamModel, reasoningEffort, startTime, includeUsage)
+	}
+	return s.handleCCBufferedFromNativeAnthropic(resp, c, originalModel, billingModel, upstreamModel, reasoningEffort, startTime)
+}
+
+// handleCCBufferedFromNativeAnthropic reads Anthropic SSE events, assembles the
+// full response, then converts Anthropic → Responses → Chat Completions.
+func (s *OpenAIGatewayService) handleCCBufferedFromNativeAnthropic(
+	resp *http.Response,
+	c *gin.Context,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	reasoningEffort *string,
+	startTime time.Time,
+) (*OpenAIForwardResult, error) {
+	requestID := resp.Header.Get("x-request-id")
+
+	scanner := bufio.NewScanner(resp.Body)
+	maxLineSize := defaultMaxLineSize
+	if s.cfg != nil && s.cfg.Gateway.MaxLineSize > 0 {
+		maxLineSize = s.cfg.Gateway.MaxLineSize
+	}
+	scanner.Buffer(make([]byte, 0, 64*1024), maxLineSize)
+
+	var finalResp *apicompat.AnthropicResponse
+	var usage ClaudeUsage
+
+	for scanner.Scan() {
+		line := scanner.Text()
+		// SSE 规范允许 `event:xxx`（冒号后无空格）：Kimi 等 Anthropic 兼容上游
+		// 返回紧凑格式，严格匹配 "event: " 会丢弃全部事件（#4653 同根因）。
+		if _, ok := extractOpenAISSEEventLine(line); !ok {
+			continue
+		}
+
+		if !scanner.Scan() {
+			break
+		}
+		payload, ok := extractOpenAISSEDataLine(scanner.Text())
+		if !ok {
+			continue
+		}
+
+		var event apicompat.AnthropicStreamEvent
+		if err := json.Unmarshal([]byte(payload), &event); err != nil {
+			continue
+		}
+
+		if event.Type == "message_start" && event.Message != nil {
+			finalResp = event.Message
+			mergeAnthropicUsage(&usage, event.Message.Usage)
+		}
+		if event.Type == "message_delta" {
+			if event.Usage != nil {
+				mergeAnthropicUsage(&usage, *event.Usage)
+			}
+			if event.Delta != nil && event.Delta.StopReason != "" && finalResp != nil {
+				finalResp.StopReason = apicompat.AnthropicStopReasonPtr(event.Delta.StopReason)
+			}
+		}
+		if event.Type == "content_block_start" && event.ContentBlock != nil && finalResp != nil {
+			finalResp.Content = append(finalResp.Content, *event.ContentBlock)
+		}
+		if event.Type == "content_block_delta" && event.Delta != nil && finalResp != nil && event.Index != nil {
+			idx := *event.Index
+			if idx < len(finalResp.Content) {
+				switch event.Delta.Type {
+				case "text_delta":
+					finalResp.Content[idx].Text += event.Delta.Text
+				case "thinking_delta":
+					finalResp.Content[idx].Thinking += event.Delta.Thinking
+				case "input_json_delta":
+					finalResp.Content[idx].Input = appendRawJSON(finalResp.Content[idx].Input, event.Delta.PartialJSON)
+				}
+			}
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai cc via native anthropic buffered: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+
+	if finalResp == nil {
+		writeChatCompletionsError(c, http.StatusBadGateway, "server_error", "Upstream stream ended without a response")
+		return nil, fmt.Errorf("upstream stream ended without response")
+	}
+
+	if usage.InputTokens > 0 || usage.OutputTokens > 0 {
+		finalResp.Usage = apicompat.AnthropicUsage{
+			InputTokens:              usage.InputTokens,
+			OutputTokens:             usage.OutputTokens,
+			CacheCreationInputTokens: usage.CacheCreationInputTokens,
+			CacheReadInputTokens:     usage.CacheReadInputTokens,
+		}
+	}
+
+	responsesResp := apicompat.AnthropicToResponsesResponse(finalResp)
+	ccResp := apicompat.ResponsesToChatCompletions(responsesResp, originalModel)
+
+	if s.responseHeaderFilter != nil {
+		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	}
+	// 非流式响应必须是 application/json（上游被强制流式，透传头会污染）。
+	c.Writer.Header().Set("Content-Type", "application/json; charset=utf-8")
+	if respBytes, err := json.Marshal(ccResp); err == nil {
+		respBytes = reverseToolNamesIfPresent(c, respBytes)
+		c.Data(http.StatusOK, "application/json; charset=utf-8", respBytes)
+	} else {
+		c.JSON(http.StatusOK, ccResp)
+	}
+
+	return &OpenAIForwardResult{
+		RequestID:        requestID,
+		Usage:            claudeUsageToOpenAIUsage(&usage),
+		Model:            originalModel,
+		BillingModel:     billingModel,
+		UpstreamModel:    upstreamModel,
+		UpstreamEndpoint: "/v1/messages",
+		ReasoningEffort:  reasoningEffort,
+		Stream:           false,
+		Duration:         time.Since(startTime),
+	}, nil
+}
+
+// handleCCStreamingFromNativeAnthropic reads Anthropic SSE events, converts each
+// to Responses events, then to Chat Completions chunks, and writes them.
+func (s *OpenAIGatewayService) handleCCStreamingFromNativeAnthropic(
+	resp *http.Response,
+	c *gin.Context,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	reasoningEffort *string,
+	startTime time.Time,
+	includeUsage bool,
+) (*OpenAIForwardResult, error) {
+	requestID := resp.Header.Get("x-request-id")
+
+	if s.responseHeaderFilter != nil {
+		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	}
+	c.Writer.Header().Set("Content-Type", "text/event-stream")
+	c.Writer.Header().Set("Cache-Control", "no-cache")
+	c.Writer.Header().Set("Connection", "keep-alive")
+	c.Writer.Header().Set("X-Accel-Buffering", "no")
+	c.Writer.WriteHeader(http.StatusOK)
+
+	anthState := apicompat.NewAnthropicEventToResponsesState()
+	anthState.Model = originalModel
+	ccState := apicompat.NewResponsesEventToChatState()
+	ccState.Model = originalModel
+	ccState.IncludeUsage = includeUsage
+
+	var usage ClaudeUsage
+	var firstTokenMs *int
+	firstChunk := true
+	clientDisconnected := false
+
+	scanner := bufio.NewScanner(resp.Body)
+	maxLineSize := defaultMaxLineSize
+	if s.cfg != nil && s.cfg.Gateway.MaxLineSize > 0 {
+		maxLineSize = s.cfg.Gateway.MaxLineSize
+	}
+	scanner.Buffer(make([]byte, 0, 64*1024), maxLineSize)
+
+	resultWithUsage := func() *OpenAIForwardResult {
+		return &OpenAIForwardResult{
+			RequestID:        requestID,
+			Usage:            claudeUsageToOpenAIUsage(&usage),
+			Model:            originalModel,
+			BillingModel:     billingModel,
+			UpstreamModel:    upstreamModel,
+			UpstreamEndpoint: "/v1/messages",
+			ReasoningEffort:  reasoningEffort,
+			Stream:           true,
+			Duration:         time.Since(startTime),
+			FirstTokenMs:     firstTokenMs,
+			ClientDisconnect: clientDisconnected,
+		}
+	}
+
+	writeChunk := func(chunk apicompat.ChatCompletionsChunk) bool {
+		if clientDisconnected {
+			return false // 已断开：不再写客户端，只排水上游累计 usage
+		}
+		sse, err := apicompat.ChatChunkToSSE(chunk)
+		if err != nil {
+			return false
+		}
+		out := string(reverseToolNamesIfPresent(c, []byte(sse)))
+		if _, err := fmt.Fprint(c.Writer, out); err != nil {
+			clientDisconnected = true
+			return false
+		}
+		return false
+	}
+
+	processAnthropicEvent := func(event *apicompat.AnthropicStreamEvent) bool {
+		if firstChunk {
+			firstChunk = false
+			ms := int(time.Since(startTime).Milliseconds())
+			firstTokenMs = &ms
+		}
+
+		// usage 恒累计（含客户端断开后的排水阶段，payg 上游照常计费）。
+		if event.Type == "message_delta" && event.Usage != nil {
+			mergeAnthropicUsage(&usage, *event.Usage)
+		}
+		if event.Type == "message_start" && event.Message != nil {
+			mergeAnthropicUsage(&usage, event.Message.Usage)
+		}
+
+		// 客户端已断开：跳过转换与写出，继续读上游直到流结束（usage 完整、
+		// 连接及时归还），不再提前 return。
+		if clientDisconnected {
+			return false
+		}
+
+		responsesEvents := apicompat.AnthropicEventToResponsesEvents(event, anthState)
+		for _, resEvt := range responsesEvents {
+			ccChunks := apicompat.ResponsesEventToChatChunks(&resEvt, ccState)
+			for _, chunk := range ccChunks {
+				writeChunk(chunk)
+			}
+		}
+		if len(responsesEvents) > 0 {
+			c.Writer.Flush()
+		}
+		return false
+	}
+
+	for scanner.Scan() {
+		line := scanner.Text()
+		if _, ok := extractOpenAISSEEventLine(line); !ok {
+			continue
+		}
+
+		if !scanner.Scan() {
+			break
+		}
+		payload, ok := extractOpenAISSEDataLine(scanner.Text())
+		if !ok {
+			continue
+		}
+
+		var event apicompat.AnthropicStreamEvent
+		if err := json.Unmarshal([]byte(payload), &event); err != nil {
+			continue
+		}
+
+		if processAnthropicEvent(&event) {
+			return resultWithUsage(), nil
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai cc via native anthropic stream: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+
+	// Finalize both state machines（客户端已断开时仍执行，保证 usage 汇总完整）。
+	finalResEvents := apicompat.FinalizeAnthropicResponsesStream(anthState)
+	for _, resEvt := range finalResEvents {
+		ccChunks := apicompat.ResponsesEventToChatChunks(&resEvt, ccState)
+		for _, chunk := range ccChunks {
+			writeChunk(chunk) //nolint:errcheck
+		}
+	}
+	finalCCChunks := apicompat.FinalizeResponsesChatStream(ccState)
+	for _, chunk := range finalCCChunks {
+		writeChunk(chunk) //nolint:errcheck
+	}
+
+	if !clientDisconnected {
+		fmt.Fprint(c.Writer, "data: [DONE]\n\n") //nolint:errcheck
+		c.Writer.Flush()
+	}
+
+	return resultWithUsage(), nil
+}
diff --git a/backend/internal/service/openai_gateway_count_tokens.go b/backend/internal/service/openai_gateway_count_tokens.go
index e56b33df4506..f20b45b5269c 100644
--- a/backend/internal/service/openai_gateway_count_tokens.go
+++ b/backend/internal/service/openai_gateway_count_tokens.go
@@ -43,6 +43,12 @@ type openAIInputTokensCountPrepared struct {
 // locally. Grok does not expose a compatible token-counting endpoint, so this
 // path deliberately avoids account selection, credentials, and upstream calls.
 func EstimateGrokCountTokens(body []byte) (int, error) {
+	return estimateAnthropicCountTokensLocally(body)
+}
+
+// estimateAnthropicCountTokensLocally 走 Anthropic→Responses→tiktoken 链本地估算
+// count_tokens，不发任何上游请求（上游无兼容端点的平台使用）。
+func estimateAnthropicCountTokensLocally(body []byte) (int, error) {
 	var anthropicReq apicompat.AnthropicRequest
 	if err := json.Unmarshal(body, &anthropicReq); err != nil {
 		return 0, fmt.Errorf("parse anthropic count_tokens request: %w", err)
@@ -64,7 +70,7 @@ func EstimateGrokCountTokens(body []byte) (int, error) {
 		ToolChoice:   responsesReq.ToolChoice,
 	})
 	if err != nil {
-		return 0, fmt.Errorf("estimate grok input tokens: %w", err)
+		return 0, fmt.Errorf("estimate input tokens: %w", err)
 	}
 	if estimated < openAIInputTokensFallbackMinimum {
 		estimated = openAIInputTokensFallbackMinimum
@@ -86,6 +92,31 @@ func (s *OpenAIGatewayService) ForwardCountTokensAsAnthropic(
 		return fmt.Errorf("count_tokens: missing account")
 	}
 
+	// 国产供应商 Anthropic 协议：上游有原生 /v1/messages/count_tokens 端点，
+	// 直接透传（仅模型名映射），不走 /v1/responses/input_tokens 估算。
+	if account.IsAnthropicProtocol() {
+		return s.forwardCountTokensViaNativeAnthropic(ctx, c, account, body, defaultMappedModel)
+	}
+
+	// 国产供应商其余协议（chat_completions / responses）：三家上游均无
+	// OpenAI 兼容的 /v1/responses/input_tokens 端点，与 Grok 一样本地估算，
+	// 不发上游请求（Claude Code 客户端会高频调用 count_tokens）。
+	if account.IsCNProvider() {
+		estimated, err := estimateAnthropicCountTokensLocally(body)
+		if err != nil {
+			writeAnthropicCountTokensError(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
+			return fmt.Errorf("count_tokens: estimate cn provider input tokens: %w", err)
+		}
+		logger.L().Debug("openai count_tokens: cn provider local estimate",
+			zap.Int64("account_id", account.ID),
+			zap.Int("estimated_input_tokens", estimated),
+		)
+		c.JSON(http.StatusOK, gin.H{
+			"input_tokens": estimated,
+		})
+		return nil
+	}
+
 	prepared, err := prepareOpenAIInputTokensCountRequest(body, account, defaultMappedModel)
 	if err != nil {
 		writeAnthropicCountTokensError(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
diff --git a/backend/internal/service/openai_gateway_forward.go b/backend/internal/service/openai_gateway_forward.go
index a3ba6ebab910..a262fb60ccb3 100644
--- a/backend/internal/service/openai_gateway_forward.go
+++ b/backend/internal/service/openai_gateway_forward.go
@@ -108,6 +108,13 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 		return s.forwardGrokResponses(ctx, c, account, body, originalModel, reqStream, startTime)
 	}
 
+	// CN 供应商 anthropic 协议账号：/v1/responses 入站是交叉协议组合
+	// （Responses 客户端 × Anthropic 上游），转成 Anthropic 请求走原生端点。
+	// 不能落到下面的 raw-CC 分支——其 URL 构造会把 anthropic base 当 CC base 用。
+	if account.IsAnthropicProtocol() {
+		return s.forwardResponsesViaNativeAnthropic(ctx, c, account, body, reqModel)
+	}
+
 	if shouldForwardOpenAIResponsesViaRawChatCompletions(account) {
 		return s.forwardResponsesViaRawChatCompletions(ctx, c, account, body)
 	}
@@ -1051,13 +1058,17 @@ func (s *OpenAIGatewayService) buildUpstreamRequest(ctx context.Context, c *gin.
 			if err != nil {
 				return nil, err
 			}
-			targetURL = buildOpenAIResponsesURL(validatedURL)
+			targetURL = buildOpenAIResponsesURLForPlatform(account.Platform, validatedURL)
 		}
 	default:
 		targetURL = openaiPlatformAPIURL
 	}
 	targetURL = appendOpenAIResponsesRequestPathSuffix(targetURL, openAIResponsesRequestPathSuffix(c))
 
+	// DeepSeek 原生 Responses 端点为无状态实现：强制 store=false、清除
+	// previous_response_id，避免携带状态字段被上游拒绝。
+	body = normalizeDeepSeekResponsesRequestBody(account, body)
+
 	req, err := http.NewRequestWithContext(ctx, "POST", targetURL, bytes.NewReader(body))
 	if err != nil {
 		return nil, err
diff --git a/backend/internal/service/openai_gateway_messages.go b/backend/internal/service/openai_gateway_messages.go
index 268aec7b5a7e..43f4977f2d84 100644
--- a/backend/internal/service/openai_gateway_messages.go
+++ b/backend/internal/service/openai_gateway_messages.go
@@ -36,6 +36,15 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic(
 ) (*OpenAIForwardResult, error) {
 	beginUpstreamResponseModelObservation(c)
 
+	// 入口分流（国产供应商 Anthropic 协议）：上游为供应商原生 Anthropic 端点时，
+	// /v1/messages 请求零转换直通（仅模型名映射 + 少量 body 清洗），完整保留
+	// thinking / tool_use / cache 语义，适配 Claude Code 等原生客户端。
+	// 必须先于 ShouldUseResponsesAPI 分流：Anthropic 协议账号经 probe 落标
+	// openai_responses_supported=false，会先命中下方的 CC 直转分支。
+	if account.IsAnthropicProtocol() {
+		return s.forwardAnthropicViaNativeAnthropicEndpoint(ctx, c, account, body, defaultMappedModel)
+	}
+
 	// 入口分流：APIKey 账号 + 上游不支持 Responses API → 走 CC 直转（与
 	// ForwardAsChatCompletions 对称）。缺少此分流时，/v1/messages 入站请求
 	// 会被无条件转为 Responses 格式发往上游 /v1/responses，导致只支持
diff --git a/backend/internal/service/openai_gateway_messages_anthropic_native.go b/backend/internal/service/openai_gateway_messages_anthropic_native.go
new file mode 100644
index 000000000000..e785f3755fda
--- /dev/null
+++ b/backend/internal/service/openai_gateway_messages_anthropic_native.go
@@ -0,0 +1,619 @@
+package service
+
+// 国产供应商（kimi/zhipu/deepseek）原生 Anthropic 端点直通路径。
+//
+// 当账号 credentials["api_protocol"] = "anthropic" 时，入站 /v1/messages 请求
+// 不再做 Anthropic→CC→Anthropic 双重转换，而是零转换直通供应商的官方
+// Anthropic 兼容端点（如 https://open.bigmodel.cn/api/anthropic/v1/messages），
+// 适配 Claude Code 等原生 Anthropic 客户端。转发骨架以
+// gateway_anthropic_passthrough.go 的 APIKey 透传为模板（字节级 SSE 中继 +
+// usage 解析），错误/failover 语义对齐 OpenAI 网关其他路径
+// （failoverOpenAIUpstreamHTTPError / handleAnthropicErrorResponse）。
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"sync/atomic"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// forwardAnthropicViaNativeAnthropicEndpoint 将 Anthropic Messages 请求零转换
+// 直通到国产供应商的原生 Anthropic 端点。仅做模型名映射与少量 body 清洗
+// （空文本块 / web-search 历史块），协议本身不转换。
+func (s *OpenAIGatewayService) forwardAnthropicViaNativeAnthropicEndpoint(
+	ctx context.Context,
+	c *gin.Context,
+	account *Account,
+	body []byte,
+	defaultMappedModel string,
+) (*OpenAIForwardResult, error) {
+	startTime := time.Now()
+
+	originalModel := strings.TrimSpace(gjson.GetBytes(body, "model").String())
+	if originalModel == "" {
+		writeAnthropicError(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return nil, fmt.Errorf("missing model in request")
+	}
+	clientStream := gjson.GetBytes(body, "stream").Bool()
+
+	billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel)
+	upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
+	if upstreamModel != originalModel {
+		rewritten, err := sjson.SetBytes(body, "model", upstreamModel)
+		if err != nil {
+			return nil, fmt.Errorf("rewrite model: %w", err)
+		}
+		body = rewritten
+	}
+
+	// 与 Anthropic 平台 passthrough 相同的 pre-filter：剥离空文本块与上游
+	// 无法接受的 web-search 历史块（GLM/Kimi/DeepSeek 对 server_tool_use 400）。
+	body = StripEmptyTextBlocks(body)
+	body = FilterWebSearchHistoryBlocks(body, upstreamModel)
+
+	logger.LegacyPrintf("service.gateway", "[CN Anthropic 直通] account=%d(%s) platform=%s model=%s upstream=%s stream=%v",
+		account.ID, account.Name, account.Platform, originalModel, upstreamModel, clientStream)
+
+	apiKey := strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
+	if apiKey == "" {
+		return nil, fmt.Errorf("account %d missing api_key", account.ID)
+	}
+	targetURL, err := s.nativeAnthropicTargetURL(account)
+	if err != nil {
+		return nil, err
+	}
+
+	proxyURL := ""
+	if account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+
+	upstreamCtx, releaseUpstreamCtx := detachStreamUpstreamContext(ctx, clientStream)
+	upstreamReq, _, err := s.buildNativeAnthropicUpstreamRequest(upstreamCtx, c, account, body, apiKey, targetURL)
+	releaseUpstreamCtx()
+	if err != nil {
+		return nil, err
+	}
+
+	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+	if err != nil {
+		return nil, s.handleOpenAIUpstreamTransportError(ctx, c, account, err, true)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	if resp.StatusCode >= 400 {
+		respBody, upstreamMsg := s.readOpenAIUpstreamError(resp)
+		if foErr := s.failoverOpenAIUpstreamHTTPError(ctx, c, account, resp, respBody, upstreamMsg, upstreamModel); foErr != nil {
+			return nil, foErr
+		}
+		// 非 failover 错误：经共享 compat handler 以 Anthropic 格式回写
+		// （透传规则、ops 记录、cyber_policy 与 CC 回退路径一致）。
+		return s.handleAnthropicErrorResponse(resp, c, account, billingModel)
+	}
+
+	if clientStream {
+		return s.handleNativeAnthropicStreamingResponse(ctx, resp, c, account, originalModel, billingModel, upstreamModel, startTime)
+	}
+	return s.handleNativeAnthropicBufferedResponse(ctx, resp, c, account, originalModel, billingModel, upstreamModel, startTime)
+}
+
+// nativeAnthropicTargetURL 组装国产供应商原生 Anthropic messages 端点。
+// 第三方端点保持朴素路径，不附加 ?beta=true。
+func (s *OpenAIGatewayService) nativeAnthropicTargetURL(account *Account) (string, error) {
+	baseURL := strings.TrimSpace(account.GetAnthropicProtocolBaseURL())
+	if baseURL == "" {
+		return "", fmt.Errorf("account %d has no anthropic protocol base url", account.ID)
+	}
+	validatedURL, err := s.validateUpstreamBaseURL(baseURL)
+	if err != nil {
+		return "", fmt.Errorf("invalid base_url: %w", err)
+	}
+	return strings.TrimRight(validatedURL, "/") + "/v1/messages", nil
+}
+
+func (s *OpenAIGatewayService) buildNativeAnthropicUpstreamRequest(
+	ctx context.Context,
+	c *gin.Context,
+	account *Account,
+	body []byte,
+	apiKey string,
+	targetURL string,
+) (*http.Request, []byte, error) {
+	// 能力维度 body sanitize：与 Anthropic 平台 passthrough 相同，按 beta
+	// header 决定是否保留 body 中的 beta 能力字段，避免客户端"body 带字段但
+	// header 忘带 token"的 bug 让第三方上游 400。
+	clientBeta := ""
+	if c != nil && c.Request != nil {
+		clientBeta = getHeaderRaw(c.Request.Header, "anthropic-beta")
+	}
+	if beta, ok := account.HeaderOverrideValue("anthropic-beta"); ok {
+		clientBeta = beta
+	}
+	if sanitized, changed := sanitizeAnthropicBodyForBetaTokens(body, clientBeta); changed {
+		body = sanitized
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, targetURL, bytes.NewReader(body))
+	if err != nil {
+		return nil, nil, err
+	}
+
+	if c != nil && c.Request != nil {
+		for key, values := range c.Request.Header {
+			lowerKey := strings.ToLower(strings.TrimSpace(key))
+			if !allowedHeaders[lowerKey] {
+				continue
+			}
+			wireKey := resolveWireCasing(key)
+			for _, v := range values {
+				addHeaderRaw(req.Header, wireKey, v)
+			}
+		}
+	}
+
+	// 覆盖入站鉴权残留，注入上游认证（默认 x-api-key；可经 extra
+	// anthropic_apikey_auth_scheme 切换 Authorization: Bearer）。
+	req.Header.Del("authorization")
+	req.Header.Del("x-api-key")
+	req.Header.Del("x-goog-api-key")
+	req.Header.Del("cookie")
+	setAnthropicAPIKeyAuthHeader(req.Header, account, apiKey)
+
+	if getHeaderRaw(req.Header, "content-type") == "" {
+		setHeaderRaw(req.Header, "content-type", "application/json")
+	}
+	if getHeaderRaw(req.Header, "anthropic-version") == "" {
+		setHeaderRaw(req.Header, "anthropic-version", "2023-06-01")
+	}
+
+	// 账号级请求头覆写（最终生效，覆盖上面所有来源的同名头）
+	account.ApplyHeaderOverrides(req.Header)
+
+	return req, body, nil
+}
+
+// handleNativeAnthropicBufferedResponse 处理非流式原生 Anthropic 响应：
+// 校验 JSON、解析 usage、透传响应头后原样回写（仅工具名反向还原）。
+func (s *OpenAIGatewayService) handleNativeAnthropicBufferedResponse(
+	ctx context.Context,
+	resp *http.Response,
+	c *gin.Context,
+	account *Account,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	startTime time.Time,
+) (*OpenAIForwardResult, error) {
+	if s.rateLimitService != nil {
+		s.rateLimitService.UpdateSessionWindow(ctx, account, resp.Header)
+	}
+
+	body, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, anthropicTooLargeError)
+	if err != nil {
+		return nil, err
+	}
+	observer := upstreamResponseModelObserverFromContext(c)
+	if observer == nil {
+		observer = beginUpstreamResponseModelObservation(c)
+	}
+	observer.ObserveAnthropic(body)
+
+	var raw json.RawMessage
+	if err := json.Unmarshal(body, &raw); err != nil {
+		return nil, invalidNonStreamingJSONFailoverError(ctx, s.rateLimitService, resp, account, body, err, billingModel)
+	}
+
+	usage := parseClaudeUsageFromResponseBody(body)
+	if IsForceCacheBilling(ctx) && usage.InputTokens > 0 {
+		body, err = classifyAnthropicResponseInputAsCacheRead(body, usage)
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	writeAnthropicPassthroughResponseHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	contentType := strings.TrimSpace(resp.Header.Get("Content-Type"))
+	if contentType == "" {
+		contentType = "application/json"
+	}
+	body = reverseToolNamesIfPresent(c, body)
+	c.Data(resp.StatusCode, contentType, body)
+
+	return &OpenAIForwardResult{
+		RequestID:        resp.Header.Get("x-request-id"),
+		Usage:            claudeUsageToOpenAIUsage(usage),
+		Model:            originalModel,
+		BillingModel:     billingModel,
+		UpstreamModel:    upstreamModel,
+		UpstreamEndpoint: "/v1/messages",
+		Stream:           false,
+		Duration:         time.Since(startTime),
+	}, nil
+}
+
+// handleNativeAnthropicStreamingResponse 处理流式原生 Anthropic 响应：
+// 字节级 SSE 中继（逐行透传、按事件边界 flush），同时解析 usage。
+// 骨架与 handleStreamingResponseAnthropicAPIKeyPassthrough 一致。
+func (s *OpenAIGatewayService) handleNativeAnthropicStreamingResponse(
+	ctx context.Context,
+	resp *http.Response,
+	c *gin.Context,
+	account *Account,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	startTime time.Time,
+) (*OpenAIForwardResult, error) {
+	observer := upstreamResponseModelObserverFromContext(c)
+	if observer == nil {
+		observer = beginUpstreamResponseModelObservation(c)
+	}
+	if s.rateLimitService != nil {
+		s.rateLimitService.UpdateSessionWindow(ctx, account, resp.Header)
+	}
+
+	writeAnthropicPassthroughResponseHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+
+	contentType := strings.TrimSpace(resp.Header.Get("Content-Type"))
+	if contentType == "" {
+		contentType = "text/event-stream"
+	}
+	c.Header("Content-Type", contentType)
+	if c.Writer.Header().Get("Cache-Control") == "" {
+		c.Header("Cache-Control", "no-cache")
+	}
+	if c.Writer.Header().Get("Connection") == "" {
+		c.Header("Connection", "keep-alive")
+	}
+	c.Header("X-Accel-Buffering", "no")
+	if v := resp.Header.Get("x-request-id"); v != "" {
+		c.Header("x-request-id", v)
+	}
+
+	w := c.Writer
+	flusher, ok := w.(http.Flusher)
+	if !ok {
+		return nil, errors.New("streaming not supported")
+	}
+
+	usage := &ClaudeUsage{}
+	var firstTokenMs *int
+	clientDisconnected := false
+	sawTerminalEvent := false
+
+	scanner := bufio.NewScanner(resp.Body)
+	maxLineSize := defaultMaxLineSize
+	if s.cfg != nil && s.cfg.Gateway.MaxLineSize > 0 {
+		maxLineSize = s.cfg.Gateway.MaxLineSize
+	}
+	scanBuf := getSSEScannerBuf64K()
+	scanner.Buffer(scanBuf[:0], maxLineSize)
+
+	type scanEvent struct {
+		line string
+		err  error
+	}
+	events := make(chan scanEvent, 16)
+	done := make(chan struct{})
+	sendEvent := func(ev scanEvent) bool {
+		select {
+		case events <- ev:
+			return true
+		case <-done:
+			return false
+		}
+	}
+	var lastReadAt int64
+	atomic.StoreInt64(&lastReadAt, time.Now().UnixNano())
+	go func(scanBuf *sseScannerBuf64K) {
+		defer putSSEScannerBuf64K(scanBuf)
+		defer close(events)
+		for scanner.Scan() {
+			atomic.StoreInt64(&lastReadAt, time.Now().UnixNano())
+			if !sendEvent(scanEvent{line: scanner.Text()}) {
+				return
+			}
+		}
+		if err := scanner.Err(); err != nil {
+			_ = sendEvent(scanEvent{err: err})
+		}
+	}(scanBuf)
+	defer close(done)
+
+	streamInterval := time.Duration(0)
+	if s.cfg != nil && s.cfg.Gateway.StreamDataIntervalTimeout > 0 {
+		streamInterval = time.Duration(s.cfg.Gateway.StreamDataIntervalTimeout) * time.Second
+	}
+	var intervalTicker *time.Ticker
+	if streamInterval > 0 {
+		intervalTicker = time.NewTicker(streamInterval)
+		defer intervalTicker.Stop()
+	}
+	var intervalCh <-chan time.Time
+	if intervalTicker != nil {
+		intervalCh = intervalTicker.C
+	}
+
+	keepaliveInterval := time.Duration(0)
+	if s.cfg != nil && s.cfg.Gateway.StreamKeepaliveInterval > 0 {
+		keepaliveInterval = time.Duration(s.cfg.Gateway.StreamKeepaliveInterval) * time.Second
+	}
+	var keepaliveTimer *time.Timer
+	if keepaliveInterval > 0 {
+		keepaliveTimer = time.NewTimer(keepaliveInterval)
+		defer keepaliveTimer.Stop()
+	}
+	var keepaliveCh <-chan time.Time
+	if keepaliveTimer != nil {
+		keepaliveCh = keepaliveTimer.C
+	}
+	lastDataAt := time.Now()
+	resetKeepaliveTimer := func() {
+		if keepaliveTimer == nil {
+			return
+		}
+		if !keepaliveTimer.Stop() {
+			select {
+			case <-keepaliveTimer.C:
+			default:
+			}
+		}
+		keepaliveTimer.Reset(keepaliveInterval)
+	}
+	inPartialEvent := false
+
+	for {
+		select {
+		case ev, ok := <-events:
+			if !ok {
+				if !clientDisconnected {
+					flusher.Flush()
+				}
+				if !sawTerminalEvent {
+					return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime),
+						fmt.Errorf("stream usage incomplete: missing terminal event")
+				}
+				return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime), nil
+			}
+			if ev.err != nil {
+				if sawTerminalEvent {
+					return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime), nil
+				}
+				if clientDisconnected {
+					return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime),
+						fmt.Errorf("stream usage incomplete after disconnect: %w", ev.err)
+				}
+				if errors.Is(ev.err, context.Canceled) || errors.Is(ev.err, context.DeadlineExceeded) {
+					return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime),
+						fmt.Errorf("stream usage incomplete: %w", ev.err)
+				}
+				if errors.Is(ev.err, bufio.ErrTooLong) {
+					logger.LegacyPrintf("service.gateway", "[CN Anthropic 直通] SSE line too long: account=%d max_size=%d error=%v", account.ID, maxLineSize, ev.err)
+					return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime), ev.err
+				}
+				return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime),
+					fmt.Errorf("stream read error: %w", ev.err)
+			}
+
+			line := ev.line
+			if data, ok := extractAnthropicSSEDataLine(line); ok {
+				trimmed := strings.TrimSpace(data)
+				observer.ObserveAnthropic([]byte(trimmed))
+				if anthropicStreamEventIsTerminal("", trimmed) {
+					sawTerminalEvent = true
+				}
+				if firstTokenMs == nil && trimmed != "" && trimmed != "[DONE]" {
+					ms := int(time.Since(startTime).Milliseconds())
+					firstTokenMs = &ms
+				}
+				parseSSEUsagePassthrough(data, usage)
+			} else {
+				trimmed := strings.TrimSpace(line)
+				if strings.HasPrefix(trimmed, "event:") && anthropicStreamEventIsTerminal(strings.TrimSpace(strings.TrimPrefix(trimmed, "event:")), "") {
+					sawTerminalEvent = true
+				}
+			}
+
+			if !clientDisconnected {
+				restored := string(reverseToolNamesIfPresent(c, []byte(line)))
+				if _, err := io.WriteString(w, restored); err != nil {
+					clientDisconnected = true
+					logger.LegacyPrintf("service.gateway", "[CN Anthropic 直通] Client disconnected during streaming, continue draining upstream for usage: account=%d", account.ID)
+				} else if _, err := io.WriteString(w, "\n"); err != nil {
+					clientDisconnected = true
+					logger.LegacyPrintf("service.gateway", "[CN Anthropic 直通] Client disconnected during streaming, continue draining upstream for usage: account=%d", account.ID)
+				} else if line == "" {
+					// 按 SSE 事件边界刷出，减少每行 flush 带来的 syscall 开销。
+					flusher.Flush()
+					lastDataAt = time.Now()
+					resetKeepaliveTimer()
+					inPartialEvent = false
+				} else {
+					inPartialEvent = true
+				}
+			}
+
+		case <-intervalCh:
+			lastRead := time.Unix(0, atomic.LoadInt64(&lastReadAt))
+			if time.Since(lastRead) < streamInterval {
+				continue
+			}
+			if clientDisconnected {
+				return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime),
+					fmt.Errorf("stream usage incomplete after timeout")
+			}
+			logger.LegacyPrintf("service.gateway", "[CN Anthropic 直通] Stream data interval timeout: account=%d model=%s interval=%s", account.ID, upstreamModel, streamInterval)
+			if s.rateLimitService != nil {
+				s.rateLimitService.HandleStreamTimeout(ctx, account, upstreamModel)
+			}
+			return s.nativeAnthropicStreamResult(c, resp, usage, firstTokenMs, clientDisconnected, originalModel, billingModel, upstreamModel, startTime),
+				fmt.Errorf("stream data interval timeout")
+
+		case <-keepaliveCh:
+			if clientDisconnected {
+				continue
+			}
+			if inPartialEvent {
+				resetKeepaliveTimer()
+				continue
+			}
+			if time.Since(lastDataAt) < keepaliveInterval {
+				resetKeepaliveTimer()
+				continue
+			}
+			if _, err := fmt.Fprint(w, "event: ping\ndata: {\"type\": \"ping\"}\n\n"); err != nil {
+				clientDisconnected = true
+				logger.LegacyPrintf("service.gateway", "[CN Anthropic 直通] Client disconnected during keepalive ping, continue draining upstream for usage: account=%d", account.ID)
+				continue
+			}
+			flusher.Flush()
+			lastDataAt = time.Now()
+			resetKeepaliveTimer()
+		}
+	}
+}
+
+// nativeAnthropicStreamResult 组装流式直通结果；流中断时同样返回已观测到的
+// usage 与错误一起带出，避免上游已计量的请求漏记漏计费（对齐 issue #5148 语义）。
+func (s *OpenAIGatewayService) nativeAnthropicStreamResult(
+	c *gin.Context,
+	resp *http.Response,
+	usage *ClaudeUsage,
+	firstTokenMs *int,
+	clientDisconnect bool,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	startTime time.Time,
+) *OpenAIForwardResult {
+	if usage == nil {
+		usage = &ClaudeUsage{}
+	}
+	return &OpenAIForwardResult{
+		RequestID:        resp.Header.Get("x-request-id"),
+		Usage:            claudeUsageToOpenAIUsage(usage),
+		Model:            originalModel,
+		BillingModel:     billingModel,
+		UpstreamModel:    upstreamModel,
+		UpstreamEndpoint: "/v1/messages",
+		Stream:           true,
+		Duration:         time.Since(startTime),
+		FirstTokenMs:     firstTokenMs,
+		ClientDisconnect: clientDisconnect,
+	}
+}
+
+// forwardCountTokensViaNativeAnthropic 把 Anthropic count_tokens 请求透传到
+// 国产供应商原生 Anthropic 端点（{base}/v1/messages/count_tokens），
+// 仅做模型名映射，不做协议转换。
+func (s *OpenAIGatewayService) forwardCountTokensViaNativeAnthropic(
+	ctx context.Context,
+	c *gin.Context,
+	account *Account,
+	body []byte,
+	defaultMappedModel string,
+) error {
+	originalModel := strings.TrimSpace(gjson.GetBytes(body, "model").String())
+	if originalModel == "" {
+		writeAnthropicCountTokensError(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return fmt.Errorf("count_tokens: missing model in request")
+	}
+	billingModel := resolveOpenAIForwardModel(account, originalModel, strings.TrimSpace(defaultMappedModel))
+	upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
+	if upstreamModel != originalModel {
+		rewritten, err := sjson.SetBytes(body, "model", upstreamModel)
+		if err != nil {
+			return fmt.Errorf("count_tokens: rewrite model: %w", err)
+		}
+		body = rewritten
+	}
+
+	apiKey := strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
+	if apiKey == "" {
+		writeAnthropicCountTokensError(c, http.StatusBadGateway, "upstream_error", "Account api_key is missing")
+		return fmt.Errorf("count_tokens: account %d missing api_key", account.ID)
+	}
+	targetURL, err := s.nativeAnthropicTargetURL(account)
+	if err != nil {
+		return fmt.Errorf("count_tokens: %w", err)
+	}
+	targetURL = strings.TrimSuffix(targetURL, "/v1/messages") + "/v1/messages/count_tokens"
+
+	upstreamReq, err := http.NewRequestWithContext(ctx, http.MethodPost, targetURL, bytes.NewReader(body))
+	if err != nil {
+		writeAnthropicCountTokensError(c, http.StatusInternalServerError, "api_error", "Failed to build request")
+		return fmt.Errorf("count_tokens: build request: %w", err)
+	}
+	reqHeader := upstreamReq.Header
+	reqHeader.Del("authorization")
+	reqHeader.Del("x-api-key")
+	setAnthropicAPIKeyAuthHeader(reqHeader, account, apiKey)
+	reqHeader.Set("content-type", "application/json")
+	reqHeader.Set("accept", "application/json")
+	account.ApplyHeaderOverrides(reqHeader)
+
+	proxyURL := ""
+	if account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+	if err != nil {
+		safeErr := sanitizeUpstreamErrorMessage(err.Error())
+		setOpsUpstreamError(c, 0, safeErr, "")
+		writeAnthropicCountTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream request failed")
+		return fmt.Errorf("count_tokens: upstream request failed: %s", safeErr)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	// count_tokens 响应体极小；与其他探测路径一致加 256KB 上限防异常上游放大内存。
+	respBody, err := io.ReadAll(io.LimitReader(resp.Body, cnQuotaMaxBodyBytes))
+	if err != nil {
+		writeAnthropicCountTokensError(c, http.StatusBadGateway, "upstream_error", "Failed to read response")
+		return fmt.Errorf("count_tokens: read response: %w", err)
+	}
+	if resp.StatusCode >= 400 {
+		if s.rateLimitService != nil {
+			s.rateLimitService.HandleUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)
+		}
+		upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(respBody)))
+		setOpsUpstreamError(c, resp.StatusCode, upstreamMsg, "")
+		writeAnthropicCountTokensError(c, resp.StatusCode, "upstream_error", "Upstream request failed")
+		return fmt.Errorf("count_tokens: upstream error: %d", resp.StatusCode)
+	}
+
+	inputTokens := gjson.GetBytes(respBody, "input_tokens")
+	if !inputTokens.Exists() {
+		writeAnthropicCountTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream response missing input_tokens")
+		return fmt.Errorf("count_tokens: response missing input_tokens field")
+	}
+	c.JSON(http.StatusOK, gin.H{
+		"input_tokens": int(inputTokens.Int()),
+	})
+	return nil
+}
+
+// claudeUsageToOpenAIUsage 把 Anthropic 格式 usage 映射到 OpenAI 网关统一的
+// 用量结构（字段一一对应）。
+func claudeUsageToOpenAIUsage(u *ClaudeUsage) OpenAIUsage {
+	if u == nil {
+		return OpenAIUsage{}
+	}
+	return OpenAIUsage{
+		InputTokens:              u.InputTokens,
+		OutputTokens:             u.OutputTokens,
+		CacheCreationInputTokens: u.CacheCreationInputTokens,
+		CacheReadInputTokens:     u.CacheReadInputTokens,
+	}
+}
diff --git a/backend/internal/service/openai_gateway_model_availability.go b/backend/internal/service/openai_gateway_model_availability.go
index a665052e602e..35adea0d723d 100644
--- a/backend/internal/service/openai_gateway_model_availability.go
+++ b/backend/internal/service/openai_gateway_model_availability.go
@@ -34,7 +34,7 @@ func (s *OpenAIGatewayService) DiagnoseModelAvailabilityForPlatform(
 		return ModelAvailabilityDiagnosis{HasAccountsInPool: true, HasModelSupport: true}
 	}
 
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 	queryGroupID := groupID
 	includeGrouped := false
 	if s.cfg != nil && s.cfg.RunMode == config.RunModeSimple {
diff --git a/backend/internal/service/openai_gateway_passthrough.go b/backend/internal/service/openai_gateway_passthrough.go
index 1fac0440f685..291f8bcc9e93 100644
--- a/backend/internal/service/openai_gateway_passthrough.go
+++ b/backend/internal/service/openai_gateway_passthrough.go
@@ -380,11 +380,14 @@ func (s *OpenAIGatewayService) buildUpstreamRequestOpenAIPassthrough(
 			if err != nil {
 				return nil, err
 			}
-			targetURL = buildOpenAIResponsesURL(validatedURL)
+			targetURL = buildOpenAIResponsesURLForPlatform(account.Platform, validatedURL)
 		}
 	}
 	targetURL = appendOpenAIResponsesRequestPathSuffix(targetURL, openAIResponsesRequestPathSuffix(c))
 
+	// DeepSeek 原生 Responses 端点为无状态实现（见 normalizeDeepSeekResponsesRequestBody）。
+	body = normalizeDeepSeekResponsesRequestBody(account, body)
+
 	req, err := http.NewRequestWithContext(ctx, http.MethodPost, targetURL, bytes.NewReader(body))
 	if err != nil {
 		return nil, err
diff --git a/backend/internal/service/openai_gateway_request_body.go b/backend/internal/service/openai_gateway_request_body.go
index 8be2d0cf2b70..3559fb400d4d 100644
--- a/backend/internal/service/openai_gateway_request_body.go
+++ b/backend/internal/service/openai_gateway_request_body.go
@@ -45,6 +45,33 @@ func buildOpenAIResponsesURL(base string) string {
 	return buildOpenAIEndpointURL(base, "/v1/responses")
 }
 
+// buildOpenAIResponsesURLForPlatform 组装 Responses 端点（平台感知）。
+// DeepSeek 官方 Responses 端点为 /responses（无 /v1 前缀，适配 Codex）；
+// 其余平台维持 /v1/responses。
+func buildOpenAIResponsesURLForPlatform(platform string, base string) string {
+	if platform == PlatformDeepseek {
+		return buildOpenAIEndpointURL(base, "/responses")
+	}
+	return buildOpenAIResponsesURL(base)
+}
+
+// normalizeDeepSeekResponsesRequestBody 适配 DeepSeek 无状态 Responses 端点：
+// 强制 store=false 并清除 previous_response_id（官方 /responses 不支持服务端
+// 状态存储，携带这些字段会被拒绝）。非 deepseek responses 协议账号原样返回。
+func normalizeDeepSeekResponsesRequestBody(account *Account, body []byte) []byte {
+	if account == nil || account.Platform != PlatformDeepseek || account.GetAPIProtocol() != APIProtocolResponses {
+		return body
+	}
+	normalized, err := sjson.SetBytes(body, "store", false)
+	if err != nil {
+		return body
+	}
+	if stripped, err := sjson.DeleteBytes(normalized, "previous_response_id"); err == nil {
+		normalized = stripped
+	}
+	return normalized
+}
+
 func trimOpenAIEncryptedReasoningItems(reqBody map[string]any) bool {
 	if len(reqBody) == 0 {
 		return false
diff --git a/backend/internal/service/openai_gateway_responses_anthropic_native.go b/backend/internal/service/openai_gateway_responses_anthropic_native.go
new file mode 100644
index 000000000000..7cb4a13d2d49
--- /dev/null
+++ b/backend/internal/service/openai_gateway_responses_anthropic_native.go
@@ -0,0 +1,414 @@
+package service
+
+// 国产供应商 Anthropic 协议账号的 Responses 入站反向路径。
+//
+// 客户端说 OpenAI Responses（/v1/responses，Codex 等）、上游是供应商原生
+// Anthropic 端点（api_protocol=anthropic）时的交叉组合：请求 Responses→Anthropic
+// 单次转换，响应 Anthropic 事件→Responses 事件转换。转换链与 Anthropic 平台的
+// gateway_forward_as_responses.go 完全一致（复用同一组 apicompat 状态机），仅上游
+// 发送/错误处理对齐 OpenAI 网关语义（模型映射、failover、transport error）。
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"go.uber.org/zap"
+)
+
+// forwardResponsesViaNativeAnthropic serves OpenAI /v1/responses clients through
+// a CN provider's native Anthropic endpoint.
+//
+// Conversion chain:
+//
+//	Request:  Responses → Anthropic (single conversion)
+//	Response: Anthropic events → Responses events (stream state machine)
+func (s *OpenAIGatewayService) forwardResponsesViaNativeAnthropic(
+	ctx context.Context,
+	c *gin.Context,
+	account *Account,
+	body []byte,
+	defaultMappedModel string,
+) (*OpenAIForwardResult, error) {
+	startTime := time.Now()
+
+	// 1. Lower Codex client-side tools to function tools understood by Anthropic.
+	adaptedBody, clientToolMapping, err := adaptResponsesClientToolsForAnthropic(body)
+	if err != nil {
+		writeResponsesError(c, http.StatusBadRequest, "invalid_request_error", "Failed to adapt request tools")
+		return nil, fmt.Errorf("adapt responses client tools: %w", err)
+	}
+
+	// 2. Parse Responses request
+	var responsesReq apicompat.ResponsesRequest
+	if err := json.Unmarshal(adaptedBody, &responsesReq); err != nil {
+		writeResponsesError(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
+		return nil, fmt.Errorf("parse responses request: %w", err)
+	}
+	originalModel := responsesReq.Model
+	if strings.TrimSpace(originalModel) == "" {
+		writeResponsesError(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return nil, fmt.Errorf("missing model in request")
+	}
+	clientStream := responsesReq.Stream
+
+	// 3. Convert Responses → Anthropic
+	anthropicReq, err := apicompat.ResponsesToAnthropicRequest(&responsesReq)
+	if err != nil {
+		writeResponsesError(c, http.StatusBadRequest, "invalid_request_error", "Failed to convert request")
+		return nil, fmt.Errorf("convert responses to anthropic: %w", err)
+	}
+
+	// 4. Model mapping（OpenAI 网关统一入口的映射语义）
+	billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel)
+	upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
+	anthropicReq.Model = upstreamModel
+
+	reasoningEffort := ExtractResponsesReasoningEffortFromBody(body)
+	reasoningEffort = ApplyThinkingEnabledFallback(reasoningEffort, body, billingModel)
+
+	// 5. Force upstream streaming（客户端原始终决定响应格式；
+	// 上游恒为流式，非流式由缓冲路径组装）。
+	anthropicReq.Stream = true
+	reqStream := true
+
+	logger.L().Debug("openai responses: forwarding via native anthropic endpoint",
+		zap.Int64("account_id", account.ID),
+		zap.String("original_model", originalModel),
+		zap.String("billing_model", billingModel),
+		zap.String("upstream_model", upstreamModel),
+		zap.Bool("client_stream", clientStream),
+	)
+
+	anthropicBody, err := json.Marshal(anthropicReq)
+	if err != nil {
+		return nil, fmt.Errorf("marshal anthropic request: %w", err)
+	}
+
+	// 与 /v1/messages 直通路径相同的 pre-filter。
+	anthropicBody = StripEmptyTextBlocks(anthropicBody)
+	anthropicBody = FilterWebSearchHistoryBlocks(anthropicBody, upstreamModel)
+	anthropicBody = enforceCacheControlLimit(anthropicBody)
+
+	apiKey := strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
+	if apiKey == "" {
+		return nil, fmt.Errorf("account %d missing api_key", account.ID)
+	}
+	targetURL, err := s.nativeAnthropicTargetURL(account)
+	if err != nil {
+		return nil, err
+	}
+
+	proxyURL := ""
+	if account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+
+	upstreamCtx, releaseUpstreamCtx := detachStreamUpstreamContext(ctx, reqStream)
+	upstreamReq, _, err := s.buildNativeAnthropicUpstreamRequest(upstreamCtx, c, account, anthropicBody, apiKey, targetURL)
+	releaseUpstreamCtx()
+	if err != nil {
+		return nil, fmt.Errorf("build upstream request: %w", err)
+	}
+
+	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+	if err != nil {
+		return nil, s.handleOpenAIUpstreamTransportError(ctx, c, account, err, true)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	if resp.StatusCode >= 400 {
+		respBody, upstreamMsg := s.readOpenAIUpstreamError(resp)
+		if foErr := s.failoverOpenAIUpstreamHTTPError(ctx, c, account, resp, respBody, upstreamMsg, upstreamModel); foErr != nil {
+			return nil, foErr
+		}
+		writeResponsesError(c, mapUpstreamStatusCode(resp.StatusCode), "server_error", upstreamMsg)
+		return nil, fmt.Errorf("upstream error: %d %s", resp.StatusCode, upstreamMsg)
+	}
+
+	if clientStream {
+		return s.handleResponsesStreamingFromNativeAnthropic(resp, c, originalModel, billingModel, upstreamModel, reasoningEffort, startTime, clientToolMapping)
+	}
+	return s.handleResponsesBufferedFromNativeAnthropic(resp, c, originalModel, billingModel, upstreamModel, reasoningEffort, startTime, clientToolMapping)
+}
+
+// handleResponsesBufferedFromNativeAnthropic reads Anthropic SSE events, assembles
+// the full response, then converts Anthropic → Responses.
+func (s *OpenAIGatewayService) handleResponsesBufferedFromNativeAnthropic(
+	resp *http.Response,
+	c *gin.Context,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	reasoningEffort *string,
+	startTime time.Time,
+	clientToolMapping apicompat.ResponsesClientToolMapping,
+) (*OpenAIForwardResult, error) {
+	requestID := resp.Header.Get("x-request-id")
+
+	scanner := bufio.NewScanner(resp.Body)
+	maxLineSize := defaultMaxLineSize
+	if s.cfg != nil && s.cfg.Gateway.MaxLineSize > 0 {
+		maxLineSize = s.cfg.Gateway.MaxLineSize
+	}
+	scanner.Buffer(make([]byte, 0, 64*1024), maxLineSize)
+
+	var finalResp *apicompat.AnthropicResponse
+	var usage ClaudeUsage
+
+	for scanner.Scan() {
+		line := scanner.Text()
+		// SSE 规范允许 `event:xxx`（冒号后无空格）：Kimi 等上游返回紧凑格式。
+		if _, ok := extractOpenAISSEEventLine(line); !ok {
+			continue
+		}
+
+		if !scanner.Scan() {
+			break
+		}
+		payload, ok := extractOpenAISSEDataLine(scanner.Text())
+		if !ok {
+			continue
+		}
+
+		var event apicompat.AnthropicStreamEvent
+		if err := json.Unmarshal([]byte(payload), &event); err != nil {
+			continue
+		}
+
+		if event.Type == "message_start" && event.Message != nil {
+			finalResp = event.Message
+			mergeAnthropicUsage(&usage, event.Message.Usage)
+		}
+		if event.Type == "message_delta" {
+			if event.Usage != nil {
+				mergeAnthropicUsage(&usage, *event.Usage)
+			}
+			if event.Delta != nil && event.Delta.StopReason != "" && finalResp != nil {
+				finalResp.StopReason = apicompat.AnthropicStopReasonPtr(event.Delta.StopReason)
+			}
+		}
+		if event.Type == "content_block_start" && event.ContentBlock != nil && finalResp != nil {
+			finalResp.Content = append(finalResp.Content, *event.ContentBlock)
+		}
+		if event.Type == "content_block_delta" && event.Delta != nil && finalResp != nil && event.Index != nil {
+			idx := *event.Index
+			if idx < len(finalResp.Content) {
+				switch event.Delta.Type {
+				case "text_delta":
+					finalResp.Content[idx].Text += event.Delta.Text
+				case "thinking_delta":
+					finalResp.Content[idx].Thinking += event.Delta.Thinking
+				case "input_json_delta":
+					finalResp.Content[idx].Input = appendRawJSON(finalResp.Content[idx].Input, event.Delta.PartialJSON)
+				}
+			}
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai responses via native anthropic buffered: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+
+	if finalResp == nil {
+		writeResponsesError(c, http.StatusBadGateway, "server_error", "Upstream stream ended without a response")
+		return nil, fmt.Errorf("upstream stream ended without response")
+	}
+
+	if usage.InputTokens > 0 || usage.OutputTokens > 0 {
+		finalResp.Usage = apicompat.AnthropicUsage{
+			InputTokens:              usage.InputTokens,
+			OutputTokens:             usage.OutputTokens,
+			CacheCreationInputTokens: usage.CacheCreationInputTokens,
+			CacheReadInputTokens:     usage.CacheReadInputTokens,
+		}
+	}
+
+	responsesResp := apicompat.AnthropicToResponsesResponse(finalResp)
+	responsesResp.Model = originalModel
+
+	if s.responseHeaderFilter != nil {
+		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	}
+	// 非流式响应必须是 application/json（上游被强制流式，透传头会污染）。
+	c.Writer.Header().Set("Content-Type", "application/json; charset=utf-8")
+	if respBytes, err := json.Marshal(responsesResp); err == nil {
+		respBytes = reverseToolNamesIfPresent(c, respBytes)
+		respBytes, _, err = apicompat.RestoreResponsesClientToolPayload(respBytes, clientToolMapping)
+		if err != nil {
+			return nil, fmt.Errorf("restore responses client tools: %w", err)
+		}
+		c.Data(http.StatusOK, "application/json; charset=utf-8", respBytes)
+	} else {
+		c.JSON(http.StatusOK, responsesResp)
+	}
+
+	return &OpenAIForwardResult{
+		RequestID:        requestID,
+		Usage:            claudeUsageToOpenAIUsage(&usage),
+		Model:            originalModel,
+		BillingModel:     billingModel,
+		UpstreamModel:    upstreamModel,
+		UpstreamEndpoint: "/v1/messages",
+		ReasoningEffort:  reasoningEffort,
+		Stream:           false,
+		Duration:         time.Since(startTime),
+	}, nil
+}
+
+// handleResponsesStreamingFromNativeAnthropic reads Anthropic SSE events, converts
+// each to Responses SSE events, and writes them to the client.
+func (s *OpenAIGatewayService) handleResponsesStreamingFromNativeAnthropic(
+	resp *http.Response,
+	c *gin.Context,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	reasoningEffort *string,
+	startTime time.Time,
+	clientToolMapping apicompat.ResponsesClientToolMapping,
+) (*OpenAIForwardResult, error) {
+	requestID := resp.Header.Get("x-request-id")
+
+	if s.responseHeaderFilter != nil {
+		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	}
+	c.Writer.Header().Set("Content-Type", "text/event-stream")
+	c.Writer.Header().Set("Cache-Control", "no-cache")
+	c.Writer.Header().Set("Connection", "keep-alive")
+	c.Writer.Header().Set("X-Accel-Buffering", "no")
+	c.Writer.WriteHeader(http.StatusOK)
+
+	state := apicompat.NewAnthropicEventToResponsesState()
+	state.Model = originalModel
+	clientToolRestorer := apicompat.NewResponsesClientToolStreamRestorer(clientToolMapping)
+
+	var usage ClaudeUsage
+	var firstTokenMs *int
+	firstChunk := true
+	clientDisconnected := false
+
+	scanner := bufio.NewScanner(resp.Body)
+	maxLineSize := defaultMaxLineSize
+	if s.cfg != nil && s.cfg.Gateway.MaxLineSize > 0 {
+		maxLineSize = s.cfg.Gateway.MaxLineSize
+	}
+	scanner.Buffer(make([]byte, 0, 64*1024), maxLineSize)
+
+	resultWithUsage := func() *OpenAIForwardResult {
+		return &OpenAIForwardResult{
+			RequestID:        requestID,
+			Usage:            claudeUsageToOpenAIUsage(&usage),
+			Model:            originalModel,
+			BillingModel:     billingModel,
+			UpstreamModel:    upstreamModel,
+			UpstreamEndpoint: "/v1/messages",
+			ReasoningEffort:  reasoningEffort,
+			Stream:           true,
+			Duration:         time.Since(startTime),
+			FirstTokenMs:     firstTokenMs,
+			ClientDisconnect: clientDisconnected,
+		}
+	}
+
+	processAnthropicEvent := func(event *apicompat.AnthropicStreamEvent) bool {
+		if firstChunk {
+			firstChunk = false
+			ms := int(time.Since(startTime).Milliseconds())
+			firstTokenMs = &ms
+		}
+
+		if event.Type == "message_delta" && event.Usage != nil {
+			mergeAnthropicUsage(&usage, *event.Usage)
+		}
+		if event.Type == "message_start" && event.Message != nil {
+			mergeAnthropicUsage(&usage, event.Message.Usage)
+		}
+
+		events := apicompat.AnthropicEventToResponsesEvents(event, state)
+		for _, evt := range events {
+			payload, err := json.Marshal(evt)
+			if err != nil {
+				continue
+			}
+			payload = reverseToolNamesIfPresent(c, payload)
+			payloads, _, err := clientToolRestorer.RestoreEvent(payload)
+			if err != nil {
+				continue
+			}
+			for _, restored := range payloads {
+				eventType := gjson.GetBytes(restored, "type").String()
+				if _, err := fmt.Fprintf(c.Writer, "event: %s\ndata: %s\n\n", eventType, restored); err != nil {
+					clientDisconnected = true
+					return true
+				}
+			}
+		}
+		if len(events) > 0 {
+			c.Writer.Flush()
+		}
+		return false
+	}
+
+	for scanner.Scan() {
+		line := scanner.Text()
+		if _, ok := extractOpenAISSEEventLine(line); !ok {
+			continue
+		}
+
+		if !scanner.Scan() {
+			break
+		}
+		payload, ok := extractOpenAISSEDataLine(scanner.Text())
+		if !ok {
+			continue
+		}
+
+		var event apicompat.AnthropicStreamEvent
+		if err := json.Unmarshal([]byte(payload), &event); err != nil {
+			continue
+		}
+
+		if processAnthropicEvent(&event) {
+			return resultWithUsage(), nil
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai responses via native anthropic stream: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+
+	// Finalize state machine（客户端已断开时仍执行，保证 usage 汇总完整）。
+	if finalEvents := apicompat.FinalizeAnthropicResponsesStream(state); len(finalEvents) > 0 {
+		for _, evt := range finalEvents {
+			sse, err := apicompat.ResponsesEventToSSE(evt)
+			if err != nil {
+				continue
+			}
+			fmt.Fprint(c.Writer, sse) //nolint:errcheck
+		}
+		c.Writer.Flush()
+	}
+
+	return resultWithUsage(), nil
+}
diff --git a/backend/internal/service/openai_gateway_scheduling.go b/backend/internal/service/openai_gateway_scheduling.go
index f7dff0909f62..7dd541472476 100644
--- a/backend/internal/service/openai_gateway_scheduling.go
+++ b/backend/internal/service/openai_gateway_scheduling.go
@@ -243,15 +243,22 @@ func (s *OpenAIGatewayService) SelectAccountForModelWithExclusions(ctx context.C
 	return s.selectAccountForModelWithExclusions(s.withOpenAIQuotaAutoPauseContext(ctx), groupID, PlatformOpenAI, sessionHash, requestedModel, excludedIDs, false, 0, "", false)
 }
 
-// noAvailableOpenAISelectionError builds the standard "no account available" error
-// while preserving the legacy /responses/compact error when applicable.
-func normalizeOpenAICompatiblePlatform(platform string) string {
-	if platform == PlatformGrok {
-		return PlatformGrok
+// NormalizeOpenAICompatiblePlatform 保留 grok 与国产 OpenAI 兼容供应商（kimi/zhipu/
+// deepseek）的原值，其他值一律归一为 openai。调度器据此对账号与请求做精确平台匹配：
+// kimi 分组请求只命中 kimi 账号，语义与 openai/grok 一致。
+// （upstream 曾将本函数改为未导出 normalizeOpenAICompatiblePlatform，本分支的
+// handler 调度入口仍需导出，保持导出名。）
+func NormalizeOpenAICompatiblePlatform(platform string) string {
+	switch platform {
+	case PlatformGrok, PlatformKimi, PlatformZhipu, PlatformDeepseek:
+		return platform
+	default:
+		return PlatformOpenAI
 	}
-	return PlatformOpenAI
 }
 
+// noAvailableOpenAISelectionError builds the standard "no account available" error
+// while preserving the legacy /responses/compact error when applicable.
 // details carries an optional machine-parseable exclusion summary (e.g.
 // "pool=2, filtered: quota_auto_pause_7d=1 runtime_blocked=1") appended in
 // parentheses. It is for server-side logs / ops diagnostics only: handlers
@@ -327,7 +334,7 @@ func isOpenAICompatibleAccountEligibleForRequest(ctx context.Context, account *A
 // ordinary scheduling gate. Legacy selection uses it before classifying the
 // profit veto so earlier failures retain their actual reason.
 func isOpenAICompatibleAccountEligibleForRequestBeforeProfit(ctx context.Context, account *Account, platform string, requestedModel string, requireCompact bool, requiredCapability OpenAIEndpointCapability) bool {
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 	if account == nil || account.Platform != platform || !account.IsOpenAICompatible() || !account.IsSchedulableForModelWithContext(ctx, requestedModel) {
 		return false
 	}
@@ -726,7 +733,7 @@ func resolveOpenAIAccountUpstreamModelForRequest(account *Account, requestedMode
 }
 
 func (s *OpenAIGatewayService) selectAccountForModelWithExclusions(ctx context.Context, groupID *int64, platform string, sessionHash string, requestedModel string, excludedIDs map[int64]struct{}, requireCompact bool, stickyAccountID int64, requiredCapability OpenAIEndpointCapability, preferLowUpstreamRate bool) (*Account, error) {
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 	if s.checkChannelPricingRestriction(ctx, groupID, requestedModel) {
 		slog.Warn("channel pricing restriction blocked request",
 			"group_id", derefGroupID(groupID),
@@ -779,7 +786,7 @@ func (s *OpenAIGatewayService) tryStickySessionHit(ctx context.Context, groupID
 	if sessionHash == "" {
 		return nil
 	}
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 
 	accountID := stickyAccountID
 	if accountID <= 0 {
@@ -846,7 +853,7 @@ func (s *OpenAIGatewayService) tryStickySessionHit(ctx context.Context, groupID
 // true); the third contains deterministic
 // exclusion diagnostics for the evaluated snapshot.
 func (s *OpenAIGatewayService) selectBestAccount(ctx context.Context, groupID *int64, platform string, accounts []Account, requestedModel string, excludedIDs map[int64]struct{}, requireCompact bool, requiredCapability OpenAIEndpointCapability, preferLowUpstreamRate bool) (*Account, bool, openAISelectionFilterStats) {
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 	compactBlocked := false
 	filterStats := openAISelectionFilterStats{pool: len(accounts)}
 	needsUpstreamCheck := s.needsUpstreamChannelRestrictionCheck(ctx, groupID)
@@ -958,7 +965,7 @@ func (s *OpenAIGatewayService) SelectAccountWithLoadAwareness(ctx context.Contex
 }
 
 func (s *OpenAIGatewayService) selectAccountWithLoadAwareness(ctx context.Context, groupID *int64, platform string, sessionHash string, requestedModel string, excludedIDs map[int64]struct{}, requireCompact bool, requiredCapability OpenAIEndpointCapability, useUpstreamTokenCost bool) (*AccountSelectionResult, error) {
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 	if s.checkChannelPricingRestriction(ctx, groupID, requestedModel) {
 		slog.Warn("channel pricing restriction blocked request",
 			"group_id", derefGroupID(groupID),
@@ -1304,7 +1311,7 @@ func (s *OpenAIGatewayService) selectAccountWithLoadAwareness(ctx context.Contex
 }
 
 func (s *OpenAIGatewayService) listSchedulableAccounts(ctx context.Context, groupID *int64, platform string) ([]Account, error) {
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 	if s.schedulerSnapshot != nil {
 		accounts, _, err := s.schedulerSnapshot.ListSchedulableAccounts(ctx, groupID, platform, false)
 		if err != nil {
@@ -1357,7 +1364,7 @@ func (s *OpenAIGatewayService) resolveFreshSchedulableOpenAIAccountBeforeProfit(
 	if account == nil {
 		return nil
 	}
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 
 	fresh := account
 	if s.schedulerSnapshot != nil {
@@ -1415,7 +1422,7 @@ func (s *OpenAIGatewayService) recheckSelectedOpenAIAccountFromDBBeforeProfit(ct
 	if account == nil {
 		return nil
 	}
-	platform = normalizeOpenAICompatiblePlatform(platform)
+	platform = NormalizeOpenAICompatiblePlatform(platform)
 	if s.schedulerSnapshot == nil || s.accountRepo == nil {
 		if !isOpenAICompatibleAccountEligibleForRequestBeforeProfit(ctx, account, platform, requestedModel, requireCompact, requiredCapability) {
 			return nil
diff --git a/backend/internal/service/openai_gateway_service.go b/backend/internal/service/openai_gateway_service.go
index a9200b10b744..25b3333f5a40 100644
--- a/backend/internal/service/openai_gateway_service.go
+++ b/backend/internal/service/openai_gateway_service.go
@@ -1205,7 +1205,7 @@ func (s *OpenAIGatewayService) GetAccessToken(ctx context.Context, account *Acco
 			}
 			return apiKey, "apikey", nil
 		}
-		apiKey := account.GetOpenAIApiKey()
+		apiKey := strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
 		if apiKey == "" {
 			return "", "", errors.New("api_key not found in credentials")
 		}
diff --git a/backend/internal/service/openai_ws_forwarder_payload.go b/backend/internal/service/openai_ws_forwarder_payload.go
index 4c02cb30f186..d783c9810420 100644
--- a/backend/internal/service/openai_ws_forwarder_payload.go
+++ b/backend/internal/service/openai_ws_forwarder_payload.go
@@ -42,7 +42,7 @@ func (s *OpenAIGatewayService) buildOpenAIResponsesWSURL(account *Account) (stri
 			if err != nil {
 				return "", err
 			}
-			targetURL = buildOpenAIResponsesURL(validatedURL)
+			targetURL = buildOpenAIResponsesURLForPlatform(account.Platform, validatedURL)
 		}
 	default:
 		targetURL = openaiPlatformAPIURL
diff --git a/backend/internal/service/ratelimit_cn_providers.go b/backend/internal/service/ratelimit_cn_providers.go
new file mode 100644
index 000000000000..4aab5a87510c
--- /dev/null
+++ b/backend/internal/service/ratelimit_cn_providers.go
@@ -0,0 +1,154 @@
+package service
+
+import (
+	"context"
+	"log/slog"
+	"net/http"
+	"strings"
+	"time"
+)
+
+// 国产供应商（kimi/zhipu/deepseek）的响应式冷却辅助。
+//
+// 与 openai/anthropic 不同：
+//   - 余额不足是「可恢复」状态（充值/检测恢复后自动重新调度），不能走 handleAuthError
+//     永久置 status=error。这里改为 SetTempUnschedulable，由 CN 余额检测周期任务
+//     （cn_provider_balance_check_service.go）在余额恢复后 ClearTempUnschedulable。
+//   - Coding Plan 滚动窗口耗尽（429）的冷却终点应是真实的窗口重置时间（已由
+//     CNProviderQuotaService 落入 account.Extra 快照），而非默认的秒级兜底。
+
+// cnBalanceExtraSuffixLow 标记账号响应过「余额不足」，供余额检测任务区分
+// 「确属余额不足」与「尚未探测」。
+const cnBalanceExtraSuffixLow = "balance_low"
+
+// cnBalanceLowReasonPrefix 是余额不足临时停调 reason 的稳定前缀。
+// 周期余额检测任务据此识别「是我们停调的」并在余额恢复后安全清除——不会误清
+// 其他子系统（阈值/限流/401）写入的临时停调。
+const cnBalanceLowReasonPrefix = "cn_balance_low"
+
+// cnBalanceLowReason 构造余额不足临时停调的 reason（带稳定前缀）。
+func cnBalanceLowReason(upstreamMsg string) string {
+	if upstreamMsg = strings.TrimSpace(upstreamMsg); upstreamMsg != "" {
+		return cnBalanceLowReasonPrefix + ": " + upstreamMsg
+	}
+	return cnBalanceLowReasonPrefix + ": 余额不足，账号临时停调"
+}
+
+// cnProviderResponseIndicatesInsufficientBalance 通过响应体文案识别余额不足
+// （智谱 payg 无独立余额端点，仅能靠响应文案识别）。
+func cnProviderResponseIndicatesInsufficientBalance(body []byte) bool {
+	if len(body) == 0 {
+		return false
+	}
+	s := strings.ToLower(string(body))
+	return strings.Contains(s, "余额不足") ||
+		strings.Contains(s, "insufficient balance") ||
+		strings.Contains(s, "insufficient_credit") ||
+		strings.Contains(s, "balance is not enough") ||
+		strings.Contains(s, "no enough balance")
+}
+
+// handleCNProviderInsufficientBalance 把余额不足标记为可恢复的临时停调：
+// 写入 balance_low 快照 + SetTempUnschedulable 一个余额检测周期，
+// 由周期任务在余额恢复后清除。返回前已通知调度阻塞。
+func (s *RateLimitService) handleCNProviderInsufficientBalance(
+	ctx context.Context,
+	account *Account,
+	upstreamMsg string,
+) {
+	msg := cnBalanceLowReason(upstreamMsg)
+
+	if err := s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{
+		cnExtraKey(account.Platform, cnBalanceExtraSuffixLow): true,
+	}); err != nil {
+		slog.Warn("cn_balance_low_mark_failed", "account_id", account.ID, "error", err)
+	}
+
+	until := time.Now().Add(s.cnBalanceCooldownDuration())
+	s.notifyAccountSchedulingBlocked(account, until, "cn_insufficient_balance")
+	if err := s.accountRepo.SetTempUnschedulable(ctx, account.ID, until, msg); err != nil {
+		slog.Warn("cn_balance_set_temp_unschedulable_failed", "account_id", account.ID, "error", err)
+		return
+	}
+	slog.Info("cn_provider_insufficient_balance",
+		"account_id", account.ID,
+		"platform", account.Platform,
+		"until", until.UTC(),
+	)
+}
+
+// cnBalanceCooldownDuration 返回余额不足临时停调的持续时长（= 2× 余额检测周期，
+// 默认 20 分钟）。周期任务会在余额恢复后提前清除，故此处只需保证冷却覆盖到下一次
+// 周期检测即可。
+func (s *RateLimitService) cnBalanceCooldownDuration() time.Duration {
+	minutes := 10
+	if s != nil && s.cfg != nil {
+		if cfgMin := s.cfg.Gateway.CNProviders.BalanceCheckIntervalMinutes; cfgMin > 0 {
+			minutes = cfgMin
+		}
+	}
+	cooldown := time.Duration(minutes) * time.Minute * 2
+	if cooldown < time.Minute {
+		cooldown = 10 * time.Minute
+	}
+	return cooldown
+}
+
+// cnProviderQuotaSnapshotReset 读取 Coding Plan 账号快照中最早一个仍在未来的窗口
+// 重置时间（5h / weekly）。429 多数由 5h 滚动窗口触发，取较早的重置点可避免
+// 把账号冷却到 weekly 重置（可达数天）的过度停调；如果确是 weekly 窗口耗尽，
+// 周期额度探测刷新快照后阈值评估会再次停调到正确的时间点。
+// 无快照或均已过期返回 nil。
+func cnProviderQuotaSnapshotReset(account *Account, now time.Time) *time.Time {
+	if account == nil || !account.IsCNProvider() || !account.IsCodingPlan() || len(account.Extra) == 0 {
+		return nil
+	}
+	provider := account.Platform
+	var earliest *time.Time
+	for _, suffix := range []string{cnExtraSuffix5hReset, cnExtraSuffixWeeklyReset} {
+		t := parseSchedulingResetAt(account.Extra[cnExtraKey(provider, suffix)])
+		if t == nil || !t.After(now) {
+			continue
+		}
+		if earliest == nil || t.Before(*earliest) {
+			earliest = t
+		}
+	}
+	return earliest
+}
+
+// applyCNProviderReactive429 处理国产供应商的 429 响应。
+// 返回 true 表示已处理（调用方应 return），false 表示未命中、继续走默认 429 逻辑。
+func (s *RateLimitService) applyCNProviderReactive429(
+	ctx context.Context,
+	account *Account,
+	headers http.Header,
+	responseBody []byte,
+) bool {
+	if !account.IsCNProvider() {
+		return false
+	}
+	// 1) 余额不足文案：可恢复临时停调（含智谱 payg 这类无余额端点的场景）。
+	if cnProviderResponseIndicatesInsufficientBalance(responseBody) {
+		s.handleCNProviderInsufficientBalance(ctx, account, extractUpstreamErrorMessage(responseBody))
+		return true
+	}
+	// 2) Coding Plan 窗口耗尽：冷却到快照中最早的窗口重置点（见
+	// cnProviderQuotaSnapshotReset：429 多由 5h 窗口触发，取较早点避免过度停调）。
+	if account.IsCodingPlan() {
+		if until := cnProviderQuotaSnapshotReset(account, time.Now()); until != nil {
+			s.notifyAccountSchedulingBlocked(account, *until, "429")
+			if err := s.accountRepo.SetRateLimited(ctx, account.ID, *until); err != nil {
+				slog.Warn("rate_limit_set_failed", "account_id", account.ID, "error", err)
+				return true
+			}
+			slog.Info("cn_coding_plan_rate_limited",
+				"account_id", account.ID,
+				"platform", account.Platform,
+				"reset_at", *until,
+			)
+			return true
+		}
+	}
+	return false
+}
diff --git a/backend/internal/service/ratelimit_service.go b/backend/internal/service/ratelimit_service.go
index ee5b09bbb4b1..0f1449191485 100644
--- a/backend/internal/service/ratelimit_service.go
+++ b/backend/internal/service/ratelimit_service.go
@@ -436,6 +436,13 @@ func (s *RateLimitService) HandleUpstreamError(ctx context.Context, account *Acc
 			shouldDisable = true
 		}
 	case 402:
+		// 国产供应商：余额不足是可恢复状态（充值/检测恢复后由周期任务自动解除），
+		// 不能走 handleAuthError 永久置 status=error。改为可恢复的临时停调。
+		if account.IsCNProvider() {
+			s.handleCNProviderInsufficientBalance(ctx, account, upstreamMsg)
+			shouldDisable = true
+			break
+		}
 		// OpenAI: deactivated_workspace 表示工作区已停用，直接标记 error
 		if account.Platform == PlatformOpenAI && gjson.GetBytes(responseBody, "detail.code").String() == "deactivated_workspace" {
 			msg := "Workspace deactivated (402): workspace has been deactivated"
@@ -1048,6 +1055,13 @@ func (s *RateLimitService) handle429(ctx context.Context, account *Account, head
 	if account.IsShadow() {
 		return
 	}
+	// 国产供应商（kimi/zhipu/deepseek）的 429 走专用可恢复路径：余额不足 → 临时停调，
+	// Coding Plan 窗口耗尽 → 冷却到快照重置点。未命中则继续默认 429 逻辑。
+	if account.IsCNProvider() {
+		if s.applyCNProviderReactive429(ctx, account, headers, responseBody) {
+			return
+		}
+	}
 	// 1. OpenAI 平台：优先尝试解析 x-codex-* 响应头（用于 rate_limit_exceeded）
 	if account.Platform == PlatformOpenAI {
 		persistOpenAI429PlanType(ctx, s.accountRepo, account, responseBody)
diff --git a/backend/internal/service/upstream_billing_probe.go b/backend/internal/service/upstream_billing_probe.go
index d66cd905dc59..c6a62bab09a5 100644
--- a/backend/internal/service/upstream_billing_probe.go
+++ b/backend/internal/service/upstream_billing_probe.go
@@ -965,7 +965,9 @@ func decodeUpstreamBillingProbeSnapshot(extra map[string]any) *UpstreamBillingPr
 
 // IsUpstreamBillingProbeIdentity reports whether an account identity may opt
 // in to the upstream billing probe. `/v1/sub2api/billing` is a key-scoped
-// sub2api convention shared by the five supported API-key platforms.
+// sub2api convention shared by the supported API-key platforms (including the
+// CN providers, whose official-domain accounts are short-circuited to
+// "unsupported" by upstreamBillingProbeTargetIsOfficialAPI).
 // Non-sub2api upstreams return 404 and the snapshot records "unsupported".
 // Only AccountTypeAPIKey is in scope. OAuth/Bedrock hold no static API key to
 // present at all; AccountTypeUpstream (antigravity relay accounts) does carry
@@ -978,7 +980,8 @@ func IsUpstreamBillingProbeIdentity(platform, accountType string) bool {
 		return false
 	}
 	switch platform {
-	case PlatformOpenAI, PlatformAnthropic, PlatformGemini, PlatformAntigravity, PlatformGrok:
+	case PlatformOpenAI, PlatformAnthropic, PlatformGemini, PlatformAntigravity, PlatformGrok,
+		PlatformKimi, PlatformZhipu, PlatformDeepseek:
 		return true
 	default:
 		return false
@@ -1002,6 +1005,9 @@ func isUpstreamBillingProbeAccount(account *Account) bool {
 // ollama.com is a first-class configuration here (Ollama Cloud accounts are
 // platform openai/anthropic with base_url https://ollama.com/v1), and it is
 // an official provider API just like the rest, so it belongs on this list.
+// CN provider domains (moonshot.cn / kimi.com / bigmodel.cn / deepseek.com)
+// serve the same role: official APIs that can never host /v1/sub2api/billing,
+// so their accounts short-circuit to "unsupported" without a request.
 var upstreamBillingProbeOfficialAPIDomains = []string{
 	"anthropic.com",
 	"googleapis.com",
@@ -1009,6 +1015,10 @@ var upstreamBillingProbeOfficialAPIDomains = []string{
 	"grok.com",
 	"openai.com",
 	"ollama.com",
+	"moonshot.cn",
+	"kimi.com",
+	"bigmodel.cn",
+	"deepseek.com",
 }
 
 func upstreamBillingProbeTargetIsOfficialAPI(baseURL string) bool {
diff --git a/backend/internal/service/upstream_billing_probe_multiplatform_test.go b/backend/internal/service/upstream_billing_probe_multiplatform_test.go
index 018f2b719857..6690a80c154c 100644
--- a/backend/internal/service/upstream_billing_probe_multiplatform_test.go
+++ b/backend/internal/service/upstream_billing_probe_multiplatform_test.go
@@ -11,11 +11,12 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
-// 探测资格：/v1/sub2api/billing 是 key 级端点，五个
-// 受支持平台的 API-key 账号都可开启探测；OAuth/Bedrock 无静态 Key 仍不合格。
+// 探测资格：/v1/sub2api/billing 是 key 级端点，全部
+// 受支持平台（含国产供应商）的 API-key 账号都可开启探测；OAuth/Bedrock 无静态 Key 仍不合格。
 func TestUpstreamBillingProbeIdentityCoversAllAPIKeyPlatforms(t *testing.T) {
 	for _, platform := range []string{
 		PlatformOpenAI, PlatformGrok, PlatformAnthropic, PlatformGemini, PlatformAntigravity,
+		PlatformKimi, PlatformZhipu, PlatformDeepseek,
 	} {
 		require.True(t, IsUpstreamBillingProbeIdentity(platform, AccountTypeAPIKey), platform)
 		require.True(t, isUpstreamBillingProbeAccount(&Account{Platform: platform, Type: AccountTypeAPIKey}), platform)
@@ -124,6 +125,14 @@ func TestUpstreamBillingProbeOfficialAPIBaseURLIsUnsupportedWithoutRequest(t *te
 		{PlatformAnthropic, "https://ollama.com/v1"},
 		{PlatformAnthropic, "https://ollama.com"},
 		{PlatformAnthropic, "https://www.ollama.com/v1"},
+		// 国产供应商官方域（含各协议端点）同样是官方 API，创建即开探测也不发请求。
+		{PlatformKimi, "https://api.moonshot.cn/v1"},
+		{PlatformKimi, "https://api.moonshot.cn/anthropic"},
+		{PlatformKimi, "https://api.kimi.com/coding"},
+		{PlatformZhipu, "https://open.bigmodel.cn/api/paas/v4"},
+		{PlatformZhipu, "https://open.bigmodel.cn/api/anthropic"},
+		{PlatformDeepseek, "https://api.deepseek.com"},
+		{PlatformDeepseek, "https://api.deepseek.com/anthropic"},
 	}
 	for i, tc := range cases {
 		account := &Account{
@@ -160,6 +169,11 @@ func TestUpstreamBillingProbeOfficialAPIHostMatchingIsNormalized(t *testing.T) {
 	require.True(t, upstreamBillingProbeTargetIsOfficialAPI("https://ollama.com:443/v1"))
 	require.True(t, upstreamBillingProbeTargetIsOfficialAPI("https://www.ollama.com/v1"))
 	require.True(t, upstreamBillingProbeTargetIsOfficialAPI("HTTPS://OLLAMA.COM./v1"))
+	// 国产供应商官方域及子域。
+	require.True(t, upstreamBillingProbeTargetIsOfficialAPI("https://api.moonshot.cn/v1"))
+	require.True(t, upstreamBillingProbeTargetIsOfficialAPI("https://api.kimi.com/coding/v1"))
+	require.True(t, upstreamBillingProbeTargetIsOfficialAPI("https://open.bigmodel.cn/api/anthropic"))
+	require.True(t, upstreamBillingProbeTargetIsOfficialAPI("https://api.deepseek.com/anthropic"))
 	// 相似但不同的注册域不拦：中转完全可能叫 *-x.ai 之外的任何名字。
 	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://relay.example/v1"))
 	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://notx.ai"))
@@ -168,6 +182,11 @@ func TestUpstreamBillingProbeOfficialAPIHostMatchingIsNormalized(t *testing.T) {
 	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://notollama.com/v1"))
 	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://ollama.com.evil.example/v1"))
 	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://ollama.example/v1"))
+	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://notmoonshot.cn/v1"))
+	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://moonshot.cn.evil.example/v1"))
+	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://kimi.example/v1"))
+	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://notbigmodel.cn"))
+	require.False(t, upstreamBillingProbeTargetIsOfficialAPI("https://deepseek.example.com"))
 }
 
 // OpenAI 语义保持不变：无自定义 base 时仍探官方域，且沿用 openai 传输画像。
diff --git a/backend/internal/service/upstream_models.go b/backend/internal/service/upstream_models.go
index 9a4e4311b057..df926b02971e 100644
--- a/backend/internal/service/upstream_models.go
+++ b/backend/internal/service/upstream_models.go
@@ -137,7 +137,8 @@ func (s *AccountTestService) buildUpstreamModelsRequest(ctx context.Context, acc
 		return s.buildAntigravityAPIKeyModelsRequest(ctx, account)
 	case account.IsGrok():
 		return s.buildGrokUpstreamModelsRequest(ctx, account)
-	case account.IsOpenAI():
+	case account.IsOpenAI() || account.IsCNProvider():
+		// 国产 OpenAI 兼容供应商（kimi/zhipu/deepseek）复用 OpenAI /v1/models 探测。
 		return s.buildOpenAIUpstreamModelsRequest(ctx, account)
 	case account.IsGemini():
 		return s.buildGeminiUpstreamModelsRequest(ctx, account)
@@ -347,12 +348,14 @@ func (s *AccountTestService) buildOpenAIUpstreamModelsRequest(ctx context.Contex
 			fmt.Sprintf("Unsupported OpenAI account type for upstream model sync: %s", account.Type), nil,
 		)
 	}
-	apiKey := strings.TrimSpace(account.GetOpenAIApiKey())
+	apiKey := strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
 	if apiKey == "" {
 		return nil, newUpstreamModelSyncConfigError("No OpenAI API key is available", nil)
 	}
 
-	baseURL := account.GetOpenAIBaseURL()
+	// 协议感知：Anthropic 协议账号的凭证 base_url 指向 /anthropic 端点，模型
+	// 列表同步需使用 OpenAI 格式 base（供应商 × 模式默认）。
+	baseURL := account.GetOpenAIFormatBaseURL()
 	if strings.TrimSpace(baseURL) == "" {
 		baseURL = "https://api.openai.com"
 	}
diff --git a/backend/internal/service/wire.go b/backend/internal/service/wire.go
index 2e26c0035c22..d6bb56fac423 100644
--- a/backend/internal/service/wire.go
+++ b/backend/internal/service/wire.go
@@ -261,6 +261,45 @@ func ProvideGrokQuotaService(
 	return service
 }
 
+// ProvideCNProviderQuotaService 构造国产供应商 Coding Plan 额度探测服务。
+func ProvideCNProviderQuotaService(
+	accountRepo AccountRepository,
+	proxyRepo ProxyRepository,
+	httpUpstream HTTPUpstream,
+	cfg *config.Config,
+) *CNProviderQuotaService {
+	return NewCNProviderQuotaService(accountRepo, proxyRepo, httpUpstream, cfg)
+}
+
+// ProvideCNProviderBalanceService 构造国产供应商余额探测服务。
+func ProvideCNProviderBalanceService(
+	accountRepo AccountRepository,
+	proxyRepo ProxyRepository,
+	httpUpstream HTTPUpstream,
+	cfg *config.Config,
+) *CNProviderBalanceService {
+	return NewCNProviderBalanceService(accountRepo, proxyRepo, httpUpstream, cfg)
+}
+
+// ProvideCNProviderBalanceCheckService 构造并启动周期余额/额度检测任务。
+// payg 账号探余额（低余额停调）；coding plan 账号探 5h/weekly 滚动窗口
+// （落 extra 快照供调度阈值评估自动停调）。
+// 间隔取自 gateway.cn_providers.balance_check_interval_minutes；<=0 或关闭时不启动。
+func ProvideCNProviderBalanceCheckService(
+	accountRepo AccountRepository,
+	balanceService *CNProviderBalanceService,
+	quotaService *CNProviderQuotaService,
+	cfg *config.Config,
+) *CNProviderBalanceCheckService {
+	minutes := 10
+	if cfg != nil && cfg.Gateway.CNProviders.BalanceCheckIntervalMinutes > 0 {
+		minutes = cfg.Gateway.CNProviders.BalanceCheckIntervalMinutes
+	}
+	svc := NewCNProviderBalanceCheckService(accountRepo, balanceService, quotaService, cfg, time.Duration(minutes)*time.Minute)
+	svc.Start()
+	return svc
+}
+
 // ProvideGeminiTokenProvider creates GeminiTokenProvider with OAuthRefreshAPI injection
 func ProvideGeminiTokenProvider(
 	accountRepo AccountRepository,
@@ -796,6 +835,9 @@ var ProviderSet = wire.NewSet(
 	ProvideOpenAITokenProvider,
 	ProvideOpenAIQuotaService,
 	ProvideGrokQuotaService,
+	ProvideCNProviderQuotaService,
+	ProvideCNProviderBalanceService,
+	ProvideCNProviderBalanceCheckService,
 	ProvideClaudeTokenProvider,
 	NewAntigravityGatewayService,
 	ProvideRateLimitService,
diff --git a/docker-compose.yml b/docker-compose.yml
new file mode 100644
index 000000000000..67edd44e2205
--- /dev/null
+++ b/docker-compose.yml
@@ -0,0 +1,302 @@
+# =============================================================================
+# Sub2API Docker Compose Configuration
+# =============================================================================
+# Quick Start:
+#   1. Copy .env.example to .env and configure
+#   2. docker-compose up -d
+#   3. Check logs: docker-compose logs -f sub2api
+#   4. Access: http://localhost:8080
+#
+# All configuration is done via environment variables.
+# No Setup Wizard needed - the system auto-initializes on first run.
+# =============================================================================
+
+services:
+  # ===========================================================================
+  # Sub2API Application
+  # ===========================================================================
+  sub2api:
+    image: randark/sub2api:latest
+    container_name: sub2api
+    restart: unless-stopped
+    security_opt:
+      - no-new-privileges:true
+    ulimits:
+      nofile:
+        soft: 100000
+        hard: 100000
+    ports:
+      - "${BIND_HOST:-0.0.0.0}:${SERVER_PORT:-8080}:8080"
+    volumes:
+      # Data persistence (config.yaml will be auto-generated here)
+      - sub2api_data:/app/data
+      # Optional: Mount custom config.yaml (uncomment and create the file first)
+      # Copy config.example.yaml to config.yaml, modify it, then uncomment:
+      # - ./config.yaml:/app/data/config.yaml
+      # Optional: Mount a custom Codex instructions template file, then point
+      # gateway.forced_codex_instructions_template_file at /app/data/codex-instructions.md.tmpl
+      # in config.yaml.
+      # - ./codex-instructions.md.tmpl:/app/data/codex-instructions.md.tmpl:ro
+    environment:
+      # =======================================================================
+      # Auto Setup (REQUIRED for Docker deployment)
+      # =======================================================================
+      - AUTO_SETUP=true
+
+      # =======================================================================
+      # Server Configuration
+      # =======================================================================
+      - SERVER_HOST=0.0.0.0
+      - SERVER_PORT=8080
+      - SERVER_MODE=${SERVER_MODE:-release}
+      - ENABLE_SERVER_TIMING=${ENABLE_SERVER_TIMING:-false}
+      - RUN_MODE=${RUN_MODE:-standard}
+      - UPDATE_GITHUB_TOKEN=${UPDATE_GITHUB_TOKEN:-}
+      - ALIPAY_MOBILE_PRECREATE_DEEP_LINK=${ALIPAY_MOBILE_PRECREATE_DEEP_LINK:-}
+
+      # =======================================================================
+      # Database Configuration (PostgreSQL)
+      # =======================================================================
+      - DATABASE_HOST=postgres
+      - DATABASE_PORT=5432
+      - DATABASE_USER=${POSTGRES_USER:-sub2api}
+      - DATABASE_PASSWORD=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required}
+      - DATABASE_DBNAME=${POSTGRES_DB:-sub2api}
+      - DATABASE_SSLMODE=disable
+      - DATABASE_MAX_OPEN_CONNS=${DATABASE_MAX_OPEN_CONNS:-50}
+      - DATABASE_MAX_IDLE_CONNS=${DATABASE_MAX_IDLE_CONNS:-10}
+      - DATABASE_CONN_MAX_LIFETIME_MINUTES=${DATABASE_CONN_MAX_LIFETIME_MINUTES:-30}
+      - DATABASE_CONN_MAX_IDLE_TIME_MINUTES=${DATABASE_CONN_MAX_IDLE_TIME_MINUTES:-5}
+
+      # =======================================================================
+      # Redis Configuration
+      # =======================================================================
+      - REDIS_HOST=redis
+      - REDIS_PORT=6379
+      - REDIS_USERNAME=${REDIS_USERNAME:-}
+      - REDIS_PASSWORD=${REDIS_PASSWORD:-}
+      - REDIS_DB=${REDIS_DB:-0}
+      - REDIS_POOL_SIZE=${REDIS_POOL_SIZE:-1024}
+      - REDIS_MIN_IDLE_CONNS=${REDIS_MIN_IDLE_CONNS:-10}
+      - REDIS_ENABLE_TLS=${REDIS_ENABLE_TLS:-false}
+
+      # =======================================================================
+      # Admin Account (auto-created on first run)
+      # =======================================================================
+      - ADMIN_EMAIL=${ADMIN_EMAIL:-admin@sub2api.local}
+      - ADMIN_PASSWORD=${ADMIN_PASSWORD:-}
+
+      # =======================================================================
+      # JWT Configuration
+      # =======================================================================
+      # IMPORTANT: Set a fixed JWT_SECRET to prevent login sessions from being
+      # invalidated after container restarts. If left empty, a random secret
+      # will be generated on each startup.
+      # Generate a secure secret: openssl rand -hex 32
+      - JWT_SECRET=${JWT_SECRET:-}
+      - JWT_EXPIRE_HOUR=${JWT_EXPIRE_HOUR:-24}
+
+      # =======================================================================
+      # Setup Configuration
+      # =======================================================================
+      - SETUP_MIGRATION_TIMEOUT_SECONDS=${SETUP_MIGRATION_TIMEOUT_SECONDS:-0}
+
+      # =======================================================================
+      # TOTP (2FA) Configuration
+      # =======================================================================
+      # IMPORTANT: Set a fixed encryption key for TOTP secrets. If left empty,
+      # a random key will be generated on each startup, causing all existing
+      # TOTP configurations to become invalid (users won't be able to login
+      # with 2FA).
+      # Generate a secure key: openssl rand -hex 32
+      - TOTP_ENCRYPTION_KEY=${TOTP_ENCRYPTION_KEY:-}
+
+      # =======================================================================
+      # Timezone Configuration
+      # This affects ALL time operations in the application:
+      # - Database timestamps
+      # - Usage statistics "today" boundary
+      # - Subscription expiry times
+      # - Log timestamps
+      # Common values: Asia/Shanghai, America/New_York, Europe/London, UTC
+      # =======================================================================
+      - TZ=${TZ:-Asia/Shanghai}
+
+      # =======================================================================
+      # Gemini OAuth Configuration (for Gemini accounts)
+      # =======================================================================
+      - GEMINI_OAUTH_CLIENT_ID=${GEMINI_OAUTH_CLIENT_ID:-}
+      - GEMINI_OAUTH_CLIENT_SECRET=${GEMINI_OAUTH_CLIENT_SECRET:-}
+      - GEMINI_OAUTH_SCOPES=${GEMINI_OAUTH_SCOPES:-}
+      - GEMINI_QUOTA_POLICY=${GEMINI_QUOTA_POLICY:-}
+
+      # Built-in OAuth client secrets (optional)
+      # SECURITY: This repo does not embed third-party client_secret.
+      - GEMINI_CLI_OAUTH_CLIENT_SECRET=${GEMINI_CLI_OAUTH_CLIENT_SECRET:-}
+      - ANTIGRAVITY_OAUTH_CLIENT_SECRET=${ANTIGRAVITY_OAUTH_CLIENT_SECRET:-}
+      - ANTIGRAVITY_USER_AGENT_VERSION=${ANTIGRAVITY_USER_AGENT_VERSION:-}
+
+      # =======================================================================
+      # Security Configuration (URL Allowlist)
+      # =======================================================================
+      # Enable URL allowlist validation (false to skip allowlist checks)
+      - SECURITY_URL_ALLOWLIST_ENABLED=${SECURITY_URL_ALLOWLIST_ENABLED:-false}
+      # Allow insecure HTTP URLs when allowlist is disabled (default: true; set to false to require https)
+      - SECURITY_URL_ALLOWLIST_ALLOW_INSECURE_HTTP=${SECURITY_URL_ALLOWLIST_ALLOW_INSECURE_HTTP:-true}
+      # Allow private IP addresses for upstream/pricing/CRS (default: true; set to false to block private hosts)
+      - SECURITY_URL_ALLOWLIST_ALLOW_PRIVATE_HOSTS=${SECURITY_URL_ALLOWLIST_ALLOW_PRIVATE_HOSTS:-true}
+      # Upstream hosts whitelist (comma-separated, only used when enabled=true)
+      - SECURITY_URL_ALLOWLIST_UPSTREAM_HOSTS=${SECURITY_URL_ALLOWLIST_UPSTREAM_HOSTS:-}
+
+      # =======================================================================
+      # Update Configuration (在线更新配置)
+      # =======================================================================
+      # Proxy for accessing GitHub (online updates + pricing data)
+      # Examples: http://host:port, socks5://host:port
+      - UPDATE_PROXY_URL=${UPDATE_PROXY_URL:-}
+
+      # =======================================================================
+      # Image Generation Stream & Concurrency
+      # =======================================================================
+      # OpenAI HTTP upstream protocol/timeout
+      - GATEWAY_OPENAI_RESPONSE_HEADER_TIMEOUT=${GATEWAY_OPENAI_RESPONSE_HEADER_TIMEOUT:-0}
+      - GATEWAY_OPENAI_HTTP2_ENABLED=${GATEWAY_OPENAI_HTTP2_ENABLED:-true}
+      - GATEWAY_OPENAI_HTTP2_ALLOW_PROXY_FALLBACK_TO_HTTP1=${GATEWAY_OPENAI_HTTP2_ALLOW_PROXY_FALLBACK_TO_HTTP1:-true}
+      - GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD=${GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD:-2}
+      - GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS:-60}
+      - GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS:-600}
+      - GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD:-2}
+      - GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS:-60}
+      - GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS:-600}
+      - GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT=${GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT:-900}
+      - GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL=${GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL:-10}
+      - GATEWAY_IMAGE_CONCURRENCY_ENABLED=${GATEWAY_IMAGE_CONCURRENCY_ENABLED:-false}
+      - GATEWAY_IMAGE_CONCURRENCY_MAX_CONCURRENT_REQUESTS=${GATEWAY_IMAGE_CONCURRENCY_MAX_CONCURRENT_REQUESTS:-0}
+      - GATEWAY_IMAGE_CONCURRENCY_OVERFLOW_MODE=${GATEWAY_IMAGE_CONCURRENCY_OVERFLOW_MODE:-reject}
+      - GATEWAY_IMAGE_CONCURRENCY_WAIT_TIMEOUT_SECONDS=${GATEWAY_IMAGE_CONCURRENCY_WAIT_TIMEOUT_SECONDS:-30}
+      - GATEWAY_IMAGE_CONCURRENCY_MAX_WAITING_REQUESTS=${GATEWAY_IMAGE_CONCURRENCY_MAX_WAITING_REQUESTS:-100}
+    depends_on:
+      postgres:
+        condition: service_healthy
+      redis:
+        condition: service_healthy
+    networks:
+      - sub2api-network
+    healthcheck:
+      test:
+        [
+          "CMD",
+          "wget",
+          "-q",
+          "-T",
+          "5",
+          "-O",
+          "/dev/null",
+          "http://localhost:8080/health",
+        ]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 30s
+
+  # ===========================================================================
+  # PostgreSQL Database
+  # ===========================================================================
+  postgres:
+    image: postgres:18-alpine
+    container_name: sub2api-postgres
+    restart: unless-stopped
+    ulimits:
+      nofile:
+        soft: 100000
+        hard: 100000
+    # Pass the tuning values documented in .env.example to postgres. The
+    # fallbacks are the postgres:18 stock defaults, so nothing changes when
+    # the variables are not set.
+    command: >
+      postgres
+      -c max_connections=${POSTGRES_MAX_CONNECTIONS:-100}
+      -c shared_buffers=${POSTGRES_SHARED_BUFFERS:-128MB}
+      -c effective_cache_size=${POSTGRES_EFFECTIVE_CACHE_SIZE:-4GB}
+      -c maintenance_work_mem=${POSTGRES_MAINTENANCE_WORK_MEM:-64MB}
+    volumes:
+      - postgres_data:/var/lib/postgresql/data
+    environment:
+      # postgres:18-alpine 默认 PGDATA=/var/lib/postgresql/18/docker（位于镜像声明的匿名卷 /var/lib/postgresql 内）。
+      # 若不显式设置 PGDATA，则即使挂载了 postgres_data 到 /var/lib/postgresql/data，数据也不会落盘到该命名卷，
+      # docker compose down/up 后会触发 initdb 重新初始化，导致用户/密码等数据丢失。
+      - PGDATA=/var/lib/postgresql/data
+      - POSTGRES_USER=${POSTGRES_USER:-sub2api}
+      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required}
+      - POSTGRES_DB=${POSTGRES_DB:-sub2api}
+      - TZ=${TZ:-Asia/Shanghai}
+    networks:
+      - sub2api-network
+    healthcheck:
+      test:
+        [
+          "CMD-SHELL",
+          "pg_isready -U ${POSTGRES_USER:-sub2api} -d ${POSTGRES_DB:-sub2api}",
+        ]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 10s
+    # 注意：不暴露端口到宿主机，应用通过内部网络连接
+    # 如需调试，可临时添加：ports: ["127.0.0.1:5433:5432"]
+
+  # ===========================================================================
+  # Redis Cache
+  # ===========================================================================
+  redis:
+    image: redis:8-alpine
+    container_name: sub2api-redis
+    restart: unless-stopped
+    ulimits:
+      nofile:
+        soft: 100000
+        hard: 100000
+    volumes:
+      - redis_data:/data
+    # The command is one quoted script for the inner `sh -c`. Compose keeps
+    # the newlines inside the quoted string, so every line needs a trailing
+    # `\` — without it, `redis-server` on the first line runs with no flags
+    # at all, and the --save/--appendonly/--appendfsync lines are never read.
+    command: >
+      sh -c '
+        redis-server \
+        --save 60 1 \
+        --appendonly yes \
+        --appendfsync everysec \
+        ${REDIS_PASSWORD:+--requirepass "$REDIS_PASSWORD"}'
+    environment:
+      - TZ=${TZ:-Asia/Shanghai}
+      # REDISCLI_AUTH is used by redis-cli for authentication (safer than -a flag)
+      - REDISCLI_AUTH=${REDIS_PASSWORD:-}
+    networks:
+      - sub2api-network
+    healthcheck:
+      test: ["CMD", "redis-cli", "ping"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 5s
+    # 注意：不暴露端口到宿主机，应用通过内部网络连接
+    # 如需调试，可临时添加：ports: ["127.0.0.1:6379:6379"]
+# =============================================================================
+# Volumes
+# =============================================================================
+volumes:
+  sub2api_data:
+    driver: local
+  postgres_data:
+    driver: local
+  redis_data:
+    driver: local
+
+# =============================================================================
+# Networks
+# =============================================================================
+networks:
+  sub2api-network:
+    driver: bridge
diff --git a/frontend/src/api/admin/cnProviders.ts b/frontend/src/api/admin/cnProviders.ts
new file mode 100644
index 000000000000..668f67a73ee9
--- /dev/null
+++ b/frontend/src/api/admin/cnProviders.ts
@@ -0,0 +1,70 @@
+/**
+ * Admin CN providers (Kimi / Zhipu / DeepSeek) API endpoints.
+ * Coding-plan rolling-window quota probe + payg balance probe.
+ */
+
+import { apiClient } from '../client'
+
+/** 滚动用量窗口档（5 小时 / 每周），对齐后端 service.CNQuotaTier。 */
+export interface CNQuotaTier {
+  window: '5h' | 'weekly'
+  used_percent: number
+  reset_at?: string
+}
+
+/** Coding Plan 额度探测结果（kimi / zhipu），对齐后端 CNProviderQuotaProbeResult。 */
+export interface CNProviderQuotaProbeResult {
+  provider: string
+  source?: string
+  success: boolean
+  credential_valid: boolean
+  tiers?: CNQuotaTier[]
+  plan_level?: string
+  status_code?: number
+  fetched_at: number
+  persisted: boolean
+  error?: string
+}
+
+/** 单币种余额明细（deepseek 双币种账号含 CNY + USD 两条）。 */
+export interface CNProviderBalanceEntry {
+  currency: string
+  balance: number
+}
+
+/** payg 余额探测结果（kimi / deepseek），对齐后端 CNProviderBalanceResult。 */
+export interface CNProviderBalanceResult {
+  provider: string
+  success: boolean
+  /** 主币种余额（balances 首条，兼容单币种展示）。 */
+  balance: number
+  currency?: string
+  /** 多币种明细；缺省时按主币种展示。 */
+  balances?: CNProviderBalanceEntry[]
+  available: boolean
+  status_code?: number
+  fetched_at: number
+  persisted: boolean
+  error?: string
+}
+
+/** 查询 Coding Plan 滚动窗口用量（5h + weekly）。 */
+export async function queryQuota(id: number): Promise<CNProviderQuotaProbeResult> {
+  const { data } = await apiClient.get<CNProviderQuotaProbeResult>(
+    `/admin/cn-providers/accounts/${id}/quota`
+  )
+  return data
+}
+
+/** 查询 payg 账号余额。 */
+export async function queryBalance(id: number): Promise<CNProviderBalanceResult> {
+  const { data } = await apiClient.get<CNProviderBalanceResult>(
+    `/admin/cn-providers/accounts/${id}/balance`
+  )
+  return data
+}
+
+export default {
+  queryQuota,
+  queryBalance
+}
diff --git a/frontend/src/api/admin/index.ts b/frontend/src/api/admin/index.ts
index 80a7073decbd..dd976daad54c 100644
--- a/frontend/src/api/admin/index.ts
+++ b/frontend/src/api/admin/index.ts
@@ -18,6 +18,7 @@ import usageAPI from './usage'
 import geminiAPI from './gemini'
 import antigravityAPI from './antigravity'
 import grokAPI from './grok'
+import cnProvidersAPI from './cnProviders'
 import userAttributesAPI from './userAttributes'
 import opsAPI from './ops'
 import errorPassthroughAPI from './errorPassthrough'
@@ -54,6 +55,7 @@ export const adminAPI = {
   gemini: geminiAPI,
   antigravity: antigravityAPI,
   grok: grokAPI,
+  cnProviders: cnProvidersAPI,
   userAttributes: userAttributesAPI,
   ops: opsAPI,
   errorPassthrough: errorPassthroughAPI,
@@ -88,6 +90,7 @@ export {
   geminiAPI,
   antigravityAPI,
   grokAPI,
+  cnProvidersAPI,
   userAttributesAPI,
   opsAPI,
   errorPassthroughAPI,
diff --git a/frontend/src/api/admin/settings.ts b/frontend/src/api/admin/settings.ts
index b176024ae6bc..c8dbe8fb78f0 100644
--- a/frontend/src/api/admin/settings.ts
+++ b/frontend/src/api/admin/settings.ts
@@ -36,13 +36,19 @@ export type SchedulingThresholdPlatformType =
   | "openai"
   | "anthropic"
   | "grok"
+  | "kimi"
+  | "zhipu"
 
 export type AccountSchedulingThresholdsMap = Record<SchedulingThresholdPlatformType, number>
 
+// 与后端 AllowedSchedulingThresholdPlatforms 保持一致（deepseek 为余额型，
+// 走余额检测而非用量阈值）。
 export const SCHEDULING_THRESHOLD_PLATFORMS: SchedulingThresholdPlatformType[] = [
   "openai",
   "anthropic",
   "grok",
+  "kimi",
+  "zhipu",
 ]
 
 export function normalizeAccountSchedulingThresholdsMap(
diff --git a/frontend/src/components/account/AccountUsageCell.vue b/frontend/src/components/account/AccountUsageCell.vue
index 485b4b870102..ae9d55e3c217 100644
--- a/frontend/src/components/account/AccountUsageCell.vue
+++ b/frontend/src/components/account/AccountUsageCell.vue
@@ -422,6 +422,21 @@
       </div>
     </template>
 
+    <!-- CN providers (Kimi / Zhipu / DeepSeek): coding-plan quota or payg balance -->
+    <template v-else-if="account.platform === 'kimi' || account.platform === 'zhipu' || account.platform === 'deepseek'">
+      <div class="space-y-1">
+        <!-- 子单元格各自按 模式×平台 判定可见；两者都不可见时（智谱 payg 无公开
+             余额端点、coding 探测也不适用）才回落到占位符。 -->
+        <div
+          v-if="!cnQuotaCellVisible && !cnBalanceCellVisible"
+          class="text-xs text-gray-400"
+          :title="t('admin.accounts.cnProviders.noBalanceEndpoint')"
+        >-</div>
+        <CNProviderQuotaCell :account="account" />
+        <CNProviderBalanceCell :account="account" />
+      </div>
+    </template>
+
     <!-- Gemini platform: show quota + local usage window -->
     <template v-else-if="account.platform === 'gemini'">
       <!-- Auth Type + Tier Badge (first line) -->
@@ -625,7 +640,10 @@ import UsageProgressBar from './UsageProgressBar.vue'
 import AccountQuotaInfo from './AccountQuotaInfo.vue'
 import OpenAIQuotaResetCell from './OpenAIQuotaResetCell.vue'
 import GrokQuotaProbeCell from './GrokQuotaProbeCell.vue'
+import CNProviderQuotaCell from './CNProviderQuotaCell.vue'
+import CNProviderBalanceCell from './CNProviderBalanceCell.vue'
 import OllamaCloudUsageCell from './OllamaCloudUsageCell.vue'
+import { cnQuotaCellVisible as cnQuotaCellVisibleFn, cnBalanceCellVisible as cnBalanceCellVisibleFn } from './credentialsBuilder'
 
 // Module-level cache shared across all AccountUsageCell instances
 const _usageCache = new Map<number, { data: AccountUsageInfo; ts: number }>()
@@ -690,6 +708,15 @@ let visibilityObserver: IntersectionObserver | null = null
 const showUsageWindows = computed(() => {
   // Gemini: we can always compute local usage windows from DB logs (simulated quotas).
   if (props.account.platform === 'gemini') return true
+  // CN providers: apikey 账号也有滚动用量窗口（coding plan）或余额（payg），
+  // 由 CNProviderQuotaCell / CNProviderBalanceCell 自行探测与展示。
+  if (
+    props.account.platform === 'kimi' ||
+    props.account.platform === 'zhipu' ||
+    props.account.platform === 'deepseek'
+  ) {
+    return true
+  }
   return props.account.type === 'oauth' || props.account.type === 'setup-token'
 })
 
@@ -712,6 +739,15 @@ const shouldFetchUsage = computed(() => {
   return false
 })
 
+// CN 供应商子单元格可见性（与 CNProviderQuotaCell / CNProviderBalanceCell 共用
+// credentialsBuilder 的单一实现）：都不可见时显示 `-` 占位符。
+const cnAccountMode = computed(() => {
+  const mode = props.account.credentials?.account_mode
+  return typeof mode === 'string' ? mode : ''
+})
+const cnQuotaCellVisible = computed(() => cnQuotaCellVisibleFn(props.account.platform, cnAccountMode.value))
+const cnBalanceCellVisible = computed(() => cnBalanceCellVisibleFn(props.account.platform, cnAccountMode.value))
+
 const isBatchManaged = computed(() => typeof props.requestBatchedUsage === 'function')
 
 const showGeminiTodayStats = computed(() => {
diff --git a/frontend/src/components/account/CNProviderBalanceCell.vue b/frontend/src/components/account/CNProviderBalanceCell.vue
new file mode 100644
index 000000000000..93127cbf986e
--- /dev/null
+++ b/frontend/src/components/account/CNProviderBalanceCell.vue
@@ -0,0 +1,168 @@
+<template>
+  <div v-if="visible" class="space-y-1">
+    <div class="flex flex-wrap items-center gap-1.5">
+      <button
+        type="button"
+        :class="[
+          'inline-flex items-center gap-0.5 rounded px-1.5 py-0.5 text-[10px] font-medium transition-colors hover:bg-gray-100 disabled:cursor-not-allowed disabled:opacity-50 dark:hover:bg-dark-600',
+          platformTextClass(account.platform)
+        ]"
+        :disabled="loading"
+        @click="handleProbe"
+      >
+        <svg
+          class="h-2.5 w-2.5"
+          :class="{ 'animate-spin': loading }"
+          fill="none"
+          stroke="currentColor"
+          viewBox="0 0 24 24"
+        >
+          <path
+            stroke-linecap="round"
+            stroke-linejoin="round"
+            stroke-width="2"
+            d="M4 4v5h.582m15.356 2A8.001 8.001 0 004.582 9m0 0H9m11 11v-5h-.581m0 0a8.003 8.003 0 01-15.357-2m15.357 2H15"
+          />
+        </svg>
+        {{ balanceLabel }}
+      </button>
+
+      <!-- Low balance badge (reactive 402/429 marker or probe-detected) -->
+      <span
+        v-if="balanceLow"
+        class="inline-flex items-center rounded bg-red-100 px-1 py-0.5 text-[10px] font-medium text-red-700 dark:bg-red-900/30 dark:text-red-300"
+      >
+        {{ t('admin.accounts.cnProviders.balanceLow') }}
+      </span>
+    </div>
+
+    <div v-if="error" class="truncate text-[10px] text-red-600 dark:text-red-400" :title="error">
+      {{ truncatedError }}
+    </div>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { computed, ref, watch } from 'vue'
+import { useI18n } from 'vue-i18n'
+import { adminAPI } from '@/api/admin'
+import type { CNProviderBalanceEntry, CNProviderBalanceResult } from '@/api/admin/cnProviders'
+import type { Account } from '@/types'
+import { platformTextClass } from '@/utils/platformColors'
+import { cnBalanceCellVisible } from './credentialsBuilder'
+
+const props = defineProps<{
+  account: Account
+}>()
+
+const { t } = useI18n()
+
+const readMode = (): string => {
+  const mode = props.account.credentials?.account_mode
+  return typeof mode === 'string' ? mode : ''
+}
+
+// 仅 kimi / deepseek payg 账号有公开余额端点（智谱 payg 无）。
+const visible = computed(() => cnBalanceCellVisible(props.account.platform, readMode()))
+
+const loading = ref(false)
+const error = ref<string | null>(null)
+const data = ref<CNProviderBalanceResult | null>(null)
+
+const extraKey = (suffix: string) => `${props.account.platform}_${suffix}`
+
+// 落库快照（后端周期探测/响应式写入 account.Extra）。
+const snapshotBalance = computed(() => {
+  const v = props.account.extra?.[extraKey('balance')]
+  return typeof v === 'number' ? v : null
+})
+const snapshotCurrency = computed(() => {
+  const v = props.account.extra?.[extraKey('balance_currency')]
+  return typeof v === 'string' ? v : ''
+})
+// 多币种快照（后端写 <platform>_balances：[{currency, balance}]，deepseek CNY+USD）。
+const snapshotBalances = computed<CNProviderBalanceEntry[]>(() => {
+  const v = props.account.extra?.[extraKey('balances')]
+  if (!Array.isArray(v)) return []
+  return v.flatMap((item): CNProviderBalanceEntry[] => {
+    if (!item || typeof item !== 'object') return []
+    const { currency, balance } = item as Record<string, unknown>
+    if (typeof currency !== 'string' || typeof balance !== 'number') return []
+    return [{ currency, balance }]
+  })
+})
+const balanceLow = computed(() => props.account.extra?.[extraKey('balance_low')] === true)
+
+// 优先用探测结果，其次落库快照。多币种返回全部明细，否则主币种单条。
+const currentEntries = computed<CNProviderBalanceEntry[]>(() => {
+  if (data.value && data.value.success) {
+    if (data.value.balances && data.value.balances.length > 0) return data.value.balances
+    return [{ currency: data.value.currency || '', balance: data.value.balance }]
+  }
+  if (snapshotBalances.value.length > 0) return snapshotBalances.value
+  if (snapshotBalance.value != null) {
+    return [{ currency: snapshotCurrency.value, balance: snapshotBalance.value }]
+  }
+  return []
+})
+
+const formatEntry = (entry: CNProviderBalanceEntry): string => {
+  const fixed = entry.balance >= 100 ? entry.balance.toFixed(0) : entry.balance.toFixed(2)
+  return `${entry.currency || '¥'} ${fixed}`
+}
+
+const balanceLabel = computed(() => {
+  if (currentEntries.value.length === 0) {
+    return t('admin.accounts.grokBalance')?.trim() || 'Balance'
+  }
+  return currentEntries.value.map(formatEntry).join(' · ')
+})
+
+const extractErrorMessage = (e: unknown): string => {
+  const err = e as {
+    message?: string
+    reason?: string
+    response?: { data?: { message?: string; error?: string } }
+  }
+  return (
+    err?.message ||
+    err?.reason ||
+    err?.response?.data?.message ||
+    err?.response?.data?.error ||
+    t('common.error')
+  )
+}
+
+const truncatedError = computed(() => {
+  if (!error.value) return ''
+  return error.value.length > 80 ? `${error.value.slice(0, 80)}...` : error.value
+})
+
+const handleProbe = async () => {
+  if (loading.value) return
+  loading.value = true
+  error.value = null
+  try {
+    const result = await adminAPI.cnProviders.queryBalance(props.account.id)
+    // 失败时保留快照展示（仅显示错误行），成功才覆盖。
+    if (result.success) {
+      data.value = result
+    } else {
+      error.value = result.error || t('common.error')
+    }
+  } catch (e) {
+    error.value = extractErrorMessage(e)
+  } finally {
+    loading.value = false
+  }
+}
+
+watch(
+  () => props.account.id,
+  () => {
+    data.value = null
+    error.value = null
+    loading.value = false
+  }
+)
+</script>
diff --git a/frontend/src/components/account/CNProviderQuotaCell.vue b/frontend/src/components/account/CNProviderQuotaCell.vue
new file mode 100644
index 000000000000..9871a8b48bda
--- /dev/null
+++ b/frontend/src/components/account/CNProviderQuotaCell.vue
@@ -0,0 +1,219 @@
+<template>
+  <div v-if="visible" class="space-y-1">
+    <div class="flex flex-wrap items-center gap-1.5">
+      <button
+        type="button"
+        :class="[
+          'inline-flex items-center gap-0.5 rounded px-1.5 py-0.5 text-[10px] font-medium transition-colors hover:bg-gray-100 disabled:cursor-not-allowed disabled:opacity-50 dark:hover:bg-dark-600',
+          platformTextClass(account.platform)
+        ]"
+        :disabled="loading"
+        :title="t('admin.accounts.cnProviders.probeTooltip')"
+        @click="handleProbe()"
+      >
+        <svg
+          class="h-2.5 w-2.5"
+          :class="{ 'animate-spin': loading }"
+          fill="none"
+          stroke="currentColor"
+          viewBox="0 0 24 24"
+        >
+          <path
+            stroke-linecap="round"
+            stroke-linejoin="round"
+            stroke-width="2"
+            d="M4 4v5h.582m15.356 2A8.001 8.001 0 004.582 9m0 0H9m11 11v-5h-.581m0 0a8.003 8.003 0 01-15.357-2m15.357 2H15"
+          />
+        </svg>
+        {{ t('admin.accounts.cnProviders.window5h') }}/{{ t('admin.accounts.cnProviders.windowWeekly') }}
+      </button>
+    </div>
+
+    <!-- Tier rows: 5h + weekly utilization bars -->
+    <div v-if="data?.success && data.tiers?.length" class="space-y-1">
+      <div v-for="tier in data.tiers" :key="tier.window" class="flex items-center gap-1.5 text-[10px]">
+        <span class="w-10 shrink-0 text-gray-500 dark:text-gray-400">{{ windowLabel(tier.window) }}</span>
+        <div class="h-1.5 w-16 shrink-0 overflow-hidden rounded-full bg-gray-200 dark:bg-dark-600">
+          <div
+            class="h-full rounded-full transition-all"
+            :class="utilizationColor(tier.used_percent)"
+            :style="{ width: `${Math.min(100, Math.max(0, tier.used_percent))}%` }"
+          />
+        </div>
+        <span :class="['shrink-0 font-medium', utilizationTextColor(tier.used_percent)]">
+          {{ Math.round(tier.used_percent) }}%
+        </span>
+        <span v-if="tier.reset_at" class="truncate text-gray-400 dark:text-gray-500" :title="tier.reset_at">
+          · {{ formatReset(tier.reset_at) }}
+        </span>
+      </div>
+    </div>
+
+    <div v-if="error" class="truncate text-[10px] text-red-600 dark:text-red-400" :title="error">
+      {{ truncatedError }}
+    </div>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { computed, onMounted, ref, watch } from 'vue'
+import { useI18n } from 'vue-i18n'
+import { adminAPI } from '@/api/admin'
+import type { CNProviderQuotaProbeResult } from '@/api/admin/cnProviders'
+import type { Account } from '@/types'
+import { platformTextClass } from '@/utils/platformColors'
+import { cnQuotaCellVisible } from './credentialsBuilder'
+
+const props = defineProps<{
+  account: Account
+}>()
+
+const { t } = useI18n()
+
+const readMode = (): string => {
+  const mode = props.account.credentials?.account_mode
+  return typeof mode === 'string' ? mode : ''
+}
+
+const visible = computed(() => cnQuotaCellVisible(props.account.platform, readMode()))
+
+const loading = ref(false)
+const error = ref<string | null>(null)
+const data = ref<CNProviderQuotaProbeResult | null>(null)
+
+// 后端周期任务/手动探测写入的 extra 快照键（<provider>_ 前缀，与后端
+// cnQuotaExtraUpdates 对齐）。页面加载即有数据，无需等待探测。
+const SNAPSHOT_STALE_MS = 15 * 60 * 1000
+
+// 自动探测去抖窗口与最近一次自动探测时间（模块级，跨实例共享）。
+const AUTO_PROBE_DEBOUNCE_MS = 5 * 60 * 1000
+const lastAutoProbeAt = new Map<number, number>()
+
+const readExtraNumber = (key: string): number | null => {
+  const v = (props.account.extra as Record<string, unknown> | undefined)?.[key]
+  return typeof v === 'number' && Number.isFinite(v) ? v : null
+}
+
+const readExtraString = (key: string): string => {
+  const v = (props.account.extra as Record<string, unknown> | undefined)?.[key]
+  return typeof v === 'string' ? v : ''
+}
+
+// 从持久化快照构造展示数据（缺少 5h/weekly 两档键时返回 null）。
+const snapshotData = computed<CNProviderQuotaProbeResult | null>(() => {
+  const platform = props.account.platform
+  const used5h = readExtraNumber(`${platform}_5h_used_percent`)
+  const usedWeekly = readExtraNumber(`${platform}_weekly_used_percent`)
+  if (used5h == null && usedWeekly == null) return null
+  const tiers: CNProviderQuotaProbeResult['tiers'] = []
+  if (used5h != null) {
+    tiers.push({ window: '5h', used_percent: used5h, reset_at: readExtraString(`${platform}_5h_reset_at`) || undefined })
+  }
+  if (usedWeekly != null) {
+    tiers.push({ window: 'weekly', used_percent: usedWeekly, reset_at: readExtraString(`${platform}_weekly_reset_at`) || undefined })
+  }
+  return { success: true, tiers } as CNProviderQuotaProbeResult
+})
+
+// 快照是否过期（无更新时间或超过 staleness 窗口）→ 挂载时需要自动探测。
+const snapshotIsStale = computed(() => {
+  const updatedAt = readExtraString(`${props.account.platform}_usage_updated_at`)
+  if (!updatedAt) return true
+  const ts = new Date(updatedAt).getTime()
+  return Number.isNaN(ts) || Date.now() - ts > SNAPSHOT_STALE_MS
+})
+
+// 挂载时：先用持久化快照渲染；快照缺失或过期再自动探测一次（失败显示错误，
+// 避免静默失败导致单元格空白无提示）。
+onMounted(() => {
+  if (!visible.value) return
+  data.value = snapshotData.value
+  if (!snapshotIsStale.value) return
+  // 模块级去抖：列表页每行一个实例，翻页/筛选/刷新会重复挂载；同一账号
+  // 短时间内已自动探测过则跳过，避免对上游形成探测风暴。
+  const last = lastAutoProbeAt.get(props.account.id) ?? 0
+  if (Date.now() - last < AUTO_PROBE_DEBOUNCE_MS) return
+  lastAutoProbeAt.set(props.account.id, Date.now())
+  handleProbe()
+})
+
+const extractErrorMessage = (e: unknown): string => {
+  const err = e as {
+    message?: string
+    reason?: string
+    response?: { data?: { message?: string; error?: string } }
+  }
+  return (
+    err?.message ||
+    err?.reason ||
+    err?.response?.data?.message ||
+    err?.response?.data?.error ||
+    t('common.error')
+  )
+}
+
+const truncatedError = computed(() => {
+  if (!error.value) return ''
+  return error.value.length > 80 ? `${error.value.slice(0, 80)}...` : error.value
+})
+
+const windowLabel = (window: string) =>
+  window === 'weekly'
+    ? t('admin.accounts.cnProviders.windowWeekly')
+    : t('admin.accounts.cnProviders.window5h')
+
+const utilizationColor = (pct: number) => {
+  if (pct >= 90) return 'bg-red-500'
+  if (pct >= 75) return 'bg-amber-500'
+  return 'bg-emerald-500'
+}
+
+const utilizationTextColor = (pct: number) => {
+  if (pct >= 90) return 'text-red-600 dark:text-red-400'
+  if (pct >= 75) return 'text-amber-600 dark:text-amber-400'
+  return 'text-emerald-600 dark:text-emerald-400'
+}
+
+// 重置时间相对/绝对简短显示。
+const formatReset = (iso: string) => {
+  const d = new Date(iso)
+  if (isNaN(d.getTime())) return iso
+  const now = Date.now()
+  const diffMs = d.getTime() - now
+  if (diffMs <= 0) return t('admin.accounts.cnProviders.resetSoon')
+  if (diffMs < 3_600_000) return `${Math.max(1, Math.round(diffMs / 60_000))}m`
+  const hours = Math.round(diffMs / 3_600_000)
+  if (hours < 48) return `${hours}h`
+  const mm = String(d.getMonth() + 1).padStart(2, '0')
+  const dd = String(d.getDate()).padStart(2, '0')
+  return `${mm}-${dd}`
+}
+
+const handleProbe = async () => {
+  if (loading.value) return
+  loading.value = true
+  error.value = null
+  try {
+    const result = await adminAPI.cnProviders.queryQuota(props.account.id)
+    // 失败时保留已渲染的快照条形图（仅显示错误行），成功才覆盖。
+    if (result.success) {
+      data.value = result
+    } else {
+      error.value = result.error || t('common.error')
+    }
+  } catch (e) {
+    error.value = extractErrorMessage(e)
+  } finally {
+    loading.value = false
+  }
+}
+
+watch(
+  () => props.account.id,
+  () => {
+    data.value = null
+    error.value = null
+    loading.value = false
+  }
+)
+</script>
diff --git a/frontend/src/components/account/CnBaseUrlPresets.vue b/frontend/src/components/account/CnBaseUrlPresets.vue
new file mode 100644
index 000000000000..5a5172555a62
--- /dev/null
+++ b/frontend/src/components/account/CnBaseUrlPresets.vue
@@ -0,0 +1,53 @@
+<template>
+  <div class="flex flex-wrap gap-2">
+    <button
+      v-for="preset in presets"
+      :key="preset.mode + ':' + preset.protocol + ':' + preset.url"
+      type="button"
+      data-testid="cn-base-url-preset"
+      :class="[
+        'rounded-lg px-3 py-1 text-xs transition-colors',
+        isActive(preset)
+          ? 'bg-primary-100 text-primary-700 dark:bg-primary-900/30 dark:text-primary-300'
+          : 'bg-gray-100 text-gray-700 hover:bg-primary-50 hover:text-primary-700 dark:bg-dark-600 dark:text-gray-300 dark:hover:bg-primary-900/30 dark:hover:text-primary-400'
+      ]"
+      @click="emit('select', preset)"
+    >
+      {{ preset.label }} ({{ displayUrl(preset.url) }})
+    </button>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { computed } from 'vue'
+import { CN_BASE_URL_PRESETS, type CnBaseUrlPreset } from './credentialsBuilder'
+
+// 国产供应商快捷端点：点击把预设地址（及对应账号类型/协议）回填到调用方。
+// 与 Grok 预设一致，仅作快速填充，输入框仍接受任意第三方转发地址。
+// 传入 protocol 时只显示该协议档的预设（协议 × 账号类型正交分档）。
+const props = defineProps<{
+  platform: 'kimi' | 'zhipu' | 'deepseek'
+  /** 当前已选账号类型，用于过滤和高亮匹配的预设 */
+  mode?: 'payg' | 'coding'
+  /** 当前已选 API 协议，用于过滤和高亮匹配的预设 */
+  protocol?: 'chat_completions' | 'anthropic' | 'responses'
+  /** 当前输入框中的 base url，用于高亮完全匹配项 */
+  currentUrl?: string
+}>()
+
+const emit = defineEmits<{
+  (e: 'select', preset: CnBaseUrlPreset): void
+}>()
+
+const presets = computed(() => {
+  const all = CN_BASE_URL_PRESETS[props.platform] ?? []
+  // 只按协议过滤：同协议下 payg/coding 两档都展示，点击即同时切换账号类型。
+  if (props.protocol == null) return all
+  return all.filter(p => p.protocol === props.protocol)
+})
+
+const isActive = (preset: CnBaseUrlPreset) =>
+  (props.mode != null && preset.mode === props.mode) || preset.url === props.currentUrl
+
+const displayUrl = (url: string) => url.replace(/^https?:\/\//i, '')
+</script>
diff --git a/frontend/src/components/account/CreateAccountModal.vue b/frontend/src/components/account/CreateAccountModal.vue
index f93b71e57436..26b4efb2b7ac 100644
--- a/frontend/src/components/account/CreateAccountModal.vue
+++ b/frontend/src/components/account/CreateAccountModal.vue
@@ -161,6 +161,48 @@
             Grok
           </button>
         </div>
+        <!-- CN providers row: Kimi / Zhipu GLM / DeepSeek -->
+        <div class="mt-2 flex flex-wrap rounded-lg bg-gray-100 p-1 dark:bg-dark-700">
+          <button
+            type="button"
+            @click="selectCNPlatform('kimi')"
+            :class="[
+              'flex flex-1 items-center justify-center gap-2 rounded-md px-4 py-2.5 text-sm font-medium transition-all',
+              form.platform === 'kimi'
+                ? 'bg-white text-pink-600 shadow-sm dark:bg-dark-600 dark:text-pink-400'
+                : 'text-gray-600 hover:text-gray-900 dark:text-gray-400 dark:hover:text-gray-200'
+            ]"
+          >
+            <PlatformIcon platform="kimi" size="sm" />
+            Kimi
+          </button>
+          <button
+            type="button"
+            @click="selectCNPlatform('zhipu')"
+            :class="[
+              'flex flex-1 items-center justify-center gap-2 rounded-md px-4 py-2.5 text-sm font-medium transition-all',
+              form.platform === 'zhipu'
+                ? 'bg-white text-indigo-600 shadow-sm dark:bg-dark-600 dark:text-indigo-400'
+                : 'text-gray-600 hover:text-gray-900 dark:text-gray-400 dark:hover:text-gray-200'
+            ]"
+          >
+            <PlatformIcon platform="zhipu" size="sm" />
+            Zhipu GLM
+          </button>
+          <button
+            type="button"
+            @click="selectCNPlatform('deepseek')"
+            :class="[
+              'flex flex-1 items-center justify-center gap-2 rounded-md px-4 py-2.5 text-sm font-medium transition-all',
+              form.platform === 'deepseek'
+                ? 'bg-white text-teal-600 shadow-sm dark:bg-dark-600 dark:text-teal-400'
+                : 'text-gray-600 hover:text-gray-900 dark:text-gray-400 dark:hover:text-gray-200'
+            ]"
+          >
+            <PlatformIcon platform="deepseek" size="sm" />
+            DeepSeek
+          </button>
+        </div>
       </div>
 
       <!-- Account Type Selection (Anthropic) -->
@@ -411,6 +453,100 @@
         </div>
       </div>
 
+      <!-- Account Mode Selection (Kimi / Zhipu / DeepSeek) -->
+      <div v-if="isCNPlatform">
+        <label class="input-label">{{ t('admin.accounts.cnProviders.accountMode.title') }}</label>
+        <div class="mt-2 grid grid-cols-1 gap-3 sm:grid-cols-2" data-tour="account-form-mode">
+          <!-- Pay-as-you-go (token balance) -->
+          <button
+            type="button"
+            @click="accountMode = 'payg'"
+            :class="[
+              'flex items-center gap-3 rounded-lg border-2 p-3 text-left transition-all',
+              accountMode === 'payg'
+                ? cnAccentActiveClass
+                : 'border-gray-200 hover:border-gray-400 dark:border-dark-600 dark:hover:border-gray-600'
+            ]"
+          >
+            <div
+              :class="[
+                'flex h-8 w-8 shrink-0 items-center justify-center rounded-lg',
+                accountMode === 'payg'
+                  ? cnAccentIconClass
+                  : 'bg-gray-100 text-gray-500 dark:bg-dark-600 dark:text-gray-400'
+              ]"
+            >
+              <Icon name="creditCard" size="sm" />
+            </div>
+            <div>
+              <span class="block text-sm font-medium text-gray-900 dark:text-white">{{ t('admin.accounts.cnProviders.accountMode.payg') }}</span>
+              <span class="text-xs text-gray-500 dark:text-gray-400">{{ t('admin.accounts.cnProviders.accountMode.paygDesc') }}</span>
+            </div>
+          </button>
+          <!-- Coding Plan (kimi / zhipu only — DeepSeek has no coding plan) -->
+          <button
+            v-if="form.platform !== 'deepseek'"
+            type="button"
+            @click="accountMode = 'coding'"
+            :class="[
+              'flex items-center gap-3 rounded-lg border-2 p-3 text-left transition-all',
+              accountMode === 'coding'
+                ? cnAccentActiveClass
+                : 'border-gray-200 hover:border-gray-400 dark:border-dark-600 dark:hover:border-gray-600'
+            ]"
+          >
+            <div
+              :class="[
+                'flex h-8 w-8 shrink-0 items-center justify-center rounded-lg',
+                accountMode === 'coding'
+                  ? cnAccentIconClass
+                  : 'bg-gray-100 text-gray-500 dark:bg-dark-600 dark:text-gray-400'
+              ]"
+            >
+              <Icon name="bolt" size="sm" />
+            </div>
+            <div>
+              <span class="block text-sm font-medium text-gray-900 dark:text-white">{{ t('admin.accounts.cnProviders.accountMode.coding') }}</span>
+              <span class="text-xs text-gray-500 dark:text-gray-400">{{ t('admin.accounts.cnProviders.accountMode.codingDesc') }}</span>
+            </div>
+          </button>
+        </div>
+      </div>
+
+      <!-- API Protocol Selection (Kimi / Zhipu / DeepSeek) -->
+      <div v-if="isCNPlatform" class="mt-4">
+        <label class="input-label">{{ t('admin.accounts.cnProviders.apiProtocol.title') }}</label>
+        <div class="mt-2 grid grid-cols-1 gap-3 sm:grid-cols-3">
+          <button
+            v-for="opt in cnProtocolOptions"
+            :key="opt.value"
+            type="button"
+            @click="apiProtocol = opt.value"
+            :class="[
+              'flex items-center gap-3 rounded-lg border-2 p-3 text-left transition-all',
+              apiProtocol === opt.value
+                ? cnAccentActiveClass
+                : 'border-gray-200 hover:border-gray-400 dark:border-dark-600 dark:hover:border-gray-600'
+            ]"
+          >
+            <div
+              :class="[
+                'flex h-8 w-8 shrink-0 items-center justify-center rounded-lg',
+                apiProtocol === opt.value
+                  ? cnAccentIconClass
+                  : 'bg-gray-100 text-gray-500 dark:bg-dark-600 dark:text-gray-400'
+              ]"
+            >
+              <Icon :name="opt.value === 'anthropic' ? 'sparkles' : opt.value === 'responses' ? 'terminal' : 'chat'" size="sm" />
+            </div>
+            <div>
+              <span class="block text-sm font-medium text-gray-900 dark:text-white">{{ t(`admin.accounts.cnProviders.apiProtocol.${opt.labelKey}`) }}</span>
+              <span class="text-xs text-gray-500 dark:text-gray-400">{{ t(`admin.accounts.cnProviders.apiProtocol.${opt.labelKey}Desc`) }}</span>
+            </div>
+          </button>
+        </div>
+      </div>
+
       <!-- Account Type Selection (Gemini) -->
       <div v-if="form.platform === 'gemini'">
         <div class="flex items-center justify-between">
@@ -1120,15 +1256,7 @@
             v-model="apiKeyBaseUrl"
             type="text"
             class="input"
-            :placeholder="
-              form.platform === 'openai'
-                ? 'https://api.openai.com'
-                : form.platform === 'gemini'
-                  ? 'https://generativelanguage.googleapis.com'
-                  : form.platform === 'grok'
-                    ? 'https://api.x.ai/v1'
-                    : 'https://api.anthropic.com'
-            "
+            :placeholder="apiKeyBaseUrlPlaceholder"
           />
           <p v-if="baseUrlHint" class="input-hint">{{ baseUrlHint }}</p>
           <GrokBaseUrlPresets
@@ -1136,6 +1264,15 @@
             class="mt-2"
             @select="apiKeyBaseUrl = $event"
           />
+          <CnBaseUrlPresets
+            v-if="isCNPlatform"
+            class="mt-2"
+            :platform="cnPresetPlatform"
+            :mode="accountMode"
+            :protocol="apiProtocol"
+            :current-url="apiKeyBaseUrl"
+            @select="onCnPresetSelect"
+          />
         </div>
         <div>
           <label class="input-label">{{ t('admin.accounts.apiKeyRequired') }}</label>
@@ -1144,15 +1281,7 @@
             type="password"
             required
             class="input font-mono"
-            :placeholder="
-              form.platform === 'openai'
-                ? 'sk-proj-...'
-                : form.platform === 'gemini'
-                  ? 'AIza...'
-                  : form.platform === 'grok'
-                    ? 'xai-...'
-                    : 'sk-ant-...'
-            "
+            :placeholder="apiKeyValuePlaceholder"
           />
           <p v-if="apiKeyHint" class="input-hint">{{ apiKeyHint }}</p>
         </div>
@@ -3609,13 +3738,17 @@ import ModelWhitelistSelector from '@/components/account/ModelWhitelistSelector.
 import QuotaLimitCard from '@/components/account/QuotaLimitCard.vue'
 import Toggle from '@/components/common/Toggle.vue'
 import GrokBaseUrlPresets from '@/components/account/GrokBaseUrlPresets.vue'
+import CnBaseUrlPresets from '@/components/account/CnBaseUrlPresets.vue'
 import HeaderOverrideEditor from '@/components/account/HeaderOverrideEditor.vue'
 import {
   applyAntigravityProjectID,
   applyHeaderOverride,
   applyInterceptWarmup,
+  defaultCNBaseUrl,
   isHeaderOverrideCapable,
   validateHeaderOverrideRows,
+  type CnAccountMode,
+  type CnApiProtocol,
   type HeaderOverrideRow
 } from '@/components/account/credentialsBuilder'
 import { formatDateTimeLocalInput, parseDateTimeLocalInput } from '@/utils/format'
@@ -3674,6 +3807,42 @@ const apiKeyHint = computed(() => {
   return t('admin.accounts.apiKeyHint')
 })
 
+// Base URL / API Key 占位符：国产供应商随账号类型变化。
+const apiKeyBaseUrlPlaceholder = computed(() => {
+  if (isCNPlatform.value) {
+    return defaultCNBaseUrl(form.platform, accountMode.value, apiProtocol.value) || 'https://api.example.com'
+  }
+  switch (form.platform) {
+    case 'openai':
+      return 'https://api.openai.com'
+    case 'gemini':
+      return 'https://generativelanguage.googleapis.com'
+    case 'grok':
+      return 'https://api.x.ai/v1'
+    default:
+      return 'https://api.anthropic.com'
+  }
+})
+
+const apiKeyValuePlaceholder = computed(() => {
+  switch (form.platform) {
+    case 'openai':
+      return 'sk-proj-...'
+    case 'gemini':
+      return 'AIza...'
+    case 'grok':
+      return 'xai-...'
+    case 'kimi':
+      return 'sk-...'
+    case 'zhipu':
+      return '<api-key>.<secret>'
+    case 'deepseek':
+      return 'sk-...'
+    default:
+      return 'sk-ant-...'
+  }
+})
+
 interface Props {
   show: boolean
   proxies: Proxy[]
@@ -3753,6 +3922,86 @@ const apiKeyBaseUrl = ref('https://api.anthropic.com')
 const apiKeyValue = ref('')
 const upstreamBillingAutoProbeEnabled = ref(true)
 
+// ── 国产供应商（Kimi / Zhipu / DeepSeek）账号类型、API 协议与端点 ──
+const accountMode = ref<CnAccountMode>('payg')
+// API 协议决定转发端点与格式：cc=现有转换链，anthropic=原生直通（Claude Code），
+// responses=deepseek 原生 Responses 端点（Codex）。与账号类型正交。
+const apiProtocol = ref<CnApiProtocol>('chat_completions')
+const isCNPlatform = computed(
+  () => form.platform === 'kimi' || form.platform === 'zhipu' || form.platform === 'deepseek'
+)
+// CnBaseUrlPresets 的 platform prop 是平台字面量联合类型，模板里不能写
+// `as` 断言（其中的 `|` 会被 eslint 误判为 Vue2 filter 语法），经此 computed 传递。
+const cnPresetPlatform = computed<'kimi' | 'zhipu' | 'deepseek'>(() => {
+  if (form.platform === 'kimi' || form.platform === 'zhipu' || form.platform === 'deepseek') {
+    return form.platform
+  }
+  return 'kimi'
+})
+// 当前平台可选的协议档（responses 仅 deepseek）。
+const cnProtocolOptions = computed<Array<{ value: CnApiProtocol; labelKey: string }>>(() => {
+  const opts: Array<{ value: CnApiProtocol; labelKey: string }> = [
+    { value: 'chat_completions', labelKey: 'chatCompletions' },
+    { value: 'anthropic', labelKey: 'anthropic' }
+  ]
+  if (form.platform === 'deepseek') {
+    opts.push({ value: 'responses', labelKey: 'responses' })
+  }
+  return opts
+})
+// 当前选中平台的品牌色（选中卡片描边 / 图标底色），与 platformColors 取色一致。
+const cnAccentActiveClass = computed(() => {
+  switch (form.platform) {
+    case 'kimi':
+      return 'border-pink-500 bg-pink-50 dark:bg-pink-900/20'
+    case 'zhipu':
+      return 'border-indigo-500 bg-indigo-50 dark:bg-indigo-900/20'
+    case 'deepseek':
+      return 'border-teal-500 bg-teal-50 dark:bg-teal-900/20'
+    default:
+      return 'border-primary-500 bg-primary-50 dark:bg-primary-900/20'
+  }
+})
+const cnAccentIconClass = computed(() => {
+  switch (form.platform) {
+    case 'kimi':
+      return 'bg-pink-500 text-white'
+    case 'zhipu':
+      return 'bg-indigo-500 text-white'
+    case 'deepseek':
+      return 'bg-teal-500 text-white'
+    default:
+      return 'bg-primary-500 text-white'
+  }
+})
+// 切换国产供应商平台：强制 apikey 类型，deepseek 无 coding 套餐故锁定 payg，
+// 协议回落 chat_completions，并把 base url 重置为该平台默认端点。
+function selectCNPlatform(platform: 'kimi' | 'zhipu' | 'deepseek') {
+  form.platform = platform
+  form.type = 'apikey'
+  accountCategory.value = 'apikey'
+  apiProtocol.value = 'chat_completions'
+  if (platform === 'deepseek') {
+    accountMode.value = 'payg'
+  }
+  apiKeyBaseUrl.value = defaultCNBaseUrl(platform, accountMode.value, apiProtocol.value)
+}
+// 账号类型 / 协议变更时同步默认 base url。
+watch(accountMode, (mode) => {
+  if (!isCNPlatform.value) return
+  apiKeyBaseUrl.value = defaultCNBaseUrl(form.platform, mode, apiProtocol.value)
+})
+watch(apiProtocol, (protocol) => {
+  if (!isCNPlatform.value) return
+  apiKeyBaseUrl.value = defaultCNBaseUrl(form.platform, accountMode.value, protocol)
+})
+// 点击预设端点：同时回填 base url、账号类型与协议。
+function onCnPresetSelect(preset: { mode: CnAccountMode; protocol: CnApiProtocol; url: string }) {
+  accountMode.value = preset.mode
+  apiProtocol.value = preset.protocol
+  apiKeyBaseUrl.value = preset.url
+}
+
 const syncPreviewCredentials = computed(() => {
   if (!apiKeyValue.value) return undefined
   return {
@@ -4248,14 +4497,18 @@ watch(
   () => form.platform,
   (newPlatform) => {
     // Reset base URL based on platform
-    apiKeyBaseUrl.value =
-      (newPlatform === 'openai')
-        ? 'https://api.openai.com'
-        : newPlatform === 'gemini'
-          ? 'https://generativelanguage.googleapis.com'
-          : newPlatform === 'grok'
-            ? 'https://api.x.ai/v1'
-            : 'https://api.anthropic.com'
+    if (newPlatform === 'kimi' || newPlatform === 'zhipu' || newPlatform === 'deepseek') {
+      apiKeyBaseUrl.value = defaultCNBaseUrl(newPlatform, accountMode.value, apiProtocol.value)
+    } else {
+      apiKeyBaseUrl.value =
+        (newPlatform === 'openai')
+          ? 'https://api.openai.com'
+          : newPlatform === 'gemini'
+            ? 'https://generativelanguage.googleapis.com'
+            : newPlatform === 'grok'
+              ? 'https://api.x.ai/v1'
+              : 'https://api.anthropic.com'
+    }
     // Clear model-related settings
     allowedModels.value = []
     modelMappings.value = []
@@ -4696,6 +4949,8 @@ const resetForm = () => {
   form.expires_at = null
   accountCategory.value = 'oauth-based'
   addMethod.value = 'oauth'
+  accountMode.value = 'payg'
+  apiProtocol.value = 'chat_completions'
   apiKeyBaseUrl.value = 'https://api.anthropic.com'
   apiKeyValue.value = ''
   upstreamBillingAutoProbeEnabled.value = true
@@ -5158,6 +5413,20 @@ const handleSubmit = async () => {
     credentials.tier_id = geminiTierAIStudio.value
   }
 
+  // 国产供应商：账号模式 + 协议 + 对应端点写入凭据；后端按 account_mode 路由
+  // 额度/余额探测，按 api_protocol 路由转发端点与格式。注意 CN apikey 走本函数
+  // 的通用路径（直接 doCreateAccount），不经过 createAccountAndFinish。
+  if (form.platform === 'kimi' || form.platform === 'zhipu' || form.platform === 'deepseek') {
+    credentials.account_mode = accountMode.value
+    credentials.api_protocol = apiProtocol.value
+    const resolvedCNBase = (
+      apiKeyBaseUrl.value.trim() || defaultCNBaseUrl(form.platform, accountMode.value, apiProtocol.value)
+    ).trim()
+    if (resolvedCNBase) {
+      credentials.base_url = resolvedCNBase
+    }
+  }
+
   // Add model mapping if configured（OpenAI 开启自动透传时不应用）
   if (!isOpenAIModelRestrictionDisabled.value) {
     const modelMapping = buildModelMappingObject(modelRestrictionMode.value, allowedModels.value, modelMappings.value)
diff --git a/frontend/src/components/account/EditAccountModal.vue b/frontend/src/components/account/EditAccountModal.vue
index 205d004b8247..1a5654bd0fdb 100644
--- a/frontend/src/components/account/EditAccountModal.vue
+++ b/frontend/src/components/account/EditAccountModal.vue
@@ -52,6 +52,57 @@
             class="mt-2"
             @select="editBaseUrl = $event"
           />
+          <CnBaseUrlPresets
+            v-if="isCNApiKeyAccount"
+            class="mt-2"
+            :platform="cnPresetPlatform"
+            :mode="editAccountMode"
+            :protocol="editApiProtocol"
+            :current-url="editBaseUrl"
+            @select="onCnPresetSelect"
+          />
+        </div>
+        <!-- Account Mode Selection (CN providers) -->
+        <div v-if="isCNApiKeyAccount">
+          <label class="input-label">{{ t('admin.accounts.cnProviders.accountMode.title') }}</label>
+          <div class="mt-2 flex flex-wrap gap-2">
+            <button
+              v-for="opt in cnAccountModeOptions"
+              :key="opt.value"
+              type="button"
+              :class="[
+                'rounded-lg border-2 px-3 py-1.5 text-xs transition-all',
+                editAccountMode === opt.value
+                  ? 'border-primary-500 bg-primary-50 font-medium text-primary-700 dark:bg-primary-900/30 dark:text-primary-300'
+                  : 'border-gray-200 text-gray-700 hover:border-gray-400 dark:border-dark-600 dark:text-gray-300 dark:hover:border-gray-600'
+              ]"
+              @click="editAccountMode = opt.value"
+            >
+              {{ t(`admin.accounts.cnProviders.accountMode.${opt.labelKey}`) }}
+            </button>
+          </div>
+          <p class="input-hint">{{ t(`admin.accounts.cnProviders.accountMode.${editAccountMode}Desc`) }}</p>
+        </div>
+        <!-- API Protocol Selection (CN providers) -->
+        <div v-if="isCNApiKeyAccount">
+          <label class="input-label">{{ t('admin.accounts.cnProviders.apiProtocol.title') }}</label>
+          <div class="mt-2 flex flex-wrap gap-2">
+            <button
+              v-for="opt in cnProtocolOptions"
+              :key="opt.value"
+              type="button"
+              :class="[
+                'rounded-lg border-2 px-3 py-1.5 text-xs transition-all',
+                editApiProtocol === opt.value
+                  ? 'border-primary-500 bg-primary-50 font-medium text-primary-700 dark:bg-primary-900/30 dark:text-primary-300'
+                  : 'border-gray-200 text-gray-700 hover:border-gray-400 dark:border-dark-600 dark:text-gray-300 dark:hover:border-gray-600'
+              ]"
+              @click="editApiProtocol = opt.value"
+            >
+              {{ t(`admin.accounts.cnProviders.apiProtocol.${opt.labelKey}`) }}
+            </button>
+          </div>
+          <p class="input-hint">{{ t(`admin.accounts.cnProviders.apiProtocol.${cnProtocolDescKey}Desc`) }}</p>
         </div>
         <div>
           <label class="input-label">{{ t('admin.accounts.apiKey') }}</label>
@@ -2710,7 +2761,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, reactive, computed, watch } from 'vue'
+import { ref, reactive, computed, watch, nextTick } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { useAppStore } from '@/stores/app'
 import { useAuthStore } from '@/stores/auth'
@@ -2737,6 +2788,7 @@ import GroupSelector from '@/components/common/GroupSelector.vue'
 import ModelWhitelistSelector from '@/components/account/ModelWhitelistSelector.vue'
 import QuotaLimitCard from '@/components/account/QuotaLimitCard.vue'
 import GrokBaseUrlPresets from '@/components/account/GrokBaseUrlPresets.vue'
+import CnBaseUrlPresets from '@/components/account/CnBaseUrlPresets.vue'
 import HeaderOverrideEditor from '@/components/account/HeaderOverrideEditor.vue'
 import OllamaCloudUsageSettings from '@/components/account/OllamaCloudUsageSettings.vue'
 import {
@@ -2750,8 +2802,11 @@ import {
   isHeaderOverrideCapable,
   splitHeaderOverridesObject,
   validateHeaderOverrideRows,
+  defaultCNBaseUrl,
   HEADER_OVERRIDE_ENABLED_CREDENTIAL_KEY,
   HEADER_OVERRIDES_CREDENTIAL_KEY,
+  type CnAccountMode,
+  type CnApiProtocol,
   type HeaderOverrideRow
 } from '@/components/account/credentialsBuilder'
 import { formatDateTime, formatDateTimeLocalInput, parseDateTimeLocalInput } from '@/utils/format'
@@ -2829,6 +2884,78 @@ interface TempUnschedRuleForm {
 const submitting = ref(false)
 const editBaseUrl = ref('https://api.anthropic.com')
 const editApiKey = ref('')
+
+// ── 国产供应商（Kimi / Zhipu / DeepSeek）account_mode / api_protocol 编辑 ──
+// account_mode 决定额度/余额监控路径，api_protocol 决定转发端点与格式；
+// 二者均可修正（早期创建的账号可能存错默认值），切换时重置 base_url 预置。
+const isCNApiKeyAccount = computed(
+  () =>
+    props.account?.type === 'apikey' &&
+    (props.account.platform === 'kimi' ||
+      props.account.platform === 'zhipu' ||
+      props.account.platform === 'deepseek')
+)
+// CnBaseUrlPresets 的 platform prop 是平台字面量联合类型，模板里不能写
+// `as` 断言（其中的 `|` 会被 eslint 误判为 Vue2 filter 语法），经此 computed 传递。
+const cnPresetPlatform = computed<'kimi' | 'zhipu' | 'deepseek'>(() => {
+  const platform = props.account?.platform
+  if (platform === 'kimi' || platform === 'zhipu' || platform === 'deepseek') {
+    return platform
+  }
+  return 'kimi'
+})
+const editApiProtocol = ref<CnApiProtocol>('chat_completions')
+const editAccountMode = ref<CnAccountMode>('payg')
+// 回填窗口标志：syncFormFromAccount 会同步改写 editAccountMode / editApiProtocol，
+// 而 watcher（pre-flush）在同步代码执行完之后才触发——若不抑制，会把刚恢复的
+// 存储版 base_url（可能是用户自定义/中转地址）覆盖为官方预设并在下次保存时持久化。
+// nextTick 后解除，此后用户主动切换模式/协议仍正常联动重置。
+const syncingForm = ref(false)
+const cnAccountModeOptions = computed<Array<{ value: CnAccountMode; labelKey: 'payg' | 'coding' }>>(
+  () => {
+    // DeepSeek 无 coding 套餐（与创建弹窗一致），仅保留按量付费。
+    if (props.account?.platform === 'deepseek') {
+      return [{ value: 'payg', labelKey: 'payg' }]
+    }
+    return [
+      { value: 'payg', labelKey: 'payg' },
+      { value: 'coding', labelKey: 'coding' }
+    ]
+  }
+)
+const cnProtocolOptions = computed<Array<{ value: CnApiProtocol; labelKey: string }>>(() => {
+  const opts: Array<{ value: CnApiProtocol; labelKey: string }> = [
+    { value: 'chat_completions', labelKey: 'chatCompletions' },
+    { value: 'anthropic', labelKey: 'anthropic' }
+  ]
+  if (props.account?.platform === 'deepseek') {
+    opts.push({ value: 'responses', labelKey: 'responses' })
+  }
+  return opts
+})
+watch(editApiProtocol, (protocol) => {
+  if (!isCNApiKeyAccount.value || syncingForm.value) return
+  editBaseUrl.value = defaultCNBaseUrl(props.account!.platform, editAccountMode.value, protocol)
+})
+watch(editAccountMode, (mode) => {
+  if (!isCNApiKeyAccount.value || syncingForm.value) return
+  // deepseek 无 coding 套餐：防御性回退（UI 已隐藏该选项）。
+  const effectiveMode = props.account!.platform === 'deepseek' && mode === 'coding' ? 'payg' : mode
+  if (effectiveMode !== mode) {
+    editAccountMode.value = effectiveMode
+    return
+  }
+  editBaseUrl.value = defaultCNBaseUrl(props.account!.platform, mode, editApiProtocol.value)
+})
+const cnProtocolDescKey = computed(
+  () => cnProtocolOptions.value.find(o => o.value === editApiProtocol.value)?.labelKey ?? 'chatCompletions'
+)
+// 点击预设端点：回填 base url 与对应模式/协议。
+function onCnPresetSelect(preset: { mode: CnAccountMode; protocol: CnApiProtocol; url: string }) {
+  editAccountMode.value = preset.mode
+  editApiProtocol.value = preset.protocol
+  editBaseUrl.value = preset.url
+}
 // Bedrock credentials
 const editBedrockAccessKeyId = ref('')
 const editBedrockSecretAccessKey = ref('')
@@ -3272,6 +3399,15 @@ const defaultBaseUrl = computed(() => {
   if (props.account?.platform === 'openai') return 'https://api.openai.com'
   if (props.account?.platform === 'gemini') return 'https://generativelanguage.googleapis.com'
   if (props.account?.platform === 'grok') return 'https://api.x.ai/v1'
+  // CN 供应商：按当前模式/协议回落到官方预设（清空输入框提交时使用），
+  // 不能落到 anthropic 默认值（会被当 CC base 拼出错误端点）。
+  if (
+    props.account?.platform === 'kimi' ||
+    props.account?.platform === 'zhipu' ||
+    props.account?.platform === 'deepseek'
+  ) {
+    return defaultCNBaseUrl(props.account.platform, editAccountMode.value, editApiProtocol.value)
+  }
   return 'https://api.anthropic.com'
 })
 
@@ -3381,6 +3517,11 @@ const syncFormFromAccount = (newAccount: Account | null) => {
   if (!newAccount) {
     return
   }
+  // 进入回填窗口：抑制 CN 模式/协议 watcher 联动重置 base_url（见 syncingForm 注释）。
+  syncingForm.value = true
+  void nextTick(() => {
+    syncingForm.value = false
+  })
   antigravityMixedChannelConfirmed.value = false
   showMixedChannelWarning.value = false
   mixedChannelWarningDetails.value = null
@@ -3622,6 +3763,17 @@ const syncFormFromAccount = (newAccount: Account | null) => {
   // Initialize API Key fields for apikey type
   if (newAccount.type === 'apikey' && newAccount.credentials) {
     const credentials = newAccount.credentials as Record<string, unknown>
+    // 国产供应商：读取 account_mode 与 api_protocol 作为可编辑初始值
+    // （编辑弹窗允许修正两者，用于修复早期存错默认值的账号）。
+    if (newAccount.platform === 'kimi' || newAccount.platform === 'zhipu' || newAccount.platform === 'deepseek') {
+      editAccountMode.value = credentials.account_mode === 'coding' ? 'coding' : 'payg'
+      const storedProtocol = credentials.api_protocol
+      editApiProtocol.value =
+        storedProtocol === 'anthropic' || storedProtocol === 'responses' ? storedProtocol : 'chat_completions'
+      if (newAccount.platform !== 'deepseek' && editApiProtocol.value === 'responses') {
+        editApiProtocol.value = 'chat_completions'
+      }
+    }
     const platformDefaultUrl =
       newAccount.platform === 'openai'
         ? 'https://api.openai.com'
@@ -3629,7 +3781,11 @@ const syncFormFromAccount = (newAccount: Account | null) => {
           ? 'https://generativelanguage.googleapis.com'
           : newAccount.platform === 'grok'
             ? 'https://api.x.ai/v1'
-            : 'https://api.anthropic.com'
+            : newAccount.platform === 'kimi' ||
+                newAccount.platform === 'zhipu' ||
+                newAccount.platform === 'deepseek'
+              ? defaultCNBaseUrl(newAccount.platform, editAccountMode.value, editApiProtocol.value)
+              : 'https://api.anthropic.com'
     editBaseUrl.value = (credentials.base_url as string) || platformDefaultUrl
 
     // Load model mappings and detect mode
@@ -4298,6 +4454,12 @@ const handleSubmit = async () => {
         base_url: newBaseUrl
       }
 
+      // 国产供应商：模式与协议写入凭据（决定额度/余额探测与转发端点/格式）。
+      if (isCNApiKeyAccount.value) {
+        newCredentials.account_mode = editAccountMode.value
+        newCredentials.api_protocol = editApiProtocol.value
+      }
+
       // Handle API key
       // 后端响应已脱敏：currentCredentials 不会再包含 api_key 原文。
       // 用户填入新值则覆盖；留空时优先看 credentials_status.has_api_key；
diff --git a/frontend/src/components/account/ModelWhitelistSelector.vue b/frontend/src/components/account/ModelWhitelistSelector.vue
index d8cec05cb977..b6086c6d6ba6 100644
--- a/frontend/src/components/account/ModelWhitelistSelector.vue
+++ b/frontend/src/components/account/ModelWhitelistSelector.vue
@@ -199,7 +199,16 @@ const normalizedPlatforms = computed(() => {
   )
 })
 
-const upstreamSyncPlatforms = new Set(['anthropic', 'openai', 'gemini', 'antigravity', 'grok'])
+const upstreamSyncPlatforms = new Set([
+  'anthropic',
+  'openai',
+  'gemini',
+  'antigravity',
+  'grok',
+  'kimi',
+  'zhipu',
+  'deepseek'
+])
 const canSyncUpstream = computed(() => {
   if (props.accountId) {
     if (normalizedPlatforms.value.length === 0) return true
diff --git a/frontend/src/components/account/credentialsBuilder.ts b/frontend/src/components/account/credentialsBuilder.ts
index 32214152506b..054fd4623898 100644
--- a/frontend/src/components/account/credentialsBuilder.ts
+++ b/frontend/src/components/account/credentialsBuilder.ts
@@ -242,6 +242,91 @@ export const GROK_BASE_URL_PRESETS: GrokBaseUrlPreset[] = [
   { label: 'eu-west-1', url: 'https://eu-west-1.api.x.ai/v1' }
 ]
 
+// ========== 国产供应商（Kimi / Zhipu / DeepSeek）base_url 预设 ==========
+// 与后端 service/domain_constants.go 的默认 base url 保持一致。
+// 账号类型（payg 按量付费 / coding 编程套餐）决定额度监控方式；
+// API 协议（chat_completions / anthropic / responses）决定转发端点与格式，
+// 两者正交。同协议请求零转换直通，跨协议组合才走转换链。
+
+export type CnAccountMode = 'payg' | 'coding'
+
+/** 仅 deepseek 支持 responses 协议（官方原生 /responses 端点，适配 Codex）。 */
+export type CnApiProtocol = 'chat_completions' | 'anthropic' | 'responses'
+
+export interface CnBaseUrlPreset {
+  mode: CnAccountMode
+  protocol: CnApiProtocol
+  /** 专有名词，不参与 i18n */
+  label: string
+  url: string
+}
+
+/** 各供应商按账号类型 × API 协议分档的快捷端点（点击快速填充，输入框仍可自由填写）。 */
+export const CN_BASE_URL_PRESETS: Record<'kimi' | 'zhipu' | 'deepseek', CnBaseUrlPreset[]> = {
+  kimi: [
+    { mode: 'payg', protocol: 'chat_completions', label: 'Moonshot', url: 'https://api.moonshot.cn/v1' },
+    { mode: 'payg', protocol: 'anthropic', label: 'Moonshot Anthropic', url: 'https://api.moonshot.cn/anthropic' },
+    { mode: 'coding', protocol: 'chat_completions', label: 'Kimi For Coding', url: 'https://api.kimi.com/coding/v1' },
+    { mode: 'coding', protocol: 'anthropic', label: 'Kimi Coding Anthropic', url: 'https://api.kimi.com/coding' }
+  ],
+  zhipu: [
+    { mode: 'payg', protocol: 'chat_completions', label: 'GLM PaaS', url: 'https://open.bigmodel.cn/api/paas/v4' },
+    { mode: 'payg', protocol: 'anthropic', label: 'GLM Anthropic', url: 'https://open.bigmodel.cn/api/anthropic' },
+    { mode: 'coding', protocol: 'chat_completions', label: 'GLM Coding', url: 'https://open.bigmodel.cn/api/coding/paas/v4' },
+    { mode: 'coding', protocol: 'anthropic', label: 'GLM Coding Anthropic', url: 'https://open.bigmodel.cn/api/anthropic' }
+  ],
+  deepseek: [
+    { mode: 'payg', protocol: 'chat_completions', label: 'DeepSeek', url: 'https://api.deepseek.com' },
+    { mode: 'payg', protocol: 'anthropic', label: 'DeepSeek Anthropic', url: 'https://api.deepseek.com/anthropic' },
+    { mode: 'payg', protocol: 'responses', label: 'DeepSeek Responses', url: 'https://api.deepseek.com' }
+  ]
+}
+
+/** 返回指定供应商 + 账号类型 + API 协议的默认 base url。 */
+export function defaultCNBaseUrl(
+  platform: string,
+  mode: CnAccountMode,
+  protocol: CnApiProtocol = 'chat_completions'
+): string {
+  if (protocol === 'anthropic') {
+    switch (platform) {
+      case 'kimi':
+        return mode === 'coding' ? 'https://api.kimi.com/coding' : 'https://api.moonshot.cn/anthropic'
+      case 'zhipu':
+        return 'https://open.bigmodel.cn/api/anthropic'
+      case 'deepseek':
+        return 'https://api.deepseek.com/anthropic'
+      default:
+        return ''
+    }
+  }
+  // responses 仅 deepseek：base 与 chat_completions 相同（端点路径差异由后端处理）。
+  switch (platform) {
+    case 'kimi':
+      return mode === 'coding' ? 'https://api.kimi.com/coding/v1' : 'https://api.moonshot.cn/v1'
+    case 'zhipu':
+      return mode === 'coding'
+        ? 'https://open.bigmodel.cn/api/coding/paas/v4'
+        : 'https://open.bigmodel.cn/api/paas/v4'
+    case 'deepseek':
+      return 'https://api.deepseek.com'
+    default:
+      return ''
+  }
+}
+
+// ===== 国产供应商用量单元格可见性（单一事实源） =====
+// CNProviderQuotaCell / CNProviderBalanceCell 与 AccountUsageCell 的占位符判定
+// 共用，避免多处复制条件后一处改另一处漏改。
+
+export function cnQuotaCellVisible(platform: string, accountMode: string): boolean {
+  return (platform === 'kimi' || platform === 'zhipu') && accountMode === 'coding'
+}
+
+export function cnBalanceCellVisible(platform: string, accountMode: string): boolean {
+  return (platform === 'kimi' || platform === 'deepseek') && accountMode !== 'coding'
+}
+
 /**
  * 将请求头覆写写入 credentials。
  * create 模式：关闭时不写入任何字段；edit 模式：关闭时删除字段（全量替换语义）。
diff --git a/frontend/src/components/common/PlatformIcon.vue b/frontend/src/components/common/PlatformIcon.vue
index 937990ad7540..7a93b2834962 100644
--- a/frontend/src/components/common/PlatformIcon.vue
+++ b/frontend/src/components/common/PlatformIcon.vue
@@ -25,6 +25,25 @@
       d="M9.27 15.29l7.978-5.897c.391-.29.95-.177 1.137.272.98 2.369.542 5.215-1.41 7.169-1.951 1.954-4.667 2.382-7.149 1.406l-2.711 1.257c3.889 2.661 8.611 2.003 11.562-.953 2.341-2.344 3.066-5.539 2.388-8.42l.006.007c-.983-4.232.242-5.924 2.75-9.383.06-.082.12-.164.179-.248l-3.301 3.305v-.01L9.267 15.292M7.623 16.723c-2.792-2.67-2.31-6.801.071-9.184 1.761-1.763 4.647-2.483 7.166-1.425l2.705-1.25a7.808 7.808 0 00-1.829-1A8.975 8.975 0 005.984 5.83c-2.533 2.536-3.33 6.436-1.962 9.764 1.022 2.487-.653 4.246-2.34 6.022-.599.63-1.199 1.259-1.682 1.925l7.62-6.815"
     />
   </svg>
+  <!-- Kimi / Moonshot official logo mark (stylized K) -->
+  <svg v-else-if="platform === 'kimi'" :class="sizeClass" viewBox="0 0 24 24" fill="currentColor">
+    <path
+      d="M21.765.351C22.998.351 24 1.353 24 2.586S22.998 4.82 21.765 4.82h-1.974c-.15 0-.26-.12-.26-.26V2.586A2.237 2.237 0 0 1 21.765.35M9.41 13.388l8.447-8.377c.16-.16.07-.471-.14-.471h-4.55s-.1.02-.14.06l-9.099 9.029c-.14.14-.35.02-.35-.21V4.81c0-.15-.1-.27-.221-.27H.22c-.12 0-.22.12-.22.27v18.57c0 .15.1.27.22.27h3.137c.12 0 .22-.12.22-.27v-3.79c0-.08.03-.16.08-.21l2.826-2.796c.07-.07.16-.08.241-.03l7.546 5.551a8.9 8.9 0 0 0 4.018 1.493c.12.01.23-.11.23-.27V19.76c0-.14-.08-.25-.19-.26a5.8 5.8 0 0 1-2.355-.942l-6.533-4.73c-.14-.09-.15-.32-.03-.441"
+    />
+  </svg>
+  <!-- Zhipu AI official logo mark (particle constellation) -->
+  <svg v-else-if="platform === 'zhipu'" :class="sizeClass" viewBox="0 0 24 24" fill="currentColor">
+    <path
+      fill-rule="evenodd"
+      d="M11.991 23.503a.24.24 0 0 0-.244.248a.24.24 0 0 0 .244.249a.24.24 0 0 0 .245-.249a.24.24 0 0 0-.22-.247zM9.671 5.365a1.697 1.697 0 0 1 1.099 2.132l-.071.172l-.016.04l-.018.054c-.07.16-.104.32-.104.498c-.035.71.47 1.279 1.186 1.314h.366c1.309.053 2.338 1.173 2.286 2.523c-.052 1.332-1.152 2.38-2.478 2.327h-.174c-.715.018-1.274.64-1.239 1.368c0 .124.018.23.053.337c.209.373.54.658.96.8c.75.23 1.517-.125 1.9-.782l.018-.035c.402-.64 1.17-.96 1.92-.711c.854.284 1.378 1.226 1.099 2.167a1.66 1.66 0 0 1-2.077 1.102a1.7 1.7 0 0 1-.907-.711l-.017-.035c-.2-.323-.463-.58-.851-.711l-.056-.018a1.646 1.646 0 0 0-1.954.746a1.66 1.66 0 0 1-1.065.764a1.677 1.677 0 0 1-1.989-1.279c-.209-.906.332-1.83 1.257-2.043a1.5 1.5 0 0 1 .296-.035h.018c.68-.071 1.151-.622 1.116-1.333a1.3 1.3 0 0 0-.227-.693a2.5 2.5 0 0 1-.366-1.403a2.4 2.4 0 0 1 .366-1.208c.14-.195.21-.444.227-.693c.018-.71-.506-1.261-1.186-1.332l-.07-.018a1.4 1.4 0 0 1-.299-.07l-.05-.019a1.7 1.7 0 0 1-1.047-2.114a1.68 1.68 0 0 1 2.094-1.101m-5.575 10.11c.26-.264.639-.367.994-.27s.633.379.728.74c.095.362-.007.748-.267 1.013c-.402.41-1.053.41-1.455 0a1.06 1.06 0 0 1 0-1.482zm14.845-.294c.359-.09.738.024.992.297c.254.274.344.665.237 1.025s-.396.634-.756.718c-.551.128-1.1-.22-1.23-.781a1.05 1.05 0 0 1 .757-1.26zm-.064-4.39c.314.32.49.753.49 1.206s-.176.886-.49 1.206c-.315.32-.74.5-1.185.5c-.444 0-.87-.18-1.184-.5a1.727 1.727 0 0 1 0-2.412a1.654 1.654 0 0 1 2.369 0m-11.243.163c.364.484.447 1.128.218 1.691a1.665 1.665 0 0 1-2.188.923c-.855-.36-1.26-1.358-.907-2.228a1.68 1.68 0 0 1 1.33-1.038a1.66 1.66 0 0 1 1.547.652m11.545-4.221c.368 0 .708.2.892.524s.184.724 0 1.048a1.03 1.03 0 0 1-.892.524a1.04 1.04 0 0 1-1.03-1.048a1.04 1.04 0 0 1 1.03-1.048m-14.358 0c.368 0 .707.2.891.524s.184.724 0 1.048a1.03 1.03 0 0 1-.891.524a1.04 1.04 0 0 1-1.03-1.048c0-.579.461-1.048 1.03-1.048m10.031-1.475c.925 0 1.675.764 1.675 1.706s-.75 1.705-1.675 1.705s-1.674-.763-1.674-1.705s.75-1.706 1.674-1.706m-2.626-.684c.362-.082.653-.356.761-.718a1.06 1.06 0 0 0-.238-1.028a1.02 1.02 0 0 0-.996-.294c-.547.14-.881.7-.752 1.257c.13.558.675.907 1.225.783m0 16.876c.359-.087.644-.36.75-.72a1.06 1.06 0 0 0-.237-1.019a1.02 1.02 0 0 0-.985-.301a1.04 1.04 0 0 0-.762.717c-.108.361-.017.754.239 1.028c.245.263.606.377.953.305zM17.19 3.5a.63.63 0 0 0 .628-.64a.63.63 0 0 0-.628-.64a.63.63 0 0 0-.628.64c0 .355.28.64.628.64m-10.38 0a.63.63 0 0 0 .628-.64c0-.355-.28-.64-.628-.64a.63.63 0 0 0-.628.64c0 .355.279.64.628.64m-5.182 7.852a.63.63 0 0 0-.628.64c0 .354.28.639.628.639a.63.63 0 0 0 .627-.606l.001-.034a.62.62 0 0 0-.628-.64zm5.182 9.13a.63.63 0 0 0-.628.64c0 .355.279.64.628.64a.63.63 0 0 0 .628-.64c0-.355-.28-.64-.628-.64m10.38.018a.63.63 0 0 0-.628.64c0 .355.28.64.628.64a.63.63 0 0 0 .628-.64a.63.63 0 0 0-.628-.64m5.182-9.148a.63.63 0 0 0-.628.64c0 .354.279.639.628.639a.63.63 0 0 0 .628-.64c0-.355-.28-.64-.628-.64zm-.384-4.992a.24.24 0 0 0 .244-.249a.24.24 0 0 0-.244-.249a.24.24 0 0 0-.244.249c0 .142.122.249.244.249M11.991.497a.24.24 0 0 0 .245-.248A.24.24 0 0 0 11.99 0a.24.24 0 0 0-.244.249c0 .133.108.236.223.247zM2.011 6.36a.24.24 0 0 0 .245-.249a.24.24 0 0 0-.244-.249a.24.24 0 0 0-.244.249a.24.24 0 0 0 .244.249zm0 11.263a.24.24 0 0 0-.243.248a.24.24 0 0 0 .244.249a.24.24 0 0 0 .244-.249a.25.25 0 0 0-.244-.248zm19.995-.018a.24.24 0 0 0-.245.248a.24.24 0 0 0 .245.25a.24.24 0 0 0 .244-.25a.25.25 0 0 0-.244-.248z"
+    />
+  </svg>
+  <!-- DeepSeek official logo mark (whale) -->
+  <svg v-else-if="platform === 'deepseek'" :class="sizeClass" viewBox="0 0 24 24" fill="currentColor">
+    <path
+      d="M23.748 4.651c-.254-.124-.364.113-.512.233-.051.04-.094.09-.137.137-.372.397-.806.657-1.373.626-.829-.046-1.537.214-2.163.848-.133-.782-.575-1.248-1.247-1.548-.352-.155-.708-.311-.955-.65-.172-.24-.219-.509-.305-.774-.055-.16-.11-.323-.293-.35-.2-.031-.278.136-.356.276-.313.572-.434 1.202-.422 1.84.027 1.436.633 2.58 1.838 3.393.137.094.172.187.129.323-.082.28-.18.553-.266.833-.055.179-.137.218-.328.14a5.5 5.5 0 0 1-1.737-1.179c-.857-.828-1.631-1.743-2.597-2.46a12 12 0 0 0-.689-.47c-.985-.957.13-1.743.387-1.836.27-.098.094-.433-.778-.428-.872.003-1.67.295-2.687.685a3 3 0 0 1-.465.136a9.6 9.6 0 0 0-2.883-.101c-1.885.21-3.39 1.1-4.497 2.622C.082 8.776-.231 10.854.152 13.02c.403 2.284 1.568 4.175 3.36 5.653 1.857 1.533 3.997 2.284 6.438 2.14 1.482-.085 3.132-.284 4.994-1.86.47.234.962.328 1.78.398.629.058 1.235-.031 1.705-.129.735-.155.684-.836.418-.961-2.155-1.004-1.682-.595-2.112-.926 1.095-1.295 2.768-3.598 3.284-6.733.05-.346.115-.834.108-1.114-.004-.171.035-.238.23-.257a4.2 4.2 0 0 0 1.545-.475c1.397-.763 1.96-2.016 2.093-3.517.02-.23-.004-.467-.247-.588M11.58 18.168c-2.088-1.642-3.101-2.183-3.52-2.16-.39.024-.32.472-.234.763.09.288.207.487.371.74.114.167.192.416-.113.603-.673.416-1.842-.14-1.897-.168-1.361-.801-2.5-1.86-3.301-3.306-.775-1.393-1.225-2.888-1.299-4.482-.02-.385.094-.522.477-.592a4.7 4.7 0 0 1 1.53-.038c2.131.311 3.946 1.264 5.467 2.774.868.86 1.525 1.887 2.202 2.89.72 1.066 1.494 2.082 2.48 2.915.348.291.626.513.892.677-.802.09-2.14.109-3.055-.615zm1.001-6.44a.306.306 0 0 1 .415-.287a.3.3 0 0 1 .113.074a.3.3 0 0 1 .086.214c0 .17-.136.307-.308.307a.303.303 0 0 1-.306-.307m3.11 1.596c-.2.081-.4.151-.591.16a1.25 1.25 0 0 1-.798-.254c-.274-.23-.47-.358-.551-.758a1.7 1.7 0 0 1 .015-.588c.07-.327-.007-.537-.238-.727-.188-.156-.426-.199-.689-.199a.6.6 0 0 1-.254-.078a.253.253 0 0 1-.114-.358a1 1 0 0 1 .192-.21c.356-.202.767-.136 1.146.016.352.144.618.408 1.001.782.392.451.462.576.685.915.176.264.336.536.446.848.066.194-.02.353-.25.45"
+    />
+  </svg>
   <!-- Composite group icon -->
   <svg v-else-if="platform === 'composite'" :class="sizeClass" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
     <circle cx="6" cy="12" r="3" />
diff --git a/frontend/src/components/common/PlatformTypeBadge.vue b/frontend/src/components/common/PlatformTypeBadge.vue
index ef86879ae04f..399a1bece8e1 100644
--- a/frontend/src/components/common/PlatformTypeBadge.vue
+++ b/frontend/src/components/common/PlatformTypeBadge.vue
@@ -90,6 +90,9 @@ const platformLabel = computed(() => {
   if (props.platform === 'openai') return 'OpenAI'
   if (props.platform === 'antigravity') return 'Antigravity'
   if (props.platform === 'grok') return 'Grok'
+  if (props.platform === 'kimi') return 'Kimi'
+  if (props.platform === 'zhipu') return 'Zhipu GLM'
+  if (props.platform === 'deepseek') return 'DeepSeek'
   return 'Gemini'
 })
 
@@ -188,6 +191,15 @@ const platformClass = computed(() => {
   if (props.platform === 'grok') {
     return 'bg-zinc-100 text-zinc-700 dark:bg-zinc-800 dark:text-zinc-300'
   }
+  if (props.platform === 'kimi') {
+    return 'bg-pink-100 text-pink-700 dark:bg-pink-900/30 dark:text-pink-400'
+  }
+  if (props.platform === 'zhipu') {
+    return 'bg-indigo-100 text-indigo-700 dark:bg-indigo-900/30 dark:text-indigo-400'
+  }
+  if (props.platform === 'deepseek') {
+    return 'bg-teal-100 text-teal-700 dark:bg-teal-900/30 dark:text-teal-400'
+  }
   return 'bg-blue-100 text-blue-700 dark:bg-blue-900/30 dark:text-blue-400'
 })
 
@@ -204,6 +216,15 @@ const typeClass = computed(() => {
   if (props.platform === 'grok') {
     return 'bg-zinc-100 text-zinc-600 dark:bg-zinc-800 dark:text-zinc-300'
   }
+  if (props.platform === 'kimi') {
+    return 'bg-pink-100 text-pink-600 dark:bg-pink-900/30 dark:text-pink-400'
+  }
+  if (props.platform === 'zhipu') {
+    return 'bg-indigo-100 text-indigo-600 dark:bg-indigo-900/30 dark:text-indigo-400'
+  }
+  if (props.platform === 'deepseek') {
+    return 'bg-teal-100 text-teal-600 dark:bg-teal-900/30 dark:text-teal-400'
+  }
   return 'bg-blue-100 text-blue-600 dark:bg-blue-900/30 dark:text-blue-400'
 })
 
diff --git a/frontend/src/composables/useModelWhitelist.ts b/frontend/src/composables/useModelWhitelist.ts
index e2c5924592f5..2ca860c673c8 100644
--- a/frontend/src/composables/useModelWhitelist.ts
+++ b/frontend/src/composables/useModelWhitelist.ts
@@ -182,7 +182,9 @@ const yiModels = [
 // Moonshot/Kimi
 const moonshotModels = [
   'moonshot-v1-8k', 'moonshot-v1-32k', 'moonshot-v1-128k',
-  'kimi-latest'
+  'kimi-latest',
+  'kimi-for-coding',
+  'kimi-k2'
 ]
 
 // 字节跳动 豆包
@@ -428,7 +430,8 @@ export function getModelsByPlatform(platform: string): string[] {
     case 'grok': return xaiModels
     case 'cohere': return cohereModels
     case 'yi': return yiModels
-    case 'moonshot': return moonshotModels
+    case 'moonshot':
+    case 'kimi': return moonshotModels
     case 'doubao': return doubaoModels
     case 'minimax': return minimaxModels
     case 'baidu': return baiduModels
diff --git a/frontend/src/i18n/locales/en/admin/accounts.ts b/frontend/src/i18n/locales/en/admin/accounts.ts
index 0eb245033f2f..de85669443ff 100644
--- a/frontend/src/i18n/locales/en/admin/accounts.ts
+++ b/frontend/src/i18n/locales/en/admin/accounts.ts
@@ -104,6 +104,33 @@ export default {
         gemini: 'Gemini',
         antigravity: 'Antigravity',
         grok: 'Grok',
+        kimi: 'Kimi',
+        zhipu: 'Zhipu GLM',
+        deepseek: 'DeepSeek',
+      },
+      cnProviders: {
+        accountMode: {
+          title: 'Account Type',
+          payg: 'Pay-as-you-go',
+          paygDesc: 'Consumes account balance, billed per token. Auto-cools down on low balance and recovers after top-up.',
+          coding: 'Coding Plan',
+          codingDesc: 'Subscription coding package, rate-limited by 5-hour / weekly rolling usage windows.',
+        },
+        apiProtocol: {
+          title: 'API Protocol',
+          chatCompletions: 'Chat Completions',
+          chatCompletionsDesc: 'Standard OpenAI-compatible endpoint; requests in other formats are converted.',
+          anthropic: 'Anthropic',
+          anthropicDesc: 'Native passthrough to the provider’s Anthropic endpoint — ideal for Claude Code.',
+          responses: 'Responses',
+          responsesDesc: 'Provider’s native Responses endpoint — ideal for Codex.',
+        },
+        window5h: '5-hour window',
+        windowWeekly: 'Weekly window',
+        probeTooltip: 'Query the provider quota endpoint for 5-hour / weekly rolling window usage',
+        balanceLow: 'Insufficient balance',
+        noBalanceEndpoint: 'This platform has no balance query endpoint',
+        resetSoon: 'reset soon',
       },
       types: {
         oauth: 'OAuth',
diff --git a/frontend/src/i18n/locales/zh/admin/accounts.ts b/frontend/src/i18n/locales/zh/admin/accounts.ts
index 1c6d000c3c75..be47ced57e87 100644
--- a/frontend/src/i18n/locales/zh/admin/accounts.ts
+++ b/frontend/src/i18n/locales/zh/admin/accounts.ts
@@ -307,6 +307,33 @@ export default {
         gemini: 'Gemini',
         antigravity: 'Antigravity',
         grok: 'Grok',
+        kimi: 'Kimi',
+        zhipu: 'Zhipu GLM',
+        deepseek: 'DeepSeek',
+      },
+      cnProviders: {
+        accountMode: {
+          title: '账号类型',
+          payg: '按量付费',
+          paygDesc: '消耗账户余额，按 Token 计费。余额不足自动冷却，充值后恢复。',
+          coding: 'Coding Plan',
+          codingDesc: '订阅制编程套餐，按 5 小时 / 每周滚动用量窗口限流。',
+        },
+        apiProtocol: {
+          title: 'API 协议',
+          chatCompletions: 'Chat Completions',
+          chatCompletionsDesc: '标准 OpenAI 兼容端点，其他格式请求将被转换。',
+          anthropic: 'Anthropic',
+          anthropicDesc: '直通供应商原生 Anthropic 端点，零转换，适配 Claude Code。',
+          responses: 'Responses',
+          responsesDesc: '供应商原生 Responses 端点，适配 Codex。',
+        },
+        window5h: '5 小时窗口',
+        windowWeekly: '每周窗口',
+        probeTooltip: '请求供应商额度端点，查询 5 小时 / 每周滚动窗口用量',
+        balanceLow: '余额不足',
+        noBalanceEndpoint: '该平台暂无余额查询接口',
+        resetSoon: '即将重置',
       },
       types: {
         oauth: 'OAuth',
diff --git a/frontend/src/types/index.ts b/frontend/src/types/index.ts
index 19ef541bc61f..fa162814d61b 100644
--- a/frontend/src/types/index.ts
+++ b/frontend/src/types/index.ts
@@ -525,7 +525,7 @@ export interface PaginationConfig {
 
 // ==================== API Key & Group Types ====================
 
-export type GroupPlatform = 'anthropic' | 'openai' | 'gemini' | 'antigravity' | 'grok' | 'composite'
+export type GroupPlatform = 'anthropic' | 'openai' | 'gemini' | 'antigravity' | 'grok' | 'kimi' | 'zhipu' | 'deepseek' | 'composite'
 
 export type VideoModelPrices = Record<string, Record<string, number>>
 
@@ -882,7 +882,7 @@ export interface UpdateGroupRequest {
 
 // ==================== Account & Proxy Types ====================
 
-export type AccountPlatform = 'anthropic' | 'openai' | 'gemini' | 'antigravity' | 'grok'
+export type AccountPlatform = 'anthropic' | 'openai' | 'gemini' | 'antigravity' | 'grok' | 'kimi' | 'zhipu' | 'deepseek'
 export type AccountType = 'oauth' | 'setup-token' | 'apikey' | 'upstream' | 'bedrock' | 'service_account'
 export type OAuthAddMethod = 'oauth' | 'setup-token'
 export type ProxyProtocol = 'http' | 'https' | 'socks5' | 'socks5h'
diff --git a/frontend/src/utils/platformColors.ts b/frontend/src/utils/platformColors.ts
index 8ab5e44138bb..97e824390e36 100644
--- a/frontend/src/utils/platformColors.ts
+++ b/frontend/src/utils/platformColors.ts
@@ -5,7 +5,16 @@
  * instead of defining their own color mappings.
  */
 
-export type Platform = 'anthropic' | 'openai' | 'antigravity' | 'gemini' | 'grok' | 'composite'
+export type Platform =
+  | 'anthropic'
+  | 'openai'
+  | 'antigravity'
+  | 'gemini'
+  | 'grok'
+  | 'kimi'
+  | 'zhipu'
+  | 'deepseek'
+  | 'composite'
 
 // ── Badge (bg + text + border, for inline badges with border) ───────
 const BADGE: Record<Platform, string> = {
@@ -14,6 +23,9 @@ const BADGE: Record<Platform, string> = {
   antigravity: 'bg-purple-500/10 text-purple-600 border-purple-500/30 dark:text-purple-400',
   gemini: 'bg-blue-500/10 text-blue-600 border-blue-500/30 dark:text-blue-400',
   grok: 'bg-zinc-800/10 text-zinc-800 border-zinc-800/30 dark:bg-zinc-500/10 dark:text-zinc-200 dark:border-zinc-500/30',
+  kimi: 'bg-pink-500/10 text-pink-600 border-pink-500/30 dark:text-pink-400',
+  zhipu: 'bg-indigo-500/10 text-indigo-600 border-indigo-500/30 dark:text-indigo-400',
+  deepseek: 'bg-teal-500/10 text-teal-600 border-teal-500/30 dark:text-teal-400',
   composite: 'bg-cyan-500/10 text-cyan-700 border-cyan-500/30 dark:text-cyan-300',
 }
 const BADGE_DEFAULT = 'bg-slate-500/10 text-slate-600 border-slate-500/30 dark:text-slate-400'
@@ -25,6 +37,9 @@ const BADGE_LIGHT: Record<Platform, string> = {
   antigravity: 'bg-purple-500/10 text-purple-600 dark:bg-purple-500/10 dark:text-purple-300',
   gemini: 'bg-blue-500/10 text-blue-600 dark:bg-blue-500/10 dark:text-blue-300',
   grok: 'bg-zinc-800/10 text-zinc-800 dark:bg-zinc-500/10 dark:text-zinc-200',
+  kimi: 'bg-pink-500/10 text-pink-600 dark:bg-pink-500/10 dark:text-pink-300',
+  zhipu: 'bg-indigo-500/10 text-indigo-600 dark:bg-indigo-500/10 dark:text-indigo-300',
+  deepseek: 'bg-teal-500/10 text-teal-600 dark:bg-teal-500/10 dark:text-teal-300',
   composite: 'bg-cyan-500/10 text-cyan-700 dark:bg-cyan-500/10 dark:text-cyan-300',
 }
 
@@ -35,6 +50,9 @@ const BORDER: Record<Platform, string> = {
   antigravity: 'border-purple-500/20 dark:border-purple-500/20',
   gemini: 'border-blue-500/20 dark:border-blue-500/20',
   grok: 'border-zinc-800/20 dark:border-zinc-500/20',
+  kimi: 'border-pink-500/20 dark:border-pink-500/20',
+  zhipu: 'border-indigo-500/20 dark:border-indigo-500/20',
+  deepseek: 'border-teal-500/20 dark:border-teal-500/20',
   composite: 'border-cyan-500/20 dark:border-cyan-500/20',
 }
 const BORDER_DEFAULT = 'border-gray-200 dark:border-dark-700'
@@ -46,6 +64,9 @@ const BORDER_STRONG: Record<Platform, string> = {
   antigravity: 'border-purple-500/35 dark:border-purple-500/30',
   gemini: 'border-blue-500/35 dark:border-blue-500/30',
   grok: 'border-zinc-800/35 dark:border-zinc-500/35',
+  kimi: 'border-pink-500/35 dark:border-pink-500/30',
+  zhipu: 'border-indigo-500/35 dark:border-indigo-500/30',
+  deepseek: 'border-teal-500/35 dark:border-teal-500/30',
   composite: 'border-cyan-500/35 dark:border-cyan-500/30',
 }
 const BORDER_STRONG_DEFAULT = 'border-gray-300 dark:border-dark-600'
@@ -58,6 +79,9 @@ const ACCENT: Record<Platform, string> = {
   antigravity: '#a855f7', // purple-500
   gemini: '#3b82f6', // blue-500
   grok: '#71717a', // zinc-500
+  kimi: '#ec4899', // pink-500
+  zhipu: '#6366f1', // indigo-500
+  deepseek: '#14b8a6', // teal-500
   composite: '#06b6d4', // cyan-500
 }
 const ACCENT_DEFAULT = '#14b8a6' // primary-500 (teal)
@@ -69,6 +93,9 @@ const ACCENT_BAR: Record<Platform, string> = {
   antigravity: 'bg-gradient-to-r from-purple-400 to-purple-500',
   gemini: 'bg-gradient-to-r from-blue-400 to-blue-500',
   grok: 'bg-gradient-to-r from-zinc-700 to-zinc-900',
+  kimi: 'bg-gradient-to-r from-pink-400 to-pink-500',
+  zhipu: 'bg-gradient-to-r from-indigo-400 to-indigo-500',
+  deepseek: 'bg-gradient-to-r from-teal-400 to-teal-500',
   composite: 'bg-gradient-to-r from-slate-500 to-cyan-500',
 }
 const ACCENT_BAR_DEFAULT = 'bg-gradient-to-r from-primary-400 to-primary-500'
@@ -80,6 +107,9 @@ const TEXT: Record<Platform, string> = {
   antigravity: 'text-purple-600 dark:text-purple-400',
   gemini: 'text-blue-600 dark:text-blue-400',
   grok: 'text-zinc-800 dark:text-zinc-200',
+  kimi: 'text-pink-600 dark:text-pink-400',
+  zhipu: 'text-indigo-600 dark:text-indigo-400',
+  deepseek: 'text-teal-600 dark:text-teal-400',
   composite: 'text-cyan-700 dark:text-cyan-300',
 }
 const TEXT_DEFAULT = 'text-primary-600 dark:text-primary-400'
@@ -91,6 +121,9 @@ const ICON: Record<Platform, string> = {
   antigravity: 'text-purple-500 dark:text-purple-400',
   gemini: 'text-blue-500 dark:text-blue-400',
   grok: 'text-zinc-800 dark:text-zinc-200',
+  kimi: 'text-pink-500 dark:text-pink-400',
+  zhipu: 'text-indigo-500 dark:text-indigo-400',
+  deepseek: 'text-teal-500 dark:text-teal-400',
   composite: 'text-cyan-600 dark:text-cyan-300',
 }
 const ICON_DEFAULT = 'text-primary-500 dark:text-primary-400'
@@ -102,6 +135,9 @@ const BUTTON: Record<Platform, string> = {
   antigravity: 'bg-purple-500 text-white hover:bg-purple-600 active:bg-purple-700 dark:bg-purple-500/80 dark:hover:bg-purple-500',
   gemini: 'bg-blue-500 text-white hover:bg-blue-600 active:bg-blue-700 dark:bg-blue-500/80 dark:hover:bg-blue-500',
   grok: 'bg-zinc-800 text-white hover:bg-zinc-900 active:bg-black dark:bg-zinc-700 dark:hover:bg-zinc-600',
+  kimi: 'bg-pink-500 text-white hover:bg-pink-600 active:bg-pink-700 dark:bg-pink-500/80 dark:hover:bg-pink-500',
+  zhipu: 'bg-indigo-500 text-white hover:bg-indigo-600 active:bg-indigo-700 dark:bg-indigo-500/80 dark:hover:bg-indigo-500',
+  deepseek: 'bg-teal-500 text-white hover:bg-teal-600 active:bg-teal-700 dark:bg-teal-500/80 dark:hover:bg-teal-500',
   composite: 'bg-cyan-700 text-white hover:bg-cyan-800 active:bg-cyan-900 dark:bg-cyan-600 dark:hover:bg-cyan-500',
 }
 const BUTTON_DEFAULT = 'bg-primary-500 text-white hover:bg-primary-600 dark:bg-primary-600 dark:hover:bg-primary-500'
@@ -113,6 +149,9 @@ const DISCOUNT: Record<Platform, string> = {
   antigravity: 'bg-purple-100 text-purple-700 dark:bg-purple-900/40 dark:text-purple-300',
   gemini: 'bg-blue-100 text-blue-700 dark:bg-blue-900/40 dark:text-blue-300',
   grok: 'bg-zinc-100 text-zinc-800 dark:bg-zinc-800 dark:text-zinc-200',
+  kimi: 'bg-pink-100 text-pink-700 dark:bg-pink-900/40 dark:text-pink-300',
+  zhipu: 'bg-indigo-100 text-indigo-700 dark:bg-indigo-900/40 dark:text-indigo-300',
+  deepseek: 'bg-teal-100 text-teal-700 dark:bg-teal-900/40 dark:text-teal-300',
   composite: 'bg-cyan-100 text-cyan-800 dark:bg-cyan-900/40 dark:text-cyan-300',
 }
 const DISCOUNT_DEFAULT = 'bg-red-100 text-red-700 dark:bg-red-900/40 dark:text-red-300'
@@ -124,6 +163,9 @@ const GRADIENT: Record<Platform, string> = {
   antigravity: 'from-purple-500 to-purple-600',
   gemini: 'from-blue-500 to-blue-600',
   grok: 'from-zinc-700 to-zinc-900',
+  kimi: 'from-pink-500 to-pink-600',
+  zhipu: 'from-indigo-500 to-indigo-600',
+  deepseek: 'from-teal-500 to-teal-600',
   composite: 'from-slate-600 to-cyan-600',
 }
 const GRADIENT_DEFAULT = 'from-primary-500 to-primary-600'
@@ -135,6 +177,9 @@ const GRADIENT_TEXT: Record<Platform, string> = {
   antigravity: 'text-purple-100',
   gemini: 'text-blue-100',
   grok: 'text-zinc-100',
+  kimi: 'text-pink-100',
+  zhipu: 'text-indigo-100',
+  deepseek: 'text-teal-100',
   composite: 'text-cyan-100',
 }
 const GRADIENT_TEXT_DEFAULT = 'text-primary-100'
@@ -145,6 +190,9 @@ const GRADIENT_SUBTEXT: Record<Platform, string> = {
   antigravity: 'text-purple-200',
   gemini: 'text-blue-200',
   grok: 'text-zinc-300',
+  kimi: 'text-pink-200',
+  zhipu: 'text-indigo-200',
+  deepseek: 'text-teal-200',
   composite: 'text-cyan-200',
 }
 const GRADIENT_SUBTEXT_DEFAULT = 'text-primary-200'
@@ -152,7 +200,17 @@ const GRADIENT_SUBTEXT_DEFAULT = 'text-primary-200'
 // ── Public API ──────────────────────────────────────────────────────
 
 function isPlatform(p: string): p is Platform {
-  return p === 'anthropic' || p === 'openai' || p === 'antigravity' || p === 'gemini' || p === 'grok' || p === 'composite'
+  return (
+    p === 'anthropic' ||
+    p === 'openai' ||
+    p === 'antigravity' ||
+    p === 'gemini' ||
+    p === 'grok' ||
+    p === 'kimi' ||
+    p === 'zhipu' ||
+    p === 'deepseek' ||
+    p === 'composite'
+  )
 }
 
 export function platformBadgeClass(p: string): string {
@@ -214,6 +272,9 @@ export function platformLabel(p: string): string {
     case 'antigravity': return 'Antigravity'
     case 'gemini': return 'Gemini'
     case 'grok': return 'Grok'
+    case 'kimi': return 'Kimi'
+    case 'zhipu': return 'Zhipu GLM'
+    case 'deepseek': return 'DeepSeek'
     case 'composite': return 'Composite'
     default: return p || 'API'
   }

From 11e1e228829ad770c7e0e42723ddc37b17b5770d Mon Sep 17 00:00:00 2001
From: Lucky <1646721+luckydududu@users.noreply.github.com>
Date: Sun, 16 Aug 2026 06:17:43 +0000
Subject: [PATCH 22/68] fix(docker): bump Go builder image to 1.26.6 to match
 go.mod

89d826be2 raised backend/go.mod to `go 1.26.6` and updated the three CI
workflows' version assertions, but left the Go builder image in all three
Dockerfiles pinned at 1.26.5. Since the official golang images set
GOTOOLCHAIN=local, the toolchain is not auto-downloaded and any image build
fails hard at `go mod download`.

CI does not catch this: the workflows build with actions/setup-go, not with
these Dockerfiles.

Also extend the Go-upgrade checklist in DEV_GUIDE.md, which listed only the
CI files -- that omission is why the Dockerfiles were missed.
---
 DEV_GUIDE.md       | 2 +-
 Dockerfile         | 2 +-
 backend/Dockerfile | 2 +-
 deploy/Dockerfile  | 2 +-
 4 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/DEV_GUIDE.md b/DEV_GUIDE.md
index 349c24449159..bf5e8867d6c3 100644
--- a/DEV_GUIDE.md
+++ b/DEV_GUIDE.md
@@ -53,7 +53,7 @@ npm install -g pnpm
 
 ### CI 要求
 
-- Go 版本必须是 **1.26.5**：三个 workflow 都用 `go-version-file: backend/go.mod` 取版本，随后硬断言 `go version | grep -q 'go1.26.5'`。升级 Go 时要同时改 `backend/go.mod` 和 `backend-ci.yml`（两处）、`release.yml`、`security-scan.yml` 里的这句断言，否则 CI 会在版本校验步骤直接失败。
+- Go 版本必须是 **1.26.6**：三个 workflow 都用 `go-version-file: backend/go.mod` 取版本，随后硬断言 `go version | grep -q 'go1.26.6'`。升级 Go 时要同时改 `backend/go.mod`、`backend-ci.yml`（两处）、`release.yml`、`security-scan.yml` 里的这句断言，**以及三个 Dockerfile 里的 Go 构建镜像**（`Dockerfile` / `deploy/Dockerfile` 的 `ARG GOLANG_IMAGE`、`backend/Dockerfile` 的 `FROM golang:`）。前者漏了 CI 会在版本校验步骤直接失败；**后者漏了 CI 不会报，而是等到有人用这些 Dockerfile 构建时才失败**（`go.mod requires go >= X (running Y; GOTOOLCHAIN=local)`）。
 - 前端使用 `pnpm install --frozen-lockfile`，必须提交 `pnpm-lock.yaml`
 
 ### 本地测试命令
diff --git a/Dockerfile b/Dockerfile
index b2dfd7034f76..c7991c5a60bf 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -8,7 +8,7 @@
 # =============================================================================
 
 ARG NODE_IMAGE=node:24-alpine
-ARG GOLANG_IMAGE=golang:1.26.5-alpine
+ARG GOLANG_IMAGE=golang:1.26.6-alpine
 ARG ALPINE_IMAGE=alpine:3.21
 ARG POSTGRES_IMAGE=postgres:18-alpine
 ARG GOPROXY=https://goproxy.cn,direct
diff --git a/backend/Dockerfile b/backend/Dockerfile
index 9976abe46b54..1efbd3859da8 100644
--- a/backend/Dockerfile
+++ b/backend/Dockerfile
@@ -1,4 +1,4 @@
-FROM golang:1.26.5-alpine
+FROM golang:1.26.6-alpine
 
 WORKDIR /app
 
diff --git a/deploy/Dockerfile b/deploy/Dockerfile
index 325520f308fb..30c7a1a240f1 100644
--- a/deploy/Dockerfile
+++ b/deploy/Dockerfile
@@ -7,7 +7,7 @@
 # =============================================================================
 
 ARG NODE_IMAGE=node:24-alpine
-ARG GOLANG_IMAGE=golang:1.26.5-alpine
+ARG GOLANG_IMAGE=golang:1.26.6-alpine
 ARG ALPINE_IMAGE=alpine:3.20
 ARG GOPROXY=https://goproxy.cn,direct
 ARG GOSUMDB=sum.golang.google.cn

From e8ff2017c061dd559b4f9ac0b7a5ada72573118a Mon Sep 17 00:00:00 2001
From: feitianbubu <feitianbubu@qq.com>
Date: Sun, 16 Aug 2026 18:04:52 +0800
Subject: [PATCH 23/68] fix(admin): show category labels in ops error
 distribution legend

---
 .../views/admin/ops/components/OpsErrorDistributionChart.vue    | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/frontend/src/views/admin/ops/components/OpsErrorDistributionChart.vue b/frontend/src/views/admin/ops/components/OpsErrorDistributionChart.vue
index ad7ce074ee20..05818c7c912b 100644
--- a/frontend/src/views/admin/ops/components/OpsErrorDistributionChart.vue
+++ b/frontend/src/views/admin/ops/components/OpsErrorDistributionChart.vue
@@ -146,7 +146,7 @@ const options = computed(() => ({
           <div class="flex flex-wrap justify-center gap-3">
             <div v-for="item in categories" :key="item.label" class="flex items-center gap-1.5 text-xs">
               <span class="h-2 w-2 rounded-full" :style="{ backgroundColor: item.color }"></span>
-              <span class="text-gray-500 dark:text-gray-400">{{ item.count }}</span>
+              <span class="text-gray-500 dark:text-gray-400">{{ item.label }} {{ item.count }}</span>
             </div>
           </div>
         </div>

From cb7841d85ce7a74a1b0ff8fa279ce8a0e11767cb Mon Sep 17 00:00:00 2001
From: feitianbubu <feitianbubu@qq.com>
Date: Fri, 14 Aug 2026 14:09:02 +0800
Subject: [PATCH 24/68] fix(i18n): add missing expired key to account status
 block used by proxy list

---
 frontend/src/i18n/locales/en/admin/accounts.ts | 1 +
 frontend/src/i18n/locales/zh/admin/accounts.ts | 1 +
 2 files changed, 2 insertions(+)

diff --git a/frontend/src/i18n/locales/en/admin/accounts.ts b/frontend/src/i18n/locales/en/admin/accounts.ts
index 0eb245033f2f..afc5c4ff7c4e 100644
--- a/frontend/src/i18n/locales/en/admin/accounts.ts
+++ b/frontend/src/i18n/locales/en/admin/accounts.ts
@@ -130,6 +130,7 @@ export default {
       status: {
         active: 'Active',
         inactive: 'Inactive',
+        expired: 'Expired',
         error: 'Error',
         cooldown: 'Cooldown',
         paused: 'Paused',
diff --git a/frontend/src/i18n/locales/zh/admin/accounts.ts b/frontend/src/i18n/locales/zh/admin/accounts.ts
index 1c6d000c3c75..aea0858b8ec5 100644
--- a/frontend/src/i18n/locales/zh/admin/accounts.ts
+++ b/frontend/src/i18n/locales/zh/admin/accounts.ts
@@ -333,6 +333,7 @@ export default {
       status: {
         active: '正常',
         inactive: '停用',
+        expired: '已过期',
         error: '错误',
         cooldown: '冷却中',
         paused: '暂停',

From 22fc0cdbf58c76d49f2b5941aa3fd97d8bd0244d Mon Sep 17 00:00:00 2001
From: InCerryGit <36690780+InCerryGit@users.noreply.github.com>
Date: Sun, 16 Aug 2026 23:49:21 +0800
Subject: [PATCH 25/68] fix(frontend): clarify OpenAI Fast/Flex policy rules

---
 .../__tests__/openaiFastPolicyLocales.spec.ts | 26 +++++++
 .../src/i18n/locales/en/admin/settings.ts     | 25 ++++---
 .../src/i18n/locales/zh/admin/settings.ts     | 25 ++++---
 frontend/src/views/admin/SettingsView.vue     | 72 +++++++++++++++++--
 .../admin/__tests__/SettingsView.spec.ts      | 42 +++++++++++
 5 files changed, 167 insertions(+), 23 deletions(-)

diff --git a/frontend/src/i18n/__tests__/openaiFastPolicyLocales.spec.ts b/frontend/src/i18n/__tests__/openaiFastPolicyLocales.spec.ts
index 7fd550c1ad64..805d356c9f62 100644
--- a/frontend/src/i18n/__tests__/openaiFastPolicyLocales.spec.ts
+++ b/frontend/src/i18n/__tests__/openaiFastPolicyLocales.spec.ts
@@ -27,4 +27,30 @@ describe('OpenAI Fast/Flex policy locale keys', () => {
       removeUser: 'Remove user'
     })
   })
+
+  it('describes target and other-model actions without whitelist terminology', () => {
+    expect(zh.admin.settings.openaiFastPolicy).toMatchObject({
+      tierAll: '全部 tier 值',
+      modelWhitelist: '目标模型',
+      fallbackAction: '其他模型处理方式',
+      summaryTargetModels: '目标模型',
+      summaryOtherModels: '其他模型'
+    })
+    expect(zh.admin.settings.openaiFastPolicy.modelWhitelistHint).toContain(
+      '留空时“处理方式”应用于全部模型'
+    )
+    expect(zh.admin.settings.openaiFastPolicy.modelWhitelistHint).not.toContain('白名单')
+
+    expect(en.admin.settings.openaiFastPolicy).toMatchObject({
+      tierAll: 'All tier values',
+      modelWhitelist: 'Target models',
+      fallbackAction: 'Other models action',
+      summaryTargetModels: 'Target models',
+      summaryOtherModels: 'Other models'
+    })
+    expect(en.admin.settings.openaiFastPolicy.modelWhitelistHint).toContain(
+      'Leave empty to apply Action to all models'
+    )
+    expect(en.admin.settings.openaiFastPolicy.modelWhitelistHint).not.toContain('whitelist')
+  })
 })
diff --git a/frontend/src/i18n/locales/en/admin/settings.ts b/frontend/src/i18n/locales/en/admin/settings.ts
index b20e2ebeee09..9aff76d251c1 100644
--- a/frontend/src/i18n/locales/en/admin/settings.ts
+++ b/frontend/src/i18n/locales/en/admin/settings.ts
@@ -1173,7 +1173,7 @@ export default {
         addRule: 'Add rule',
         saveHint: 'Saved together with system settings (click the global Save button at the bottom of the page).',
         serviceTier: 'service_tier match',
-        tierAll: 'All tiers',
+        tierAll: 'All tier values',
         tierPriority: 'priority (fast)',
         tierFlex: 'flex',
         action: 'Action',
@@ -1196,13 +1196,22 @@ export default {
         errorMessage: 'Error message',
         errorMessagePlaceholder: 'Custom error message when blocked',
         errorMessageHint: 'Leave empty for the default message.',
-        modelWhitelist: 'Model whitelist',
-        modelWhitelistHint: 'Leave empty to apply to all models. Supports exact match and wildcard prefix (e.g., gpt-5.5*).',
-        modelPatternPlaceholder: 'e.g., gpt-5.5 or gpt-5.5*',
-        addModelPattern: 'Add model pattern',
-        fallbackAction: 'Fallback action',
-        fallbackActionHint: 'Action for models not matching the whitelist.',
-        fallbackErrorMessagePlaceholder: 'Custom error message when non-whitelisted models are blocked'
+        modelWhitelist: 'Target models',
+        modelWhitelistHint: 'Models in this list use Action; models outside it use Other models action. Leave empty to apply Action to all models. Supports exact matches and wildcard prefixes (e.g., gpt-5.6*).',
+        modelPatternPlaceholder: 'e.g., gpt-5.6-sol or gpt-5.6*',
+        addModelPattern: 'Add target model',
+        fallbackAction: 'Other models action',
+        fallbackActionHint: 'Applies only to models outside the target list.',
+        fallbackErrorMessagePlaceholder: 'Custom error message when other models are blocked',
+        summaryTargetModels: 'Target models',
+        summaryAllModels: 'All models',
+        summaryOtherModels: 'Other models',
+        summaryAction: {
+          pass: 'Pass',
+          filter: 'Filter',
+          block: 'Block',
+          force_priority: 'Force priority'
+        }
       },
       wechatConnect: {
         title: 'WeChat Connect',
diff --git a/frontend/src/i18n/locales/zh/admin/settings.ts b/frontend/src/i18n/locales/zh/admin/settings.ts
index 3967f25ed73a..82a932e5641f 100644
--- a/frontend/src/i18n/locales/zh/admin/settings.ts
+++ b/frontend/src/i18n/locales/zh/admin/settings.ts
@@ -1167,7 +1167,7 @@ export default {
         addRule: '新增规则',
         saveHint: '保存时随系统设置一起提交（点击页面底部「保存」按钮）。',
         serviceTier: 'service_tier 匹配',
-        tierAll: '全部 tier',
+        tierAll: '全部 tier 值',
         tierPriority: 'priority（fast）',
         tierFlex: 'flex',
         action: '处理方式',
@@ -1190,13 +1190,22 @@ export default {
         errorMessage: '错误消息',
         errorMessagePlaceholder: '拦截时返回的自定义错误消息',
         errorMessageHint: '留空则使用默认错误消息。',
-        modelWhitelist: '模型白名单',
-        modelWhitelistHint: '留空表示对所有模型生效；支持精确匹配与通配符（如 gpt-5.5*）。',
-        modelPatternPlaceholder: '例如: gpt-5.5 或 gpt-5.5*',
-        addModelPattern: '添加模型规则',
-        fallbackAction: '未匹配模型处理方式',
-        fallbackActionHint: '当请求模型不在白名单中时的处理方式。',
-        fallbackErrorMessagePlaceholder: '未匹配模型被拦截时返回的自定义错误消息'
+        modelWhitelist: '目标模型',
+        modelWhitelistHint: '列表内模型使用“处理方式”，列表外模型使用“其他模型处理方式”。留空时“处理方式”应用于全部模型。支持精确匹配与前缀通配符（如 gpt-5.6*）。',
+        modelPatternPlaceholder: '例如: gpt-5.6-sol 或 gpt-5.6*',
+        addModelPattern: '添加目标模型',
+        fallbackAction: '其他模型处理方式',
+        fallbackActionHint: '仅应用于目标列表之外的模型。',
+        fallbackErrorMessagePlaceholder: '其他模型被拦截时返回的自定义错误消息',
+        summaryTargetModels: '目标模型',
+        summaryAllModels: '全部模型',
+        summaryOtherModels: '其他模型',
+        summaryAction: {
+          pass: '透传',
+          filter: '过滤',
+          block: '拦截',
+          force_priority: '强制 priority'
+        }
       },
       wechatConnect: {
         title: '微信登录',
diff --git a/frontend/src/views/admin/SettingsView.vue b/frontend/src/views/admin/SettingsView.vue
index b153865ff3bc..bb4d021b25de 100644
--- a/frontend/src/views/admin/SettingsView.vue
+++ b/frontend/src/views/admin/SettingsView.vue
@@ -1128,6 +1128,47 @@
                   </button>
                 </div>
 
+                <div
+                  class="mb-4 flex flex-wrap items-center gap-x-2 gap-y-1 text-xs text-gray-500 dark:text-gray-400"
+                  :data-testid="`openai-fast-policy-summary-${ruleIndex}`"
+                >
+                  <span class="font-medium text-gray-700 dark:text-gray-300">
+                    {{
+                      t(
+                        hasOpenAIFastPolicyTargetModels(rule)
+                          ? "admin.settings.openaiFastPolicy.summaryTargetModels"
+                          : "admin.settings.openaiFastPolicy.summaryAllModels",
+                      )
+                    }}
+                  </span>
+                  <span aria-hidden="true">→</span>
+                  <span
+                    class="inline-flex items-center rounded bg-primary-50 px-2 py-0.5 font-medium text-primary-700 dark:bg-primary-900/30 dark:text-primary-300"
+                  >
+                    {{ openaiFastPolicyActionSummary(rule.action) }}
+                  </span>
+                  <template v-if="hasOpenAIFastPolicyTargetModels(rule)">
+                    <span aria-hidden="true">·</span>
+                    <span class="font-medium text-gray-700 dark:text-gray-300">
+                      {{
+                        t(
+                          "admin.settings.openaiFastPolicy.summaryOtherModels",
+                        )
+                      }}
+                    </span>
+                    <span aria-hidden="true">→</span>
+                    <span
+                      class="inline-flex items-center rounded bg-gray-100 px-2 py-0.5 font-medium text-gray-700 dark:bg-dark-600 dark:text-gray-300"
+                    >
+                      {{
+                        openaiFastPolicyActionSummary(
+                          rule.fallback_action || "pass",
+                        )
+                      }}
+                    </span>
+                  </template>
+                </div>
+
                 <div class="grid grid-cols-1 gap-4 md:grid-cols-3">
                   <!-- Service Tier -->
                   <div>
@@ -1227,14 +1268,23 @@
                   </p>
                 </div>
 
-                <!-- Model Whitelist -->
-                <div class="mt-3">
+                <!-- Target Models -->
+                <div
+                  class="mt-3"
+                  role="group"
+                  :aria-labelledby="`openai-fast-policy-models-label-${ruleIndex}`"
+                  :aria-describedby="`openai-fast-policy-models-hint-${ruleIndex}`"
+                >
                   <label
+                    :id="`openai-fast-policy-models-label-${ruleIndex}`"
                     class="mb-1 block text-xs font-medium text-gray-600 dark:text-gray-400"
                   >
                     {{ t("admin.settings.openaiFastPolicy.modelWhitelist") }}
                   </label>
-                  <p class="mb-2 text-xs text-gray-400 dark:text-gray-500">
+                  <p
+                    :id="`openai-fast-policy-models-hint-${ruleIndex}`"
+                    class="mb-2 text-xs text-gray-400 dark:text-gray-500"
+                  >
                     {{
                       t("admin.settings.openaiFastPolicy.modelWhitelistHint")
                     }}
@@ -1298,11 +1348,9 @@
                   </button>
                 </div>
 
-                <!-- Fallback Action (only when model_whitelist is non-empty) -->
+                <!-- Other Models Action (only when target models are non-empty) -->
                 <div
-                  v-if="
-                    rule.model_whitelist && rule.model_whitelist.length > 0
-                  "
+                  v-if="hasOpenAIFastPolicyTargetModels(rule)"
                   class="mt-3"
                 >
                   <label
@@ -11975,6 +12023,16 @@ const openaiFastPolicyActionOptions = computed(() => [
   { value: "block", label: t("admin.settings.openaiFastPolicy.actionBlock") },
 ]);
 
+function openaiFastPolicyActionSummary(
+  action: OpenAIFastPolicyRule["action"],
+) {
+  return t(`admin.settings.openaiFastPolicy.summaryAction.${action}`);
+}
+
+function hasOpenAIFastPolicyTargetModels(rule: OpenAIFastPolicyRule) {
+  return Boolean(rule.model_whitelist?.some((pattern) => pattern.trim() !== ""));
+}
+
 const openaiFastPolicyScopeOptions = computed(() => [
   { value: "all", label: t("admin.settings.openaiFastPolicy.scopeAll") },
   { value: "oauth", label: t("admin.settings.openaiFastPolicy.scopeOAuth") },
diff --git a/frontend/src/views/admin/__tests__/SettingsView.spec.ts b/frontend/src/views/admin/__tests__/SettingsView.spec.ts
index aa81d8f4763e..bf6fe5e68dc4 100644
--- a/frontend/src/views/admin/__tests__/SettingsView.spec.ts
+++ b/frontend/src/views/admin/__tests__/SettingsView.spec.ts
@@ -229,6 +229,11 @@ vi.mock("vue-i18n", async () => {
     "admin.settings.upstreamBillingProbe.intervalHint": "范围 5–1440 分钟。",
     "admin.settings.upstreamBillingProbe.saved": "上游倍率自动探测设置已保存",
     "admin.settings.upstreamBillingProbe.saveFailed": "保存上游倍率自动探测设置失败",
+    "admin.settings.openaiFastPolicy.summaryTargetModels": "目标模型",
+    "admin.settings.openaiFastPolicy.summaryAllModels": "全部模型",
+    "admin.settings.openaiFastPolicy.summaryOtherModels": "其他模型",
+    "admin.settings.openaiFastPolicy.summaryAction.filter": "过滤",
+    "admin.settings.openaiFastPolicy.summaryAction.pass": "透传",
     "admin.settings.security.passkeyDeploymentHint":
       "请由服务器运维在部署配置中将 webauthn.enabled 设为 true，填写 webauthn.rp_id（仅域名）与 webauthn.rp_origins（完整 HTTPS 来源），然后重启服务。",
     "admin.settings.site.uploadImage": "上传图片",
@@ -1261,6 +1266,43 @@ describe("admin SettingsView payment visible method controls", () => {
     expect(wrapper.text()).not.toContain("OpenAI 高级调度器");
   });
 
+  it("summarizes target and other-model actions, then switches to all models", async () => {
+    getSettings.mockResolvedValueOnce({
+      ...baseSettingsResponse,
+      openai_fast_policy_settings: {
+        rules: [
+          {
+            service_tier: "all",
+            action: "filter",
+            scope: "all",
+            model_whitelist: ["gpt-5.6-sol"],
+            fallback_action: "pass",
+          },
+        ],
+      },
+    });
+    const wrapper = mountView();
+
+    await flushPromises();
+    await openGatewayTab(wrapper);
+
+    const summary = wrapper.get('[data-testid="openai-fast-policy-summary-0"]');
+    expect(summary.text()).toContain("目标模型");
+    expect(summary.text()).toContain("过滤");
+    expect(summary.text()).toContain("其他模型");
+    expect(summary.text()).toContain("透传");
+
+    await wrapper
+      .get(
+        '[role="group"][aria-labelledby="openai-fast-policy-models-label-0"] input[type="text"]',
+      )
+      .setValue("");
+    expect(summary.text()).toContain("全部模型");
+    expect(summary.text()).toContain("过滤");
+    expect(summary.text()).not.toContain("其他模型");
+    expect(summary.text()).not.toContain("透传");
+  });
+
   it("loads and saves upstream billing probe settings from the gateway tab", async () => {
     getUpstreamBillingProbeSettings.mockResolvedValueOnce({
       enabled: false,

From 5f19433103f5d5fb78ca31fced71e86bbf235868 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Mon, 17 Aug 2026 10:52:57 +0800
Subject: [PATCH 26/68] fix(ops): avoid single-insert fallback after batch
 failure

---
 backend/internal/service/ops_service.go            | 13 ++-----------
 backend/internal/service/ops_service_batch_test.go |  6 +++---
 2 files changed, 5 insertions(+), 14 deletions(-)

diff --git a/backend/internal/service/ops_service.go b/backend/internal/service/ops_service.go
index b1d9b9f57c92..c651b9184d90 100644
--- a/backend/internal/service/ops_service.go
+++ b/backend/internal/service/ops_service.go
@@ -439,17 +439,8 @@ func (s *OpsService) RecordErrorBatch(ctx context.Context, entries []*OpsInsertE
 	}
 
 	if _, err := s.opsRepo.BatchInsertErrorLogs(ctx, prepared); err != nil {
-		log.Printf("[Ops] RecordErrorBatch failed, fallback to single inserts: %v", err)
-		var firstErr error
-		for _, entry := range prepared {
-			if _, insertErr := s.opsRepo.InsertErrorLog(ctx, entry); insertErr != nil {
-				log.Printf("[Ops] RecordErrorBatch fallback insert failed: %v", insertErr)
-				if firstErr == nil {
-					firstErr = insertErr
-				}
-			}
-		}
-		return firstErr
+		log.Printf("[Ops] RecordErrorBatch failed: %v", err)
+		return err
 	}
 	return nil
 }
diff --git a/backend/internal/service/ops_service_batch_test.go b/backend/internal/service/ops_service_batch_test.go
index a027f4873729..e6a87191dc46 100644
--- a/backend/internal/service/ops_service_batch_test.go
+++ b/backend/internal/service/ops_service_batch_test.go
@@ -69,7 +69,7 @@ func TestOpsServiceRecordErrorBatch_SanitizesAndBatches(t *testing.T) {
 	require.False(t, second.CreatedAt.IsZero())
 }
 
-func TestOpsServiceRecordErrorBatch_FallsBackToSingleInsert(t *testing.T) {
+func TestOpsServiceRecordErrorBatch_DoesNotFallbackToSingleInsertsWhenBatchFails(t *testing.T) {
 	t.Parallel()
 
 	var (
@@ -92,9 +92,9 @@ func TestOpsServiceRecordErrorBatch_FallsBackToSingleInsert(t *testing.T) {
 		{ErrorMessage: "first"},
 		{ErrorMessage: "second"},
 	})
-	require.NoError(t, err)
+	require.Error(t, err)
 	require.Equal(t, 1, batchCalls)
-	require.Equal(t, 2, singleCalls)
+	require.Zero(t, singleCalls)
 }
 
 func TestOpsServiceRecordErrorPersistsExplicitAccountAuthStatusZero(t *testing.T) {

From 79c2eb50201568b47f6ebff4d9c4665347d4eba4 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Mon, 17 Aug 2026 10:59:43 +0800
Subject: [PATCH 27/68] fix: skip expiry reminders without SMTP config

---
 backend/internal/service/leader_lock_test.go  | 12 ++--
 .../service/subscription_expiry_service.go    | 29 +++++++++
 .../subscription_expiry_service_test.go       | 61 +++++++++++++++++--
 3 files changed, 92 insertions(+), 10 deletions(-)

diff --git a/backend/internal/service/leader_lock_test.go b/backend/internal/service/leader_lock_test.go
index aba1c62c3ee3..a2adf0126733 100644
--- a/backend/internal/service/leader_lock_test.go
+++ b/backend/internal/service/leader_lock_test.go
@@ -92,10 +92,10 @@ func TestSubscriptionExpiryService_ReminderSkipsScanWhenNotLeader(t *testing.T)
 	_, _ = cache.TryAcquireLeaderLock(context.Background(), subscriptionExpiryReminderLeaderLockKey, "peer", time.Minute)
 
 	repo := &subscriptionExpiryRepoStub{}
-	settingRepo := &subscriptionExpirySettingRepoStub{values: map[string]string{}}
+	settingRepo := &subscriptionExpirySettingRepoStub{values: map[string]string{SettingKeySMTPHost: "smtp.example.com"}}
 	svc := NewSubscriptionExpiryService(repo, time.Minute)
 	svc.SetSettingRepository(settingRepo)
-	svc.SetNotificationEmailService(NewNotificationEmailService(settingRepo, nil))
+	svc.SetNotificationEmailService(NewNotificationEmailService(settingRepo, NewEmailService(settingRepo, nil)))
 	svc.SetLeaderLock(cache, nil)
 
 	svc.sendExpiryReminders(context.Background())
@@ -105,10 +105,10 @@ func TestSubscriptionExpiryService_ReminderSkipsScanWhenNotLeader(t *testing.T)
 
 func TestSubscriptionExpiryService_ReminderScansWhenLeader(t *testing.T) {
 	repo := &subscriptionExpiryRepoStub{}
-	settingRepo := &subscriptionExpirySettingRepoStub{values: map[string]string{}}
+	settingRepo := &subscriptionExpirySettingRepoStub{values: map[string]string{SettingKeySMTPHost: "smtp.example.com"}}
 	svc := NewSubscriptionExpiryService(repo, time.Minute)
 	svc.SetSettingRepository(settingRepo)
-	svc.SetNotificationEmailService(NewNotificationEmailService(settingRepo, nil))
+	svc.SetNotificationEmailService(NewNotificationEmailService(settingRepo, NewEmailService(settingRepo, nil)))
 	svc.SetLeaderLock(&fakeLeaderLockCache{}, nil)
 
 	svc.sendExpiryReminders(context.Background())
@@ -127,10 +127,10 @@ func TestSubscriptionExpiryService_ReminderRunsEveryCycleSingleInstance(t *testi
 	for name, cache := range cases {
 		t.Run(name, func(t *testing.T) {
 			repo := &subscriptionExpiryRepoStub{}
-			settingRepo := &subscriptionExpirySettingRepoStub{values: map[string]string{}}
+			settingRepo := &subscriptionExpirySettingRepoStub{values: map[string]string{SettingKeySMTPHost: "smtp.example.com"}}
 			svc := NewSubscriptionExpiryService(repo, time.Minute)
 			svc.SetSettingRepository(settingRepo)
-			svc.SetNotificationEmailService(NewNotificationEmailService(settingRepo, nil))
+			svc.SetNotificationEmailService(NewNotificationEmailService(settingRepo, NewEmailService(settingRepo, nil)))
 			svc.SetLeaderLock(cache, nil)
 
 			// Three consecutive cycles, mimicking the ticker loop.
diff --git a/backend/internal/service/subscription_expiry_service.go b/backend/internal/service/subscription_expiry_service.go
index c93c763f029e..e27da3661d88 100644
--- a/backend/internal/service/subscription_expiry_service.go
+++ b/backend/internal/service/subscription_expiry_service.go
@@ -15,6 +15,7 @@ import (
 )
 
 const (
+	subscriptionExpiryReminderSMTPWarningInterval = time.Minute
 	// subscriptionExpiryReminderLeaderLockKey gates the per-cycle reminder scan so
 	// that only one instance walks all active subscriptions and sends reminder
 	// emails, avoiding redundant full scans and duplicate emails.
@@ -37,6 +38,9 @@ type SubscriptionExpiryService struct {
 	lockCache  LeaderLockCache
 	db         *sql.DB
 	instanceID string
+
+	smtpWarningMu   sync.Mutex
+	lastSMTPWarning time.Time
 }
 
 func NewSubscriptionExpiryService(userSubRepo UserSubscriptionRepository, interval time.Duration) *SubscriptionExpiryService {
@@ -121,6 +125,9 @@ func (s *SubscriptionExpiryService) sendExpiryReminders(ctx context.Context) {
 	if !s.expiryReminderEnabled(ctx) {
 		return
 	}
+	if !s.smtpConfigured(ctx) {
+		return
+	}
 
 	// Multi-instance guard: only the leader walks every active subscription and
 	// sends reminders, avoiding N× full scans and duplicate reminder emails.
@@ -159,6 +166,28 @@ func (s *SubscriptionExpiryService) expiryReminderEnabled(ctx context.Context) b
 	return !isFalseSettingValue(value)
 }
 
+func (s *SubscriptionExpiryService) smtpConfigured(ctx context.Context) bool {
+	if s == nil || s.notificationEmailService == nil || s.notificationEmailService.emailService == nil {
+		return false
+	}
+	_, err := s.notificationEmailService.emailService.GetSMTPConfig(ctx)
+	if err == nil {
+		return true
+	}
+	if errors.Is(err, ErrEmailNotConfigured) {
+		s.smtpWarningMu.Lock()
+		defer s.smtpWarningMu.Unlock()
+		now := time.Now()
+		if s.lastSMTPWarning.IsZero() || now.Sub(s.lastSMTPWarning) >= subscriptionExpiryReminderSMTPWarningInterval {
+			log.Printf("[SubscriptionExpiry] SMTP is not configured; skipping expiry reminders")
+			s.lastSMTPWarning = now
+		}
+		return false
+	}
+	log.Printf("[SubscriptionExpiry] Read SMTP configuration failed; skipping expiry reminders: %v", err)
+	return false
+}
+
 func (s *SubscriptionExpiryService) sendExpiryReminderIfDue(ctx context.Context, sub *UserSubscription) {
 	if sub == nil || sub.User == nil || sub.Group == nil || sub.User.Email == "" {
 		return
diff --git a/backend/internal/service/subscription_expiry_service_test.go b/backend/internal/service/subscription_expiry_service_test.go
index 74b82e56c3b0..decb816e8490 100644
--- a/backend/internal/service/subscription_expiry_service_test.go
+++ b/backend/internal/service/subscription_expiry_service_test.go
@@ -1,8 +1,10 @@
 package service
 
 import (
+	"bytes"
 	"context"
 	"errors"
+	"log"
 	"testing"
 	"time"
 
@@ -116,8 +118,9 @@ func (r *subscriptionExpiryRepoStub) BatchUpdateExpiredStatus(context.Context) (
 }
 
 type subscriptionExpirySettingRepoStub struct {
-	values map[string]string
-	err    error
+	values   map[string]string
+	err      error
+	multiErr error
 }
 
 func (r *subscriptionExpirySettingRepoStub) Get(context.Context, string) (*Setting, error) {
@@ -139,8 +142,17 @@ func (r *subscriptionExpirySettingRepoStub) Set(context.Context, string, string)
 	return nil
 }
 
-func (r *subscriptionExpirySettingRepoStub) GetMultiple(context.Context, []string) (map[string]string, error) {
-	return nil, nil
+func (r *subscriptionExpirySettingRepoStub) GetMultiple(_ context.Context, keys []string) (map[string]string, error) {
+	if r.multiErr != nil {
+		return nil, r.multiErr
+	}
+	values := make(map[string]string, len(keys))
+	for _, key := range keys {
+		if value, ok := r.values[key]; ok {
+			values[key] = value
+		}
+	}
+	return values, nil
 }
 
 func (r *subscriptionExpirySettingRepoStub) SetMultiple(context.Context, map[string]string) error {
@@ -182,3 +194,44 @@ func TestSubscriptionExpiryService_ExpiryReminderSettingReadErrorFailsClosed(t *
 
 	require.False(t, svc.expiryReminderEnabled(context.Background()))
 }
+
+func TestSubscriptionExpiryService_MissingSMTPSkipsReminderScanAndLogsOncePerInterval(t *testing.T) {
+	repo := &subscriptionExpiryRepoStub{}
+	settingRepo := &subscriptionExpirySettingRepoStub{values: map[string]string{}}
+	emailService := NewEmailService(settingRepo, nil)
+	svc := NewSubscriptionExpiryService(repo, time.Minute)
+	svc.SetSettingRepository(settingRepo)
+	svc.SetNotificationEmailService(NewNotificationEmailService(settingRepo, emailService))
+
+	var logs bytes.Buffer
+	previousWriter := log.Writer()
+	previousFlags := log.Flags()
+	log.SetOutput(&logs)
+	log.SetFlags(0)
+	t.Cleanup(func() {
+		log.SetOutput(previousWriter)
+		log.SetFlags(previousFlags)
+	})
+
+	svc.sendExpiryReminders(context.Background())
+	svc.sendExpiryReminders(context.Background())
+
+	require.Zero(t, repo.listCalls)
+	require.Equal(t, 1, bytes.Count(logs.Bytes(), []byte("SMTP is not configured")))
+}
+
+func TestSubscriptionExpiryService_SMTPConfigReadErrorSkipsReminderScan(t *testing.T) {
+	repo := &subscriptionExpiryRepoStub{}
+	settingRepo := &subscriptionExpirySettingRepoStub{
+		values:   map[string]string{},
+		multiErr: errors.New("db down"),
+	}
+	emailService := NewEmailService(settingRepo, nil)
+	svc := NewSubscriptionExpiryService(repo, time.Minute)
+	svc.SetSettingRepository(settingRepo)
+	svc.SetNotificationEmailService(NewNotificationEmailService(settingRepo, emailService))
+
+	svc.sendExpiryReminders(context.Background())
+
+	require.Zero(t, repo.listCalls)
+}

From 1977810cf24fa36d939cce694e7808347e516259 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Mon, 17 Aug 2026 11:04:50 +0800
Subject: [PATCH 28/68] fix(frontend): isolate account helper data loading

---
 frontend/src/views/admin/AccountsView.vue     | 19 +++++++++++++------
 .../AccountsView.usageWindowsHint.spec.ts     | 15 ++++++++++++++-
 2 files changed, 27 insertions(+), 7 deletions(-)

diff --git a/frontend/src/views/admin/AccountsView.vue b/frontend/src/views/admin/AccountsView.vue
index cfab5c07e3a8..982c21b6f5a6 100644
--- a/frontend/src/views/admin/AccountsView.vue
+++ b/frontend/src/views/admin/AccountsView.vue
@@ -2454,12 +2454,19 @@ onMounted(async () => {
 
   load()
   loadUpstreamBillingProbeGlobalState()
-  try {
-    const [p, g] = await Promise.all([adminAPI.proxies.getAll(), adminAPI.groups.getAll()])
-    proxies.value = p
-    groups.value = g
-  } catch (error) {
-    console.error('Failed to load proxies/groups:', error)
+  const [proxiesResult, groupsResult] = await Promise.allSettled([
+    adminAPI.proxies.getAll(),
+    adminAPI.groups.getAll()
+  ])
+  if (proxiesResult.status === 'fulfilled') {
+    proxies.value = proxiesResult.value
+  } else {
+    console.error('Failed to load proxies:', proxiesResult.reason)
+  }
+  if (groupsResult.status === 'fulfilled') {
+    groups.value = groupsResult.value
+  } else {
+    console.error('Failed to load groups:', groupsResult.reason)
   }
   window.addEventListener('scroll', handleScroll, true)
   window.addEventListener('resize', handleViewportResize)
diff --git a/frontend/src/views/admin/__tests__/AccountsView.usageWindowsHint.spec.ts b/frontend/src/views/admin/__tests__/AccountsView.usageWindowsHint.spec.ts
index 559536315652..9f25d0dcd342 100644
--- a/frontend/src/views/admin/__tests__/AccountsView.usageWindowsHint.spec.ts
+++ b/frontend/src/views/admin/__tests__/AccountsView.usageWindowsHint.spec.ts
@@ -101,7 +101,10 @@ function mountView() {
         Pagination: true,
         ConfirmDialog: true,
         AccountTableActions: { template: '<div><slot name="beforeCreate" /><slot name="after" /></div>' },
-        AccountTableFilters: { template: '<div></div>' },
+        AccountTableFilters: {
+          props: ['groups'],
+          template: '<div data-test="account-filters" :data-group-count="groups.length"></div>'
+        },
         AccountBulkActionsBar: true,
         AccountActionMenu: true,
         ImportDataModal: true,
@@ -155,6 +158,16 @@ describe('admin AccountsView usage windows hint', () => {
     getAllGroups.mockResolvedValue([])
   })
 
+  it('keeps groups available when loading proxies fails', async () => {
+    getAllProxies.mockRejectedValue(new Error('proxy service unavailable'))
+    getAllGroups.mockResolvedValue([{ id: 7, name: 'production' }])
+
+    const wrapper = mountView()
+    await flushPromises()
+
+    expect(wrapper.get('[data-test="account-filters"]').attributes('data-group-count')).toBe('1')
+  })
+
   it('renders an explanatory tooltip next to the usage windows column header', async () => {
     const wrapper = mountView()
     await flushPromises()

From 4b667ccd45747162ac58545cdbb6f6d88737bf04 Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Mon, 17 Aug 2026 04:07:06 +0000
Subject: [PATCH 29/68] =?UTF-8?q?fix(review):=20=E5=A4=84=E7=90=86=20PR=20?=
 =?UTF-8?q?#5666=20=E5=9B=9B=E4=B8=AA=E9=98=BB=E6=96=AD=E9=A1=B9=EF=BC=88B?=
 =?UTF-8?q?1-B4=EF=BC=89?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- B1: 移除根目录 docker-compose.yml（个人镜像测试产物混入）
- B2: 新增迁移 224 放宽 user_platform_quotas CHECK 至 8 平台，补
  BulkInsertInitial 国产平台集成测试与迁移内容断言测试
- B3: CC/Responses×anthropic 四个上游读循环接入间隔超时泵
  （gateway.stream_data_interval_timeout，默认 180s）：上游挂住 SSE
  不发数据也不断连时结束排水/组装、关闭 resp.Body 归还连接池位，
  排水超时仍按已累计 usage 返回（与 messages 主路径同语义）
- B4: 配额/余额探测发起前过 cnValidateProbeURL（与网关转发同一套
  security.url_allowlist 策略），被拒时零出站、API key 不离开本机
---
 ...er_platform_quota_repo_integration_test.go |  29 ++
 .../service/cn_provider_balance_service.go    |   7 +
 .../internal/service/cn_provider_probe_url.go |  48 +++
 .../service/cn_provider_probe_url_test.go     | 129 ++++++++
 .../service/cn_provider_quota_service.go      |   8 +
 .../openai_gateway_anthropic_native_pump.go   | 126 ++++++++
 ...enai_gateway_anthropic_native_pump_test.go | 244 ++++++++++++++
 ...teway_chat_completions_anthropic_native.go | 111 +++++--
 ...enai_gateway_responses_anthropic_native.go | 106 ++++--
 ..._user_platform_quotas_add_cn_providers.sql |  17 +
 ...tform_quota_cn_providers_migration_test.go |  22 ++
 docker-compose.yml                            | 302 ------------------
 12 files changed, 795 insertions(+), 354 deletions(-)
 create mode 100644 backend/internal/service/cn_provider_probe_url.go
 create mode 100644 backend/internal/service/cn_provider_probe_url_test.go
 create mode 100644 backend/internal/service/openai_gateway_anthropic_native_pump.go
 create mode 100644 backend/internal/service/openai_gateway_anthropic_native_pump_test.go
 create mode 100644 backend/migrations/224_user_platform_quotas_add_cn_providers.sql
 create mode 100644 backend/migrations/user_platform_quota_cn_providers_migration_test.go
 delete mode 100644 docker-compose.yml

diff --git a/backend/internal/repository/user_platform_quota_repo_integration_test.go b/backend/internal/repository/user_platform_quota_repo_integration_test.go
index 4a8d1b58967f..1bd4b08786c9 100644
--- a/backend/internal/repository/user_platform_quota_repo_integration_test.go
+++ b/backend/internal/repository/user_platform_quota_repo_integration_test.go
@@ -98,6 +98,35 @@ func TestUserPlatformQuotaRepository_BulkInsertInitial_GrokAllowed(t *testing.T)
 	require.InDelta(t, 9.0, *rec.DailyLimitUSD, 1e-9)
 }
 
+// TestUserPlatformQuotaRepository_BulkInsertInitial_CNProvidersAllowed 回归迁移 224：
+// kimi/zhipu/deepseek 平台必须能写入 user_platform_quotas（CHECK 约束已含国产供应商）。
+// 历史 bug：三个平台不在约束内 → 注册预填充 8 平台默认配额时整条多行 INSERT 中止 →
+// fail-open 吞错 → 新用户拿到零条配额记录（缺失配额行 = 无限额）。
+func TestUserPlatformQuotaRepository_BulkInsertInitial_CNProvidersAllowed(t *testing.T) {
+	ctx := context.Background()
+	tx := testEntTx(t)
+	txCtx := dbent.NewTxContext(ctx, tx)
+	client := tx.Client()
+
+	userID := mustCreateUserForQuota(t, client)
+	repo := NewUserPlatformQuotaRepository(client)
+
+	daily := 12.0
+	records := []UserPlatformQuotaRecord{
+		{UserID: userID, Platform: "kimi", DailyLimitUSD: &daily},
+		{UserID: userID, Platform: "zhipu"},
+		{UserID: userID, Platform: "deepseek"},
+	}
+	require.NoError(t, repo.BulkInsertInitial(txCtx, records),
+		"kimi/zhipu/deepseek 平台应可写入（迁移 224 后 CHECK 约束已含国产供应商）")
+
+	for _, platform := range []string{"kimi", "zhipu", "deepseek"} {
+		rec, err := repo.GetByUserPlatform(txCtx, userID, platform)
+		require.NoError(t, err)
+		require.NotNil(t, rec, "%s 配额行应已写入", platform)
+	}
+}
+
 func TestUserPlatformQuotaRepository_GetByUserPlatform(t *testing.T) {
 	ctx := context.Background()
 	tx := testEntTx(t)
diff --git a/backend/internal/service/cn_provider_balance_service.go b/backend/internal/service/cn_provider_balance_service.go
index 2f306e8b6235..d01c9b6d087c 100644
--- a/backend/internal/service/cn_provider_balance_service.go
+++ b/backend/internal/service/cn_provider_balance_service.go
@@ -125,6 +125,13 @@ func (s *CNProviderBalanceService) queryBalance(ctx context.Context, accountID i
 	}
 
 	targetURL := cnBalanceURL(account)
+	// 探测发起前过出站 URL 安全策略（与网关转发/Grok 探测同一套校验）：
+	// DeepSeek 端点由账号 base_url 衍生，不得把 API key 发往策略外主机。
+	validatedURL, err := cnValidateProbeURL(s.cfg, targetURL)
+	if err != nil {
+		return nil, infraerrors.New(http.StatusForbidden, "CN_BALANCE_URL_REJECTED", err.Error())
+	}
+	targetURL = validatedURL
 	proxyURL := s.resolveProxyURL(ctx, account)
 	callCtx, cancel := context.WithTimeout(ctx, cnBalanceUpstreamTimeout)
 	defer cancel()
diff --git a/backend/internal/service/cn_provider_probe_url.go b/backend/internal/service/cn_provider_probe_url.go
new file mode 100644
index 000000000000..95b0ff783fcd
--- /dev/null
+++ b/backend/internal/service/cn_provider_probe_url.go
@@ -0,0 +1,48 @@
+package service
+
+// CN 供应商探测端点的出站 URL 安全策略校验（配额/余额探测共用）。
+//
+// 背景（review B4）：这两条探测路径会把账号 API key 发往 base_url 衍生端点，
+// 此前完全绕过 security.url_allowlist——在加固部署里构成任意外发与内网探测
+// 面（本项目此前发生过账号测试 SSRF 生产事件）。与网关转发
+// （validateUpstreamBaseURL）、Grok 探测（grokOperatorPolicyValidator）一致，
+// 探测发起前必须过同一套运营者策略。
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/util/urlvalidator"
+)
+
+// cnValidateProbeURL 按全局出站 URL 安全策略校验探测端点，返回规范化 URL。
+// 白名单开启时强制 UpstreamHosts（阻断私网与未列名主机）；关闭时仅做格式
+// 校验（HTTP 允许与否跟随配置）；cfg 为 nil 时退化为纯格式校验。
+func cnValidateProbeURL(cfg *config.Config, raw string) (string, error) {
+	trimmed := strings.TrimSpace(raw)
+	if trimmed == "" {
+		return "", errors.New("probe url is required")
+	}
+	if cfg != nil && cfg.Security.URLAllowlist.Enabled {
+		normalized, err := urlvalidator.ValidateHTTPSURL(trimmed, urlvalidator.ValidationOptions{
+			AllowedHosts:     cfg.Security.URLAllowlist.UpstreamHosts,
+			RequireAllowlist: true,
+			AllowPrivate:     cfg.Security.URLAllowlist.AllowPrivateHosts,
+		})
+		if err != nil {
+			return "", fmt.Errorf("probe target rejected by URL security policy: %w", err)
+		}
+		return normalized, nil
+	}
+	var allowInsecureHTTP bool
+	if cfg != nil {
+		allowInsecureHTTP = cfg.Security.URLAllowlist.AllowInsecureHTTP
+	}
+	normalized, err := urlvalidator.ValidateURLFormat(trimmed, allowInsecureHTTP)
+	if err != nil {
+		return "", fmt.Errorf("probe target rejected by URL security policy: %w", err)
+	}
+	return normalized, nil
+}
diff --git a/backend/internal/service/cn_provider_probe_url_test.go b/backend/internal/service/cn_provider_probe_url_test.go
new file mode 100644
index 000000000000..f28508840053
--- /dev/null
+++ b/backend/internal/service/cn_provider_probe_url_test.go
@@ -0,0 +1,129 @@
+package service
+
+// CN 供应商探测端点 URL 安全策略回归测试（review B4）：
+// 配额/余额探测不得绕过 security.url_allowlist——base_url 衍生的探测端点
+// 必须先过运营者策略，被拒绝时不得发起任何上游请求（API key 不出站）。
+
+import (
+	"context"
+	"net/http"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/tlsfingerprint"
+	"github.com/stretchr/testify/require"
+)
+
+func cnProbeAllowlistConfig(hosts ...string) *config.Config {
+	return &config.Config{
+		Security: config.SecurityConfig{
+			URLAllowlist: config.URLAllowlistConfig{
+				Enabled:       true,
+				UpstreamHosts: hosts,
+			},
+		},
+	}
+}
+
+func TestCNValidateProbeURL_AllowlistPolicy(t *testing.T) {
+	cfg := cnProbeAllowlistConfig("api.moonshot.cn", "api.deepseek.com")
+
+	// 白名单内主机放行（保留完整路径）。
+	ok, err := cnValidateProbeURL(cfg, "https://api.moonshot.cn/v1/users/me/balance")
+	require.NoError(t, err)
+	require.Equal(t, "https://api.moonshot.cn/v1/users/me/balance", ok)
+
+	// 白名单外主机拒绝。
+	_, err = cnValidateProbeURL(cfg, "https://relay.attacker.example/v1/usages")
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "rejected by URL security policy")
+
+	// 私网主机拒绝（内网探测面）。
+	_, err = cnValidateProbeURL(cfg, "http://169.254.169.254/latest/meta-data")
+	require.Error(t, err)
+
+	// 白名单关闭：仅格式校验，任意 https 主机放行。
+	formatOnly, err := cnValidateProbeURL(&config.Config{
+		Security: config.SecurityConfig{URLAllowlist: config.URLAllowlistConfig{Enabled: false}},
+	}, "https://relay.attacker.example/v1/usages")
+	require.NoError(t, err)
+	require.Equal(t, "https://relay.attacker.example/v1/usages", formatOnly)
+}
+
+// recordingHTTPUpstream 断言探测被策略拒绝时没有任何上游请求发出。
+type recordingHTTPUpstream struct{ calls int }
+
+func (u *recordingHTTPUpstream) Do(req *http.Request, proxyURL string, accountID int64, accountConcurrency int) (*http.Response, error) {
+	u.calls++
+	return nil, context.DeadlineExceeded
+}
+
+func (u *recordingHTTPUpstream) DoWithTLS(req *http.Request, proxyURL string, accountID int64, accountConcurrency int, profile *tlsfingerprint.Profile) (*http.Response, error) {
+	u.calls++
+	return nil, context.DeadlineExceeded
+}
+
+type fakeCNProbeAccountRepo struct {
+	AccountRepository
+	account *Account
+}
+
+func (r *fakeCNProbeAccountRepo) GetByID(ctx context.Context, id int64) (*Account, error) {
+	return r.account, nil
+}
+
+// kimi coding 账号的 base_url 指向中转（含 api.kimi.com/coding 路径段即可被识别
+// 为 kimi coding plan）→ 衍生额度端点落在中转主机上，白名单未列名必须拒绝。
+func TestCNProviderQuotaService_RejectsURLBlockedByPolicy(t *testing.T) {
+	repo := &fakeCNProbeAccountRepo{account: &Account{
+		ID: 1, Platform: PlatformKimi, Type: AccountTypeAPIKey, Status: StatusActive,
+		Credentials: map[string]any{
+			"account_mode": "coding",
+			"api_key":      "sk-test",
+			"base_url":     "https://relay.attacker.example/api.kimi.com/coding",
+		},
+	}}
+	upstream := &recordingHTTPUpstream{}
+	svc := NewCNProviderQuotaService(repo, nil, upstream, cnProbeAllowlistConfig("api.kimi.com"))
+
+	_, err := svc.QueryUsage(context.Background(), 1)
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "CN_QUOTA_URL_REJECTED")
+	require.Zero(t, upstream.calls, "probe must not issue any upstream request when URL policy rejects the target")
+}
+
+// deepseek payg 账号自定义 base_url → 余额端点落在中转主机上，必须先过策略。
+func TestCNProviderBalanceService_RejectsURLBlockedByPolicy(t *testing.T) {
+	repo := &fakeCNProbeAccountRepo{account: &Account{
+		ID: 2, Platform: PlatformDeepseek, Type: AccountTypeAPIKey, Status: StatusActive,
+		Credentials: map[string]any{
+			"account_mode": "payg",
+			"api_key":      "sk-test",
+			"base_url":     "https://relay.attacker.example",
+		},
+	}}
+	upstream := &recordingHTTPUpstream{}
+	svc := NewCNProviderBalanceService(repo, nil, upstream, cnProbeAllowlistConfig("api.deepseek.com"))
+
+	_, err := svc.QueryBalance(context.Background(), 2)
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "CN_BALANCE_URL_REJECTED")
+	require.Zero(t, upstream.calls, "probe must not issue any upstream request when URL policy rejects the target")
+}
+
+// 白名单包含官方主机的正常路径：URL 校验通过后才发出上游请求（此处允许到达
+// httpUpstream 层即视为通过校验，不发真实网络）。
+func TestCNProviderBalanceService_OfficialHostPassesValidation(t *testing.T) {
+	repo := &fakeCNProbeAccountRepo{account: &Account{
+		ID: 3, Platform: PlatformDeepseek, Type: AccountTypeAPIKey, Status: StatusActive,
+		Credentials: map[string]any{
+			"account_mode": "payg",
+			"api_key":      "sk-test",
+		},
+	}}
+	upstream := &recordingHTTPUpstream{}
+	svc := NewCNProviderBalanceService(repo, nil, upstream, cnProbeAllowlistConfig("api.deepseek.com"))
+
+	_, _ = svc.QueryBalance(context.Background(), 3)
+	require.Equal(t, 1, upstream.calls, "official host must pass URL policy and reach the upstream layer")
+}
diff --git a/backend/internal/service/cn_provider_quota_service.go b/backend/internal/service/cn_provider_quota_service.go
index e8fdba9c5609..f4cda8d6d121 100644
--- a/backend/internal/service/cn_provider_quota_service.go
+++ b/backend/internal/service/cn_provider_quota_service.go
@@ -144,6 +144,14 @@ func (s *CNProviderQuotaService) queryUsage(ctx context.Context, accountID int64
 		authHeader = apiKey // 智谱额度端点鉴权不加 Bearer 前缀
 	}
 
+	// 探测发起前过出站 URL 安全策略（与网关转发/Grok 探测同一套校验）：
+	// 端点多由账号 base_url 衍生，不得把 API key 发往策略外主机。
+	validatedURL, err := cnValidateProbeURL(s.cfg, targetURL)
+	if err != nil {
+		return nil, infraerrors.New(http.StatusForbidden, "CN_QUOTA_URL_REJECTED", err.Error())
+	}
+	targetURL = validatedURL
+
 	proxyURL := s.resolveProxyURL(ctx, account)
 	callCtx, cancel := context.WithTimeout(ctx, cnQuotaUpstreamTimeout)
 	defer cancel()
diff --git a/backend/internal/service/openai_gateway_anthropic_native_pump.go b/backend/internal/service/openai_gateway_anthropic_native_pump.go
new file mode 100644
index 000000000000..8a1967770568
--- /dev/null
+++ b/backend/internal/service/openai_gateway_anthropic_native_pump.go
@@ -0,0 +1,126 @@
+package service
+
+// 国产供应商 Anthropic 协议转换路径的上游 SSE 行泵。
+//
+// 这两条转换链（CC×anthropic / Responses×anthropic）的上游 ctx 是
+// WithoutCancel（detachStreamUpstreamContext）、http.Client 无整体 Timeout，
+// 客户端断开后的排水阶段若上游挂住 SSE（不发数据也不断连），
+// scanner.Scan() 将永久阻塞：goroutine 钉死、resp.Body 不归还、连接池位
+// 被占用、usage 永不落库。
+//
+// 与 handleAnthropicStreamingResponse / readOpenAICompatBufferedTerminal 的
+// 同类排水一致，本泵用 gateway.stream_data_interval_timeout（默认 180s）作为
+// 逐行读间隔上限，超时即向调用方返回 errAnthropicNativeStreamIdle，由调用方
+// 关闭 resp.Body 解除阻塞的读并结束排水。
+
+import (
+	"bufio"
+	"errors"
+	"io"
+	"time"
+)
+
+// errAnthropicNativeStreamIdle 表示上游流读间隔超时（见上方文件注释）。
+var errAnthropicNativeStreamIdle = errors.New("stream data interval timeout")
+
+// anthropicNativeLineEvent 是行泵交付的单次读取结果：line 为一行 SSE 文本，
+// err 为 scanner 读错误（流自然结束时 next 返回 io.EOF，不经过本字段）。
+type anthropicNativeLineEvent struct {
+	line string
+	err  error
+}
+
+// anthropicNativeLinePump 以独立 goroutine 泵送 scanner 的行，并对逐行到达
+// 间隔施加 interval 上限（<=0 表示禁用，保持无界读的旧行为）。
+type anthropicNativeLinePump struct {
+	events   chan anthropicNativeLineEvent
+	done     chan struct{}
+	timer    *time.Timer
+	interval time.Duration
+}
+
+// newAnthropicNativeLinePump 启动泵 goroutine；调用方 defer pump.stop()。
+func newAnthropicNativeLinePump(scanner *bufio.Scanner, interval time.Duration) *anthropicNativeLinePump {
+	p := &anthropicNativeLinePump{
+		events:   make(chan anthropicNativeLineEvent, 16),
+		done:     make(chan struct{}),
+		interval: interval,
+	}
+	if interval > 0 {
+		p.timer = time.NewTimer(interval)
+	}
+	go func() {
+		defer close(p.events)
+		for scanner.Scan() {
+			select {
+			case p.events <- anthropicNativeLineEvent{line: scanner.Text()}:
+			case <-p.done:
+				return
+			}
+		}
+		if err := scanner.Err(); err != nil {
+			select {
+			case p.events <- anthropicNativeLineEvent{err: err}:
+			case <-p.done:
+			}
+		}
+	}()
+	return p
+}
+
+// next 阻塞返回下一行。返回 io.EOF 表示上游正常收流；errAnthropicNativeStreamIdle
+// 表示 interval 内无任何数据到达（计时从收到上一行时起算，事件处理耗时不算入，
+// 与 readOpenAICompatBufferedTerminal 的 resetTimeout 语义一致）。
+func (p *anthropicNativeLinePump) next() (string, error) {
+	var timeoutCh <-chan time.Time
+	if p.timer != nil {
+		timeoutCh = p.timer.C
+	}
+	select {
+	case ev, ok := <-p.events:
+		if !ok {
+			return "", io.EOF
+		}
+		p.resetTimer()
+		return ev.line, ev.err
+	case <-timeoutCh:
+		return "", errAnthropicNativeStreamIdle
+	}
+}
+
+// resetTimer 在收到一行后重启间隔计时器。
+func (p *anthropicNativeLinePump) resetTimer() {
+	if p.timer == nil {
+		return
+	}
+	if !p.timer.Stop() {
+		select {
+		case <-p.timer.C:
+		default:
+		}
+	}
+	p.timer.Reset(p.interval)
+}
+
+// stop 终止泵 goroutine。注意：goroutine 若正阻塞在 scanner.Read 上，需由
+// 调用方关闭 resp.Body（间隔超时分支已做）才能真正退出。
+func (p *anthropicNativeLinePump) stop() {
+	close(p.done)
+	if p.timer != nil {
+		if !p.timer.Stop() {
+			select {
+			case <-p.timer.C:
+			default:
+			}
+		}
+	}
+}
+
+// anthropicNativeStreamInterval 返回本组转换路径适用的读间隔上限；
+// gateway.stream_data_interval_timeout <= 0 时视为禁用。
+func (s *OpenAIGatewayService) anthropicNativeStreamInterval() time.Duration {
+	if s.cfg != nil && s.cfg.Gateway.StreamDataIntervalTimeout > 0 {
+		return time.Duration(s.cfg.Gateway.StreamDataIntervalTimeout) * time.Second
+	}
+	return 0
+}
diff --git a/backend/internal/service/openai_gateway_anthropic_native_pump_test.go b/backend/internal/service/openai_gateway_anthropic_native_pump_test.go
new file mode 100644
index 000000000000..1e9e1b39a635
--- /dev/null
+++ b/backend/internal/service/openai_gateway_anthropic_native_pump_test.go
@@ -0,0 +1,244 @@
+package service
+
+// 国产供应商 Anthropic 协议转换路径的上游读间隔超时回归测试（B3）：
+// 上游挂住 SSE（不发数据也不断连）时，CC×anthropic / Responses×anthropic
+// 的读循环必须按 gateway.stream_data_interval_timeout 结束，而不是永久阻塞。
+
+import (
+	"bufio"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/gin-gonic/gin"
+)
+
+func newNativeAnthropicHangTestService(intervalSec int) *OpenAIGatewayService {
+	return &OpenAIGatewayService{
+		cfg: &config.Config{
+			Gateway: config.GatewayConfig{
+				StreamDataIntervalTimeout: intervalSec,
+				MaxLineSize:               defaultMaxLineSize,
+			},
+		},
+	}
+}
+
+func newHangingUpstreamResponse() (*http.Response, *io.PipeReader, *io.PipeWriter) {
+	pr, pw := io.Pipe()
+	return &http.Response{StatusCode: http.StatusOK, Body: pr, Header: http.Header{}}, pr, pw
+}
+
+// miniAnthropicSSEStream 是一段最小可转换的 Anthropic 事件流。
+func miniAnthropicSSEStream() string {
+	return strings.Join([]string{
+		"event: message_start",
+		`data: {"type":"message_start","message":{"id":"msg_1","type":"message","role":"assistant","content":[],"model":"glm-4.7","usage":{"input_tokens":10,"output_tokens":1}}}`,
+		"",
+		"event: content_block_start",
+		`data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}`,
+		"",
+		"event: content_block_delta",
+		`data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"Hello"}}`,
+		"",
+		"event: content_block_stop",
+		`data: {"type":"content_block_stop","index":0}`,
+		"",
+		"event: message_delta",
+		`data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":5}}`,
+		"",
+		"event: message_stop",
+		`data: {"type":"message_stop"}`,
+		"",
+		"",
+	}, "\n")
+}
+
+func TestAnthropicNativeLinePump_TimesOutWithoutData(t *testing.T) {
+	pr, _ := io.Pipe()
+	scanner := bufio.NewScanner(pr)
+	defer func() { _ = pr.Close() }()
+
+	pump := newAnthropicNativeLinePump(scanner, 50*time.Millisecond)
+	defer pump.stop()
+
+	start := time.Now()
+	_, err := pump.next()
+	if err == nil || !strings.Contains(err.Error(), "stream data interval timeout") {
+		t.Fatalf("expected interval timeout, got %v", err)
+	}
+	if elapsed := time.Since(start); elapsed > 5*time.Second {
+		t.Fatalf("timeout not respected: %v", elapsed)
+	}
+}
+
+func TestAnthropicNativeLinePump_DataResetsTimer(t *testing.T) {
+	pr, pw := io.Pipe()
+	scanner := bufio.NewScanner(pr)
+	pump := newAnthropicNativeLinePump(scanner, 1*time.Second)
+	defer pump.stop()
+
+	go func() {
+		_, _ = pw.Write([]byte("event: ping\n"))
+		// 保持流打开且不再发数据：第二次 next 必须超时。
+		time.Sleep(3 * time.Second)
+		_ = pw.Close()
+	}()
+	defer func() { _ = pr.Close() }()
+
+	line, err := pump.next()
+	if err != nil || line != "event: ping" {
+		t.Fatalf("expected first line, got %q err=%v", line, err)
+	}
+
+	start := time.Now()
+	_, err = pump.next()
+	if err == nil || !strings.Contains(err.Error(), "stream data interval timeout") {
+		t.Fatalf("expected interval timeout after data stops, got %v", err)
+	}
+	if elapsed := time.Since(start); elapsed > 5*time.Second {
+		t.Fatalf("timeout not respected: %v", elapsed)
+	}
+}
+
+func TestCCStreamingFromNativeAnthropic_HangTimesOut(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newNativeAnthropicHangTestService(1)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	resp, pr, pw := newHangingUpstreamResponse()
+	start := time.Now()
+	res, err := svc.handleCCStreamingFromNativeAnthropic(resp, c, "glm-4.7", "glm-4.7", "glm-4.7", nil, start, true)
+	_ = pw.Close()
+	_ = pr.Close()
+
+	if err == nil || !strings.Contains(err.Error(), "stream data interval timeout") {
+		t.Fatalf("expected stream timeout error, got %v", err)
+	}
+	if res == nil {
+		t.Fatalf("expected result carrying accumulated usage")
+	}
+	if elapsed := time.Since(start); elapsed > 5*time.Second {
+		t.Fatalf("handler did not respect interval bound: %v", elapsed)
+	}
+}
+
+func TestCCBufferedFromNativeAnthropic_HangTimesOut(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newNativeAnthropicHangTestService(1)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	resp, pr, pw := newHangingUpstreamResponse()
+	start := time.Now()
+	_, err := svc.handleCCBufferedFromNativeAnthropic(resp, c, "glm-4.7", "glm-4.7", "glm-4.7", nil, start)
+	_ = pw.Close()
+	_ = pr.Close()
+
+	if err == nil || !strings.Contains(err.Error(), "stream data interval timeout") {
+		t.Fatalf("expected stream timeout error, got %v", err)
+	}
+	if elapsed := time.Since(start); elapsed > 5*time.Second {
+		t.Fatalf("handler did not respect interval bound: %v", elapsed)
+	}
+	if !strings.Contains(rec.Body.String(), "Upstream stream data interval timeout") {
+		t.Fatalf("expected 502 error body, got %q", rec.Body.String())
+	}
+}
+
+func TestResponsesStreamingFromNativeAnthropic_HangTimesOut(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newNativeAnthropicHangTestService(1)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	resp, pr, pw := newHangingUpstreamResponse()
+	start := time.Now()
+	res, err := svc.handleResponsesStreamingFromNativeAnthropic(resp, c, "glm-4.7", "glm-4.7", "glm-4.7", nil, start, apicompat.ResponsesClientToolMapping{})
+	_ = pw.Close()
+	_ = pr.Close()
+
+	if err == nil || !strings.Contains(err.Error(), "stream data interval timeout") {
+		t.Fatalf("expected stream timeout error, got %v", err)
+	}
+	if res == nil {
+		t.Fatalf("expected result carrying accumulated usage")
+	}
+	if elapsed := time.Since(start); elapsed > 5*time.Second {
+		t.Fatalf("handler did not respect interval bound: %v", elapsed)
+	}
+}
+
+func TestCCStreamingFromNativeAnthropic_HappyPathStillConverts(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newNativeAnthropicHangTestService(5)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	resp, pr, pw := newHangingUpstreamResponse()
+	go func() {
+		_, _ = pw.Write([]byte(miniAnthropicSSEStream()))
+		_ = pw.Close()
+	}()
+	defer func() { _ = pr.Close() }()
+
+	res, err := svc.handleCCStreamingFromNativeAnthropic(resp, c, "glm-4.7", "glm-4.7", "glm-4.7", nil, time.Now(), true)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	if res == nil {
+		t.Fatalf("expected result")
+	}
+	body := rec.Body.String()
+	if !strings.Contains(body, "Hello") {
+		t.Fatalf("expected converted text chunk, got %q", body)
+	}
+	if !strings.Contains(body, "data: [DONE]") {
+		t.Fatalf("expected [DONE] terminator, got %q", body)
+	}
+}
+
+func TestCCBufferedFromNativeAnthropic_HappyPathStillConverts(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newNativeAnthropicHangTestService(5)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	resp, pr, pw := newHangingUpstreamResponse()
+	go func() {
+		_, _ = pw.Write([]byte(miniAnthropicSSEStream()))
+		_ = pw.Close()
+	}()
+	defer func() { _ = pr.Close() }()
+
+	res, err := svc.handleCCBufferedFromNativeAnthropic(resp, c, "glm-4.7", "glm-4.7", "glm-4.7", nil, time.Now())
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	if res == nil {
+		t.Fatalf("expected result")
+	}
+	body := rec.Body.String()
+	if !strings.Contains(body, "Hello") {
+		t.Fatalf("expected converted text in buffered response, got %q", body)
+	}
+	if res.Usage.InputTokens != 10 || res.Usage.OutputTokens != 5 {
+		t.Fatalf("expected usage 10/5, got %+v", res.Usage)
+	}
+}
diff --git a/backend/internal/service/openai_gateway_chat_completions_anthropic_native.go b/backend/internal/service/openai_gateway_chat_completions_anthropic_native.go
index 0ef89a7ed02e..b41bc644c62e 100644
--- a/backend/internal/service/openai_gateway_chat_completions_anthropic_native.go
+++ b/backend/internal/service/openai_gateway_chat_completions_anthropic_native.go
@@ -14,6 +14,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
 	"net/http"
 	"strings"
 	"time"
@@ -163,18 +164,53 @@ func (s *OpenAIGatewayService) handleCCBufferedFromNativeAnthropic(
 	var finalResp *apicompat.AnthropicResponse
 	var usage ClaudeUsage
 
-	for scanner.Scan() {
-		line := scanner.Text()
+	// 读间隔上限：上游挂住 SSE 时中止组装（缓冲路径尚未提交响应头，可回 502）。
+	streamInterval := s.anthropicNativeStreamInterval()
+	pump := newAnthropicNativeLinePump(scanner, streamInterval)
+	defer pump.stop()
+
+	logReadErr := func(err error) {
+		if !errors.Is(err, io.EOF) && !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai cc via native anthropic buffered: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+	onIdle := func() (*OpenAIForwardResult, error) {
+		_ = resp.Body.Close()
+		logger.L().Warn("openai cc via native anthropic buffered: data interval timeout",
+			zap.String("request_id", requestID),
+			zap.Duration("interval", streamInterval),
+		)
+		writeChatCompletionsError(c, http.StatusBadGateway, "server_error", "Upstream stream data interval timeout")
+		return nil, fmt.Errorf("stream data interval timeout")
+	}
+
+	for {
+		line, rerr := pump.next()
+		if rerr != nil {
+			if errors.Is(rerr, errAnthropicNativeStreamIdle) {
+				return onIdle()
+			}
+			logReadErr(rerr)
+			break
+		}
 		// SSE 规范允许 `event:xxx`（冒号后无空格）：Kimi 等 Anthropic 兼容上游
 		// 返回紧凑格式，严格匹配 "event: " 会丢弃全部事件（#4653 同根因）。
 		if _, ok := extractOpenAISSEEventLine(line); !ok {
 			continue
 		}
 
-		if !scanner.Scan() {
+		dataLine, rerr := pump.next()
+		if rerr != nil {
+			if errors.Is(rerr, errAnthropicNativeStreamIdle) {
+				return onIdle()
+			}
+			logReadErr(rerr)
 			break
 		}
-		payload, ok := extractOpenAISSEDataLine(scanner.Text())
+		payload, ok := extractOpenAISSEDataLine(dataLine)
 		if !ok {
 			continue
 		}
@@ -214,15 +250,6 @@ func (s *OpenAIGatewayService) handleCCBufferedFromNativeAnthropic(
 		}
 	}
 
-	if err := scanner.Err(); err != nil {
-		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
-			logger.L().Warn("openai cc via native anthropic buffered: read error",
-				zap.Error(err),
-				zap.String("request_id", requestID),
-			)
-		}
-	}
-
 	if finalResp == nil {
 		writeChatCompletionsError(c, http.StatusBadGateway, "server_error", "Upstream stream ended without a response")
 		return nil, fmt.Errorf("upstream stream ended without response")
@@ -322,6 +349,34 @@ func (s *OpenAIGatewayService) handleCCStreamingFromNativeAnthropic(
 		}
 	}
 
+	// 读间隔上限：上游挂住 SSE（不发数据也不断连）时结束排水。上游 ctx 为
+	// WithoutCancel 且 http.Client 无整体 Timeout，无此界限则客户端断开后
+	// scanner.Scan() 永久阻塞（见 anthropic native pump 文件注释）。
+	streamInterval := s.anthropicNativeStreamInterval()
+	pump := newAnthropicNativeLinePump(scanner, streamInterval)
+	defer pump.stop()
+
+	logReadErr := func(err error) {
+		if !errors.Is(err, io.EOF) && !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai cc via native anthropic stream: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+	// onIdle 关闭上游连接（解除阻塞的读、归还连接池位），并按已累计 usage
+	// 返回——与 messages 主路径 "stream usage incomplete after timeout" 同语义。
+	onIdle := func() (*OpenAIForwardResult, error) {
+		_ = resp.Body.Close()
+		if !clientDisconnected {
+			logger.L().Warn("openai cc via native anthropic stream: data interval timeout",
+				zap.String("request_id", requestID),
+				zap.Duration("interval", streamInterval),
+			)
+		}
+		return resultWithUsage(), fmt.Errorf("stream data interval timeout")
+	}
+
 	writeChunk := func(chunk apicompat.ChatCompletionsChunk) bool {
 		if clientDisconnected {
 			return false // 已断开：不再写客户端，只排水上游累计 usage
@@ -372,16 +427,29 @@ func (s *OpenAIGatewayService) handleCCStreamingFromNativeAnthropic(
 		return false
 	}
 
-	for scanner.Scan() {
-		line := scanner.Text()
+	for {
+		line, rerr := pump.next()
+		if rerr != nil {
+			if errors.Is(rerr, errAnthropicNativeStreamIdle) {
+				return onIdle()
+			}
+			logReadErr(rerr)
+			break
+		}
 		if _, ok := extractOpenAISSEEventLine(line); !ok {
 			continue
 		}
 
-		if !scanner.Scan() {
+		dataLine, rerr := pump.next()
+		if rerr != nil {
+			if errors.Is(rerr, errAnthropicNativeStreamIdle) {
+				return onIdle()
+			}
+			// EOF / 读错误：事件行后流终止，进入 finalize。
+			logReadErr(rerr)
 			break
 		}
-		payload, ok := extractOpenAISSEDataLine(scanner.Text())
+		payload, ok := extractOpenAISSEDataLine(dataLine)
 		if !ok {
 			continue
 		}
@@ -396,15 +464,6 @@ func (s *OpenAIGatewayService) handleCCStreamingFromNativeAnthropic(
 		}
 	}
 
-	if err := scanner.Err(); err != nil {
-		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
-			logger.L().Warn("openai cc via native anthropic stream: read error",
-				zap.Error(err),
-				zap.String("request_id", requestID),
-			)
-		}
-	}
-
 	// Finalize both state machines（客户端已断开时仍执行，保证 usage 汇总完整）。
 	finalResEvents := apicompat.FinalizeAnthropicResponsesStream(anthState)
 	for _, resEvt := range finalResEvents {
diff --git a/backend/internal/service/openai_gateway_responses_anthropic_native.go b/backend/internal/service/openai_gateway_responses_anthropic_native.go
index 7cb4a13d2d49..b72067fc4d77 100644
--- a/backend/internal/service/openai_gateway_responses_anthropic_native.go
+++ b/backend/internal/service/openai_gateway_responses_anthropic_native.go
@@ -14,6 +14,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
 	"net/http"
 	"strings"
 	"time"
@@ -166,17 +167,52 @@ func (s *OpenAIGatewayService) handleResponsesBufferedFromNativeAnthropic(
 	var finalResp *apicompat.AnthropicResponse
 	var usage ClaudeUsage
 
-	for scanner.Scan() {
-		line := scanner.Text()
+	// 读间隔上限：上游挂住 SSE 时中止组装（缓冲路径尚未提交响应头，可回 502）。
+	streamInterval := s.anthropicNativeStreamInterval()
+	pump := newAnthropicNativeLinePump(scanner, streamInterval)
+	defer pump.stop()
+
+	logReadErr := func(err error) {
+		if !errors.Is(err, io.EOF) && !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai responses via native anthropic buffered: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+	onIdle := func() (*OpenAIForwardResult, error) {
+		_ = resp.Body.Close()
+		logger.L().Warn("openai responses via native anthropic buffered: data interval timeout",
+			zap.String("request_id", requestID),
+			zap.Duration("interval", streamInterval),
+		)
+		writeResponsesError(c, http.StatusBadGateway, "server_error", "Upstream stream data interval timeout")
+		return nil, fmt.Errorf("stream data interval timeout")
+	}
+
+	for {
+		line, rerr := pump.next()
+		if rerr != nil {
+			if errors.Is(rerr, errAnthropicNativeStreamIdle) {
+				return onIdle()
+			}
+			logReadErr(rerr)
+			break
+		}
 		// SSE 规范允许 `event:xxx`（冒号后无空格）：Kimi 等上游返回紧凑格式。
 		if _, ok := extractOpenAISSEEventLine(line); !ok {
 			continue
 		}
 
-		if !scanner.Scan() {
+		dataLine, rerr := pump.next()
+		if rerr != nil {
+			if errors.Is(rerr, errAnthropicNativeStreamIdle) {
+				return onIdle()
+			}
+			logReadErr(rerr)
 			break
 		}
-		payload, ok := extractOpenAISSEDataLine(scanner.Text())
+		payload, ok := extractOpenAISSEDataLine(dataLine)
 		if !ok {
 			continue
 		}
@@ -216,15 +252,6 @@ func (s *OpenAIGatewayService) handleResponsesBufferedFromNativeAnthropic(
 		}
 	}
 
-	if err := scanner.Err(); err != nil {
-		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
-			logger.L().Warn("openai responses via native anthropic buffered: read error",
-				zap.Error(err),
-				zap.String("request_id", requestID),
-			)
-		}
-	}
-
 	if finalResp == nil {
 		writeResponsesError(c, http.StatusBadGateway, "server_error", "Upstream stream ended without a response")
 		return nil, fmt.Errorf("upstream stream ended without response")
@@ -326,6 +353,30 @@ func (s *OpenAIGatewayService) handleResponsesStreamingFromNativeAnthropic(
 		}
 	}
 
+	// 读间隔上限：上游挂住 SSE（不发数据也不断连）时结束转换循环。上游 ctx 为
+	// WithoutCancel 且 http.Client 无整体 Timeout，无此界限则 scanner.Scan()
+	// 永久阻塞（见 anthropic native pump 文件注释）。
+	streamInterval := s.anthropicNativeStreamInterval()
+	pump := newAnthropicNativeLinePump(scanner, streamInterval)
+	defer pump.stop()
+
+	logReadErr := func(err error) {
+		if !errors.Is(err, io.EOF) && !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai responses via native anthropic stream: read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+	onIdle := func() (*OpenAIForwardResult, error) {
+		_ = resp.Body.Close()
+		logger.L().Warn("openai responses via native anthropic stream: data interval timeout",
+			zap.String("request_id", requestID),
+			zap.Duration("interval", streamInterval),
+		)
+		return resultWithUsage(), fmt.Errorf("stream data interval timeout")
+	}
+
 	processAnthropicEvent := func(event *apicompat.AnthropicStreamEvent) bool {
 		if firstChunk {
 			firstChunk = false
@@ -365,16 +416,28 @@ func (s *OpenAIGatewayService) handleResponsesStreamingFromNativeAnthropic(
 		return false
 	}
 
-	for scanner.Scan() {
-		line := scanner.Text()
+	for {
+		line, rerr := pump.next()
+		if rerr != nil {
+			if errors.Is(rerr, errAnthropicNativeStreamIdle) {
+				return onIdle()
+			}
+			logReadErr(rerr)
+			break
+		}
 		if _, ok := extractOpenAISSEEventLine(line); !ok {
 			continue
 		}
 
-		if !scanner.Scan() {
+		dataLine, rerr := pump.next()
+		if rerr != nil {
+			if errors.Is(rerr, errAnthropicNativeStreamIdle) {
+				return onIdle()
+			}
+			logReadErr(rerr)
 			break
 		}
-		payload, ok := extractOpenAISSEDataLine(scanner.Text())
+		payload, ok := extractOpenAISSEDataLine(dataLine)
 		if !ok {
 			continue
 		}
@@ -389,15 +452,6 @@ func (s *OpenAIGatewayService) handleResponsesStreamingFromNativeAnthropic(
 		}
 	}
 
-	if err := scanner.Err(); err != nil {
-		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
-			logger.L().Warn("openai responses via native anthropic stream: read error",
-				zap.Error(err),
-				zap.String("request_id", requestID),
-			)
-		}
-	}
-
 	// Finalize state machine（客户端已断开时仍执行，保证 usage 汇总完整）。
 	if finalEvents := apicompat.FinalizeAnthropicResponsesStream(state); len(finalEvents) > 0 {
 		for _, evt := range finalEvents {
diff --git a/backend/migrations/224_user_platform_quotas_add_cn_providers.sql b/backend/migrations/224_user_platform_quotas_add_cn_providers.sql
new file mode 100644
index 000000000000..011a7762afe7
--- /dev/null
+++ b/backend/migrations/224_user_platform_quotas_add_cn_providers.sql
@@ -0,0 +1,17 @@
+-- 把 kimi/zhipu/deepseek 平台加入 user_platform_quotas.platform 的 CHECK 约束。
+--
+-- 背景：国产供应商进入 AllowedQuotaPlatforms（internal/service/domain_constants.go），
+-- 注册时 GetDefaultPlatformQuotas 会为全部 8 平台预填充默认配额行，但 157 号迁移的
+-- CHECK 仍只允许 5 平台。BulkInsertInitial 是单条多行 INSERT，任一违约行会中止整条
+-- 语句 → 注册路径 fail-open 吞错 → 新用户拿到零条配额记录（含原有 5 平台，缺失配额
+-- 行 = 无限额）。与 157 头注释记载的 grok 同型事故一致。
+--
+-- 修复：把约束与代码平台列表（PlatformKimi/PlatformZhipu/PlatformDeepseek）对齐。
+-- DROP ... IF EXISTS 保证可重入；新约束是旧约束的超集，存量行（仅 5 平台）瞬时校验通过。
+ALTER TABLE user_platform_quotas
+    DROP CONSTRAINT IF EXISTS user_platform_quotas_platform_check;
+
+ALTER TABLE user_platform_quotas
+    ADD CONSTRAINT user_platform_quotas_platform_check
+    CHECK (platform IN ('anthropic', 'openai', 'gemini', 'antigravity', 'grok',
+                        'kimi', 'zhipu', 'deepseek'));
diff --git a/backend/migrations/user_platform_quota_cn_providers_migration_test.go b/backend/migrations/user_platform_quota_cn_providers_migration_test.go
new file mode 100644
index 000000000000..cce672aaaf63
--- /dev/null
+++ b/backend/migrations/user_platform_quota_cn_providers_migration_test.go
@@ -0,0 +1,22 @@
+package migrations
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+// TestUserPlatformQuotasCNProvidersMigration 校验 224 号迁移把 kimi/zhipu/deepseek
+// 加入 user_platform_quotas.platform 的 CHECK 约束（对照 157 号 grok 迁移）。
+// 约束未放宽时，注册预填充 8 平台默认配额会整条 INSERT 中止 → 新用户零配额行
+//（缺失配额行 = 无限额），管理端设置国产平台配额直接 500。
+func TestUserPlatformQuotasCNProvidersMigration(t *testing.T) {
+	content, err := FS.ReadFile("224_user_platform_quotas_add_cn_providers.sql")
+	require.NoError(t, err)
+
+	sql := strings.Join(strings.Fields(string(content)), " ")
+	require.Contains(t, sql, "DROP CONSTRAINT IF EXISTS user_platform_quotas_platform_check")
+	require.Contains(t, sql,
+		"CHECK (platform IN ('anthropic', 'openai', 'gemini', 'antigravity', 'grok', 'kimi', 'zhipu', 'deepseek'))")
+}
diff --git a/docker-compose.yml b/docker-compose.yml
deleted file mode 100644
index 67edd44e2205..000000000000
--- a/docker-compose.yml
+++ /dev/null
@@ -1,302 +0,0 @@
-# =============================================================================
-# Sub2API Docker Compose Configuration
-# =============================================================================
-# Quick Start:
-#   1. Copy .env.example to .env and configure
-#   2. docker-compose up -d
-#   3. Check logs: docker-compose logs -f sub2api
-#   4. Access: http://localhost:8080
-#
-# All configuration is done via environment variables.
-# No Setup Wizard needed - the system auto-initializes on first run.
-# =============================================================================
-
-services:
-  # ===========================================================================
-  # Sub2API Application
-  # ===========================================================================
-  sub2api:
-    image: randark/sub2api:latest
-    container_name: sub2api
-    restart: unless-stopped
-    security_opt:
-      - no-new-privileges:true
-    ulimits:
-      nofile:
-        soft: 100000
-        hard: 100000
-    ports:
-      - "${BIND_HOST:-0.0.0.0}:${SERVER_PORT:-8080}:8080"
-    volumes:
-      # Data persistence (config.yaml will be auto-generated here)
-      - sub2api_data:/app/data
-      # Optional: Mount custom config.yaml (uncomment and create the file first)
-      # Copy config.example.yaml to config.yaml, modify it, then uncomment:
-      # - ./config.yaml:/app/data/config.yaml
-      # Optional: Mount a custom Codex instructions template file, then point
-      # gateway.forced_codex_instructions_template_file at /app/data/codex-instructions.md.tmpl
-      # in config.yaml.
-      # - ./codex-instructions.md.tmpl:/app/data/codex-instructions.md.tmpl:ro
-    environment:
-      # =======================================================================
-      # Auto Setup (REQUIRED for Docker deployment)
-      # =======================================================================
-      - AUTO_SETUP=true
-
-      # =======================================================================
-      # Server Configuration
-      # =======================================================================
-      - SERVER_HOST=0.0.0.0
-      - SERVER_PORT=8080
-      - SERVER_MODE=${SERVER_MODE:-release}
-      - ENABLE_SERVER_TIMING=${ENABLE_SERVER_TIMING:-false}
-      - RUN_MODE=${RUN_MODE:-standard}
-      - UPDATE_GITHUB_TOKEN=${UPDATE_GITHUB_TOKEN:-}
-      - ALIPAY_MOBILE_PRECREATE_DEEP_LINK=${ALIPAY_MOBILE_PRECREATE_DEEP_LINK:-}
-
-      # =======================================================================
-      # Database Configuration (PostgreSQL)
-      # =======================================================================
-      - DATABASE_HOST=postgres
-      - DATABASE_PORT=5432
-      - DATABASE_USER=${POSTGRES_USER:-sub2api}
-      - DATABASE_PASSWORD=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required}
-      - DATABASE_DBNAME=${POSTGRES_DB:-sub2api}
-      - DATABASE_SSLMODE=disable
-      - DATABASE_MAX_OPEN_CONNS=${DATABASE_MAX_OPEN_CONNS:-50}
-      - DATABASE_MAX_IDLE_CONNS=${DATABASE_MAX_IDLE_CONNS:-10}
-      - DATABASE_CONN_MAX_LIFETIME_MINUTES=${DATABASE_CONN_MAX_LIFETIME_MINUTES:-30}
-      - DATABASE_CONN_MAX_IDLE_TIME_MINUTES=${DATABASE_CONN_MAX_IDLE_TIME_MINUTES:-5}
-
-      # =======================================================================
-      # Redis Configuration
-      # =======================================================================
-      - REDIS_HOST=redis
-      - REDIS_PORT=6379
-      - REDIS_USERNAME=${REDIS_USERNAME:-}
-      - REDIS_PASSWORD=${REDIS_PASSWORD:-}
-      - REDIS_DB=${REDIS_DB:-0}
-      - REDIS_POOL_SIZE=${REDIS_POOL_SIZE:-1024}
-      - REDIS_MIN_IDLE_CONNS=${REDIS_MIN_IDLE_CONNS:-10}
-      - REDIS_ENABLE_TLS=${REDIS_ENABLE_TLS:-false}
-
-      # =======================================================================
-      # Admin Account (auto-created on first run)
-      # =======================================================================
-      - ADMIN_EMAIL=${ADMIN_EMAIL:-admin@sub2api.local}
-      - ADMIN_PASSWORD=${ADMIN_PASSWORD:-}
-
-      # =======================================================================
-      # JWT Configuration
-      # =======================================================================
-      # IMPORTANT: Set a fixed JWT_SECRET to prevent login sessions from being
-      # invalidated after container restarts. If left empty, a random secret
-      # will be generated on each startup.
-      # Generate a secure secret: openssl rand -hex 32
-      - JWT_SECRET=${JWT_SECRET:-}
-      - JWT_EXPIRE_HOUR=${JWT_EXPIRE_HOUR:-24}
-
-      # =======================================================================
-      # Setup Configuration
-      # =======================================================================
-      - SETUP_MIGRATION_TIMEOUT_SECONDS=${SETUP_MIGRATION_TIMEOUT_SECONDS:-0}
-
-      # =======================================================================
-      # TOTP (2FA) Configuration
-      # =======================================================================
-      # IMPORTANT: Set a fixed encryption key for TOTP secrets. If left empty,
-      # a random key will be generated on each startup, causing all existing
-      # TOTP configurations to become invalid (users won't be able to login
-      # with 2FA).
-      # Generate a secure key: openssl rand -hex 32
-      - TOTP_ENCRYPTION_KEY=${TOTP_ENCRYPTION_KEY:-}
-
-      # =======================================================================
-      # Timezone Configuration
-      # This affects ALL time operations in the application:
-      # - Database timestamps
-      # - Usage statistics "today" boundary
-      # - Subscription expiry times
-      # - Log timestamps
-      # Common values: Asia/Shanghai, America/New_York, Europe/London, UTC
-      # =======================================================================
-      - TZ=${TZ:-Asia/Shanghai}
-
-      # =======================================================================
-      # Gemini OAuth Configuration (for Gemini accounts)
-      # =======================================================================
-      - GEMINI_OAUTH_CLIENT_ID=${GEMINI_OAUTH_CLIENT_ID:-}
-      - GEMINI_OAUTH_CLIENT_SECRET=${GEMINI_OAUTH_CLIENT_SECRET:-}
-      - GEMINI_OAUTH_SCOPES=${GEMINI_OAUTH_SCOPES:-}
-      - GEMINI_QUOTA_POLICY=${GEMINI_QUOTA_POLICY:-}
-
-      # Built-in OAuth client secrets (optional)
-      # SECURITY: This repo does not embed third-party client_secret.
-      - GEMINI_CLI_OAUTH_CLIENT_SECRET=${GEMINI_CLI_OAUTH_CLIENT_SECRET:-}
-      - ANTIGRAVITY_OAUTH_CLIENT_SECRET=${ANTIGRAVITY_OAUTH_CLIENT_SECRET:-}
-      - ANTIGRAVITY_USER_AGENT_VERSION=${ANTIGRAVITY_USER_AGENT_VERSION:-}
-
-      # =======================================================================
-      # Security Configuration (URL Allowlist)
-      # =======================================================================
-      # Enable URL allowlist validation (false to skip allowlist checks)
-      - SECURITY_URL_ALLOWLIST_ENABLED=${SECURITY_URL_ALLOWLIST_ENABLED:-false}
-      # Allow insecure HTTP URLs when allowlist is disabled (default: true; set to false to require https)
-      - SECURITY_URL_ALLOWLIST_ALLOW_INSECURE_HTTP=${SECURITY_URL_ALLOWLIST_ALLOW_INSECURE_HTTP:-true}
-      # Allow private IP addresses for upstream/pricing/CRS (default: true; set to false to block private hosts)
-      - SECURITY_URL_ALLOWLIST_ALLOW_PRIVATE_HOSTS=${SECURITY_URL_ALLOWLIST_ALLOW_PRIVATE_HOSTS:-true}
-      # Upstream hosts whitelist (comma-separated, only used when enabled=true)
-      - SECURITY_URL_ALLOWLIST_UPSTREAM_HOSTS=${SECURITY_URL_ALLOWLIST_UPSTREAM_HOSTS:-}
-
-      # =======================================================================
-      # Update Configuration (在线更新配置)
-      # =======================================================================
-      # Proxy for accessing GitHub (online updates + pricing data)
-      # Examples: http://host:port, socks5://host:port
-      - UPDATE_PROXY_URL=${UPDATE_PROXY_URL:-}
-
-      # =======================================================================
-      # Image Generation Stream & Concurrency
-      # =======================================================================
-      # OpenAI HTTP upstream protocol/timeout
-      - GATEWAY_OPENAI_RESPONSE_HEADER_TIMEOUT=${GATEWAY_OPENAI_RESPONSE_HEADER_TIMEOUT:-0}
-      - GATEWAY_OPENAI_HTTP2_ENABLED=${GATEWAY_OPENAI_HTTP2_ENABLED:-true}
-      - GATEWAY_OPENAI_HTTP2_ALLOW_PROXY_FALLBACK_TO_HTTP1=${GATEWAY_OPENAI_HTTP2_ALLOW_PROXY_FALLBACK_TO_HTTP1:-true}
-      - GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD=${GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD:-2}
-      - GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS:-60}
-      - GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS:-600}
-      - GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD:-2}
-      - GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS:-60}
-      - GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS:-600}
-      - GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT=${GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT:-900}
-      - GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL=${GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL:-10}
-      - GATEWAY_IMAGE_CONCURRENCY_ENABLED=${GATEWAY_IMAGE_CONCURRENCY_ENABLED:-false}
-      - GATEWAY_IMAGE_CONCURRENCY_MAX_CONCURRENT_REQUESTS=${GATEWAY_IMAGE_CONCURRENCY_MAX_CONCURRENT_REQUESTS:-0}
-      - GATEWAY_IMAGE_CONCURRENCY_OVERFLOW_MODE=${GATEWAY_IMAGE_CONCURRENCY_OVERFLOW_MODE:-reject}
-      - GATEWAY_IMAGE_CONCURRENCY_WAIT_TIMEOUT_SECONDS=${GATEWAY_IMAGE_CONCURRENCY_WAIT_TIMEOUT_SECONDS:-30}
-      - GATEWAY_IMAGE_CONCURRENCY_MAX_WAITING_REQUESTS=${GATEWAY_IMAGE_CONCURRENCY_MAX_WAITING_REQUESTS:-100}
-    depends_on:
-      postgres:
-        condition: service_healthy
-      redis:
-        condition: service_healthy
-    networks:
-      - sub2api-network
-    healthcheck:
-      test:
-        [
-          "CMD",
-          "wget",
-          "-q",
-          "-T",
-          "5",
-          "-O",
-          "/dev/null",
-          "http://localhost:8080/health",
-        ]
-      interval: 30s
-      timeout: 10s
-      retries: 3
-      start_period: 30s
-
-  # ===========================================================================
-  # PostgreSQL Database
-  # ===========================================================================
-  postgres:
-    image: postgres:18-alpine
-    container_name: sub2api-postgres
-    restart: unless-stopped
-    ulimits:
-      nofile:
-        soft: 100000
-        hard: 100000
-    # Pass the tuning values documented in .env.example to postgres. The
-    # fallbacks are the postgres:18 stock defaults, so nothing changes when
-    # the variables are not set.
-    command: >
-      postgres
-      -c max_connections=${POSTGRES_MAX_CONNECTIONS:-100}
-      -c shared_buffers=${POSTGRES_SHARED_BUFFERS:-128MB}
-      -c effective_cache_size=${POSTGRES_EFFECTIVE_CACHE_SIZE:-4GB}
-      -c maintenance_work_mem=${POSTGRES_MAINTENANCE_WORK_MEM:-64MB}
-    volumes:
-      - postgres_data:/var/lib/postgresql/data
-    environment:
-      # postgres:18-alpine 默认 PGDATA=/var/lib/postgresql/18/docker（位于镜像声明的匿名卷 /var/lib/postgresql 内）。
-      # 若不显式设置 PGDATA，则即使挂载了 postgres_data 到 /var/lib/postgresql/data，数据也不会落盘到该命名卷，
-      # docker compose down/up 后会触发 initdb 重新初始化，导致用户/密码等数据丢失。
-      - PGDATA=/var/lib/postgresql/data
-      - POSTGRES_USER=${POSTGRES_USER:-sub2api}
-      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required}
-      - POSTGRES_DB=${POSTGRES_DB:-sub2api}
-      - TZ=${TZ:-Asia/Shanghai}
-    networks:
-      - sub2api-network
-    healthcheck:
-      test:
-        [
-          "CMD-SHELL",
-          "pg_isready -U ${POSTGRES_USER:-sub2api} -d ${POSTGRES_DB:-sub2api}",
-        ]
-      interval: 10s
-      timeout: 5s
-      retries: 5
-      start_period: 10s
-    # 注意：不暴露端口到宿主机，应用通过内部网络连接
-    # 如需调试，可临时添加：ports: ["127.0.0.1:5433:5432"]
-
-  # ===========================================================================
-  # Redis Cache
-  # ===========================================================================
-  redis:
-    image: redis:8-alpine
-    container_name: sub2api-redis
-    restart: unless-stopped
-    ulimits:
-      nofile:
-        soft: 100000
-        hard: 100000
-    volumes:
-      - redis_data:/data
-    # The command is one quoted script for the inner `sh -c`. Compose keeps
-    # the newlines inside the quoted string, so every line needs a trailing
-    # `\` — without it, `redis-server` on the first line runs with no flags
-    # at all, and the --save/--appendonly/--appendfsync lines are never read.
-    command: >
-      sh -c '
-        redis-server \
-        --save 60 1 \
-        --appendonly yes \
-        --appendfsync everysec \
-        ${REDIS_PASSWORD:+--requirepass "$REDIS_PASSWORD"}'
-    environment:
-      - TZ=${TZ:-Asia/Shanghai}
-      # REDISCLI_AUTH is used by redis-cli for authentication (safer than -a flag)
-      - REDISCLI_AUTH=${REDIS_PASSWORD:-}
-    networks:
-      - sub2api-network
-    healthcheck:
-      test: ["CMD", "redis-cli", "ping"]
-      interval: 10s
-      timeout: 5s
-      retries: 5
-      start_period: 5s
-    # 注意：不暴露端口到宿主机，应用通过内部网络连接
-    # 如需调试，可临时添加：ports: ["127.0.0.1:6379:6379"]
-# =============================================================================
-# Volumes
-# =============================================================================
-volumes:
-  sub2api_data:
-    driver: local
-  postgres_data:
-    driver: local
-  redis_data:
-    driver: local
-
-# =============================================================================
-# Networks
-# =============================================================================
-networks:
-  sub2api-network:
-    driver: bridge

From e7285453829930e2889432dcd18d7a0c2ba18481 Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Mon, 17 Aug 2026 04:26:22 +0000
Subject: [PATCH 30/68] =?UTF-8?q?style:=20gofmt=20=E8=BF=81=E7=A7=BB?=
 =?UTF-8?q?=E6=B5=8B=E8=AF=95=E6=B3=A8=E9=87=8A=EF=BC=88CI=20golangci-lint?=
 =?UTF-8?q?=20gofmt=EF=BC=89?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../user_platform_quota_cn_providers_migration_test.go          | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/backend/migrations/user_platform_quota_cn_providers_migration_test.go b/backend/migrations/user_platform_quota_cn_providers_migration_test.go
index cce672aaaf63..73bfd057677a 100644
--- a/backend/migrations/user_platform_quota_cn_providers_migration_test.go
+++ b/backend/migrations/user_platform_quota_cn_providers_migration_test.go
@@ -10,7 +10,7 @@ import (
 // TestUserPlatformQuotasCNProvidersMigration 校验 224 号迁移把 kimi/zhipu/deepseek
 // 加入 user_platform_quotas.platform 的 CHECK 约束（对照 157 号 grok 迁移）。
 // 约束未放宽时，注册预填充 8 平台默认配额会整条 INSERT 中止 → 新用户零配额行
-//（缺失配额行 = 无限额），管理端设置国产平台配额直接 500。
+// （缺失配额行 = 无限额），管理端设置国产平台配额直接 500。
 func TestUserPlatformQuotasCNProvidersMigration(t *testing.T) {
 	content, err := FS.ReadFile("224_user_platform_quotas_add_cn_providers.sql")
 	require.NoError(t, err)

From 9aac3b73f7f6d03ca40a151647f48a49757b41d8 Mon Sep 17 00:00:00 2001
From: xiaobo <467359395@qq.com>
Date: Sun, 16 Aug 2026 21:33:53 -0700
Subject: [PATCH 31/68] add Ollama usage query action

---
 .../components/account/AccountUsageCell.vue   |  5 ++
 .../account/OllamaCloudUsageCell.vue          | 54 +++++++++++++++++--
 .../__tests__/AccountUsageCell.spec.ts        | 11 +++-
 .../__tests__/OllamaCloudUsageCell.spec.ts    | 53 +++++++++++++++---
 4 files changed, 111 insertions(+), 12 deletions(-)

diff --git a/frontend/src/components/account/AccountUsageCell.vue b/frontend/src/components/account/AccountUsageCell.vue
index 485b4b870102..5ad110087c5b 100644
--- a/frontend/src/components/account/AccountUsageCell.vue
+++ b/frontend/src/components/account/AccountUsageCell.vue
@@ -547,6 +547,7 @@
       <OllamaCloudUsageCell
         v-if="account.ollama_cloud_usage?.eligible"
         :account="account"
+        @updated="handleOllamaCloudUsageUpdated"
       />
       <!-- Today stats row (requests, tokens, cost, user_cost) -->
       <div
@@ -1477,6 +1478,10 @@ const handleQuotaResetAccountUpdated = (account: Account) => {
   emit('account-updated', account)
 }
 
+const handleOllamaCloudUsageUpdated = (state: NonNullable<Account['ollama_cloud_usage']>) => {
+  emit('account-updated', { ...props.account, ollama_cloud_usage: state })
+}
+
 // ===== Key account today stats formatters =====
 
 const formatKeyRequests = computed(() => {
diff --git a/frontend/src/components/account/OllamaCloudUsageCell.vue b/frontend/src/components/account/OllamaCloudUsageCell.vue
index 201e271fe530..17c09050c0a9 100644
--- a/frontend/src/components/account/OllamaCloudUsageCell.vue
+++ b/frontend/src/components/account/OllamaCloudUsageCell.vue
@@ -20,16 +20,64 @@
       color="emerald"
       data-testid="ollama-cloud-seven-day"
     />
+    <div v-if="state.configured" class="flex items-center pt-0.5">
+      <button
+        type="button"
+        class="inline-flex items-center gap-0.5 rounded px-1.5 py-0.5 text-[10px] font-medium text-blue-600 transition-colors hover:bg-blue-50 disabled:cursor-not-allowed disabled:opacity-50 dark:text-blue-400 dark:hover:bg-blue-900/30"
+        :disabled="refreshing"
+        data-testid="ollama-cloud-usage-query"
+        @click="refreshUsage"
+      >
+        <svg
+          class="h-2.5 w-2.5"
+          :class="{ 'animate-spin': refreshing }"
+          fill="none"
+          stroke="currentColor"
+          viewBox="0 0 24 24"
+        >
+          <path
+            stroke-linecap="round"
+            stroke-linejoin="round"
+            stroke-width="2"
+            d="M4 4v5h.582m15.356 2A8.001 8.001 0 004.582 9m0 0H9m11 11v-5h-.581m0 0a8.003 8.003 0 01-15.357-2m15.357 2H15"
+          />
+        </svg>
+        {{ t('admin.accounts.usageWindow.activeQuery') }}
+      </button>
+    </div>
   </div>
   <span v-else class="text-sm text-gray-400 dark:text-dark-500">-</span>
 </template>
 
 <script setup lang="ts">
-import { computed } from 'vue'
-import type { Account } from '@/types'
+import { computed, ref, watch } from 'vue'
+import { useI18n } from 'vue-i18n'
+import { adminAPI } from '@/api/admin'
+import type { Account, OllamaCloudUsageState } from '@/types'
 import UsageProgressBar from './UsageProgressBar.vue'
 
 const props = defineProps<{ account: Account }>()
-const state = computed(() => props.account.ollama_cloud_usage)
+const emit = defineEmits<{ updated: [state: OllamaCloudUsageState] }>()
+const { t } = useI18n()
+const state = ref(props.account.ollama_cloud_usage)
+const refreshing = ref(false)
 const snapshot = computed(() => state.value?.snapshot)
+
+watch(() => props.account.ollama_cloud_usage, (next) => {
+  state.value = next
+})
+
+const refreshUsage = async () => {
+  if (refreshing.value) return
+  refreshing.value = true
+  try {
+    const next = await adminAPI.accounts.refreshOllamaCloudUsage(props.account.id)
+    state.value = next
+    emit('updated', next)
+  } catch (error) {
+    console.error('Failed to refresh Ollama Cloud usage:', error)
+  } finally {
+    refreshing.value = false
+  }
+}
 </script>
diff --git a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
index ff0e94d6647e..dc071b6de57a 100644
--- a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
+++ b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
@@ -72,7 +72,7 @@ describe('AccountUsageCell', () => {
     })
   })
 
-  it('renders eligible Ollama Cloud state inside the unified usage cell', () => {
+  it('renders eligible Ollama Cloud state and forwards query updates', async () => {
     const wrapper = mount(AccountUsageCell, {
       props: {
         account: makeAccount({
@@ -101,7 +101,8 @@ describe('AccountUsageCell', () => {
         stubs: {
           OllamaCloudUsageCell: {
             props: ['account'],
-            template: '<div data-test="embedded-ollama">{{ account.ollama_cloud_usage.snapshot.data.five_hour.used_percent }}</div>'
+            emits: ['updated'],
+            template: '<button data-test="embedded-ollama" @click="$emit(\'updated\', { ...account.ollama_cloud_usage, auto_refresh_enabled: false })">{{ account.ollama_cloud_usage.snapshot.data.five_hour.used_percent }}</button>'
           },
           UsageProgressBar: true,
           AccountQuotaInfo: true
@@ -111,6 +112,12 @@ describe('AccountUsageCell', () => {
 
     expect(wrapper.get('[data-test="embedded-ollama"]').text()).toBe('12')
     expect(getUsage).not.toHaveBeenCalled()
+
+    await wrapper.get('[data-test="embedded-ollama"]').trigger('click')
+
+    const updatedAccount = wrapper.emitted<Account[]>('account-updated')?.[0]?.[0]
+    expect(updatedAccount?.id).toBe(9001)
+    expect(updatedAccount?.ollama_cloud_usage?.auto_refresh_enabled).toBe(false)
   })
 
   it('Antigravity 图片用量会聚合新旧 image 模型', async () => {
diff --git a/frontend/src/components/account/__tests__/OllamaCloudUsageCell.spec.ts b/frontend/src/components/account/__tests__/OllamaCloudUsageCell.spec.ts
index 5f6dfee054ad..e38f3d3729ff 100644
--- a/frontend/src/components/account/__tests__/OllamaCloudUsageCell.spec.ts
+++ b/frontend/src/components/account/__tests__/OllamaCloudUsageCell.spec.ts
@@ -1,9 +1,21 @@
-import { mount } from '@vue/test-utils'
-import { describe, expect, it, vi } from 'vitest'
+import { flushPromises, mount } from '@vue/test-utils'
+import { beforeEach, describe, expect, it, vi } from 'vitest'
 import OllamaCloudUsageCell from '../OllamaCloudUsageCell.vue'
 import UsageProgressBar from '../UsageProgressBar.vue'
 import type { Account, OllamaCloudUsageState } from '@/types'
 
+const { refreshOllamaCloudUsage } = vi.hoisted(() => ({
+  refreshOllamaCloudUsage: vi.fn()
+}))
+
+vi.mock('@/api/admin', () => ({
+  adminAPI: {
+    accounts: {
+      refreshOllamaCloudUsage
+    }
+  }
+}))
+
 vi.mock('vue-i18n', async () => {
   const actual = await vi.importActual<typeof import('vue-i18n')>('vue-i18n')
   return {
@@ -64,7 +76,11 @@ const account = (state = usageState()): Account => ({
 })
 
 describe('OllamaCloudUsageCell', () => {
-  it('renders only native 5h and 7d windows in a shrinkable mobile-safe cell', () => {
+  beforeEach(() => {
+    refreshOllamaCloudUsage.mockReset()
+  })
+
+  it('renders native 5h and 7d windows with the configured-account query action', () => {
     const wrapper = mount(OllamaCloudUsageCell, { props: { account: account() } })
     const cell = wrapper.get('[data-testid="ollama-cloud-usage-cell"]')
     expect(cell.classes()).toEqual(expect.arrayContaining(['min-w-0', 'max-w-full']))
@@ -84,14 +100,15 @@ describe('OllamaCloudUsageCell', () => {
     })
 
     expect(wrapper.find('[data-testid="ollama-cloud-usage-details"]').exists()).toBe(false)
-    expect(wrapper.find('[data-testid="ollama-cloud-usage-refresh"]').exists()).toBe(false)
-    expect(wrapper.findAll('button')).toHaveLength(0)
+    const query = wrapper.get('[data-testid="ollama-cloud-usage-query"]')
+    expect(query.classes()).toEqual(expect.arrayContaining(['text-blue-600', 'hover:bg-blue-50']))
+    expect(query.text()).toContain('admin.accounts.usageWindow.activeQuery')
     expect(wrapper.text()).not.toContain('max')
     expect(wrapper.text()).not.toContain('$0')
     expect(wrapper.text()).not.toContain('gpt-oss:120b-cloud')
   })
 
-  it('reacts to an account snapshot update without a list-cell refresh action', async () => {
+  it('reacts to an account snapshot update', async () => {
     const wrapper = mount(OllamaCloudUsageCell, { props: { account: account() } })
     const next = usageState()
     next.snapshot!.data!.five_hour!.used_percent = 43
@@ -99,6 +116,28 @@ describe('OllamaCloudUsageCell', () => {
     await wrapper.setProps({ account: account(next) })
 
     expect(wrapper.findAllComponents(UsageProgressBar)[0].props('utilization')).toBe(43)
-    expect(wrapper.findAll('button')).toHaveLength(0)
+  })
+
+  it('queries through the edit-page refresh endpoint and emits the updated state', async () => {
+    const next = usageState()
+    next.snapshot!.data!.five_hour!.used_percent = 43
+    refreshOllamaCloudUsage.mockResolvedValueOnce(next)
+    const wrapper = mount(OllamaCloudUsageCell, { props: { account: account() } })
+
+    await wrapper.get('[data-testid="ollama-cloud-usage-query"]').trigger('click')
+    await flushPromises()
+
+    expect(refreshOllamaCloudUsage).toHaveBeenCalledWith(7)
+    expect(wrapper.findAllComponents(UsageProgressBar)[0].props('utilization')).toBe(43)
+    expect(wrapper.emitted<OllamaCloudUsageState[]>('updated')?.[0]?.[0]).toEqual(next)
+  })
+
+  it('hides the query action until a browser session is configured', () => {
+    const state = usageState()
+    state.configured = false
+
+    const wrapper = mount(OllamaCloudUsageCell, { props: { account: account(state) } })
+
+    expect(wrapper.find('[data-testid="ollama-cloud-usage-query"]').exists()).toBe(false)
   })
 })

From cb5e03a72027e26d18abbfa0be4dce0f77864693 Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Mon, 17 Aug 2026 12:36:58 +0800
Subject: [PATCH 32/68] fix(antigravity): preserve mixed Gemini tool config

---
 .../service/antigravity_gateway_compat.go     | 36 +++++++++++++++
 .../antigravity_gateway_compat_test.go        | 45 +++++++++++++++++++
 .../antigravity_gateway_service_test.go       | 36 +++++++++++++++
 3 files changed, 117 insertions(+)

diff --git a/backend/internal/service/antigravity_gateway_compat.go b/backend/internal/service/antigravity_gateway_compat.go
index ab8f0d17a2a4..49ca07688489 100644
--- a/backend/internal/service/antigravity_gateway_compat.go
+++ b/backend/internal/service/antigravity_gateway_compat.go
@@ -264,6 +264,10 @@ func (s *AntigravityGatewayService) buildAntigravityCompatGeminiBody(
 		if err != nil {
 			return nil, err
 		}
+		body, err = enableMixedGeminiToolInvocations(body)
+		if err != nil {
+			return nil, err
+		}
 		body = ensureGeminiFunctionCallThoughtSignatures(body)
 		body, err = injectIdentityPatchToGeminiRequest(body)
 		if err != nil {
@@ -280,6 +284,38 @@ func (s *AntigravityGatewayService) buildAntigravityCompatGeminiBody(
 	return antigravity.TransformClaudeToGeminiWithOptions(claudeRequest, projectID, mappedModel, options)
 }
 
+func enableMixedGeminiToolInvocations(body []byte) ([]byte, error) {
+	var request map[string]any
+	if err := json.Unmarshal(body, &request); err != nil {
+		return nil, err
+	}
+
+	var hasGoogleSearch, hasFunctionDeclarations bool
+	if tools, ok := request["tools"].([]any); ok {
+		for _, rawTool := range tools {
+			tool, ok := rawTool.(map[string]any)
+			if !ok {
+				continue
+			}
+			_, hasSearch := tool["googleSearch"]
+			declarations, hasFunctions := tool["functionDeclarations"].([]any)
+			hasGoogleSearch = hasGoogleSearch || hasSearch
+			hasFunctionDeclarations = hasFunctionDeclarations || hasFunctions && len(declarations) > 0
+		}
+	}
+	if !hasGoogleSearch || !hasFunctionDeclarations {
+		return body, nil
+	}
+
+	toolConfig, _ := request["toolConfig"].(map[string]any)
+	if toolConfig == nil {
+		toolConfig = make(map[string]any)
+		request["toolConfig"] = toolConfig
+	}
+	toolConfig["includeServerSideToolInvocations"] = true
+	return json.Marshal(request)
+}
+
 func antigravityCompatProxyURL(account *Account) string {
 	if account.ProxyID == nil || account.Proxy == nil {
 		return ""
diff --git a/backend/internal/service/antigravity_gateway_compat_test.go b/backend/internal/service/antigravity_gateway_compat_test.go
index ff3d195cbf63..3d2176a00135 100644
--- a/backend/internal/service/antigravity_gateway_compat_test.go
+++ b/backend/internal/service/antigravity_gateway_compat_test.go
@@ -3,6 +3,7 @@ package service
 import (
 	"bytes"
 	"context"
+	"encoding/json"
 	"io"
 	"net/http"
 	"net/http/httptest"
@@ -222,6 +223,50 @@ func TestAntigravityCompatRejectsUnsupportedAccountType(t *testing.T) {
 	}
 }
 
+func TestBuildAntigravityCompatGeminiBody_ConfiguresMixedToolInvocations(t *testing.T) {
+	svc := &AntigravityGatewayService{}
+	tests := []struct {
+		name      string
+		tools     string
+		wantField bool
+	}{
+		{
+			name:      "mixed server and client tools",
+			tools:     `[{"name":"get_weather","input_schema":{"type":"object"}},{"type":"web_search_20250305","name":"web_search"}]`,
+			wantField: true,
+		},
+		{
+			name:  "client tools only",
+			tools: `[{"name":"get_weather","input_schema":{"type":"object"}}]`,
+		},
+		{
+			name:  "server tools only",
+			tools: `[{"type":"web_search_20250305","name":"web_search"}]`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			claudeBody := []byte(`{"messages":[{"role":"user","content":"hello"}],"tools":` + tt.tools + `}`)
+			claudeBody = bytes.ReplaceAll(claudeBody, []byte{92}, nil)
+			body, err := svc.buildAntigravityCompatGeminiBody(context.Background(), claudeBody, nil, "project-1", "gemini-2.5-flash")
+			require.NoError(t, err)
+
+			var wrapped map[string]any
+			require.NoError(t, json.Unmarshal(body, &wrapped))
+			request := wrapped["request"].(map[string]any)
+			toolConfig, exists := request["toolConfig"].(map[string]any)
+			if !tt.wantField {
+				require.False(t, exists)
+				return
+			}
+			require.True(t, exists)
+			require.Equal(t, true, toolConfig["includeServerSideToolInvocations"])
+			require.NotContains(t, toolConfig, "include_server_side_tool_invocations")
+		})
+	}
+}
+
 func TestAntigravityCompatPreservesChatTokenLimit(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	tests := []struct {
diff --git a/backend/internal/service/antigravity_gateway_service_test.go b/backend/internal/service/antigravity_gateway_service_test.go
index 00fee73a1152..1a1bfc998b2c 100644
--- a/backend/internal/service/antigravity_gateway_service_test.go
+++ b/backend/internal/service/antigravity_gateway_service_test.go
@@ -340,6 +340,42 @@ func TestAntigravityGatewayService_ForwardGemini_UsesConfiguredProjectFallback(t
 	require.Equal(t, "configured-project", wrapped["project"])
 }
 
+func TestAntigravityGatewayService_ForwardGemini_PreservesServerSideToolInvocationConfig(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	body := []byte(`{"contents":[{"role":"user","parts":[{"text":"hello"}]}],"tools":[{"functionDeclarations":[{"name":"get_weather","parameters":{"type":"object","additionalProperties":false}}]},{"googleSearch":{}}],"toolConfig":{"includeServerSideToolInvocations":true}}`)
+	writer := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(writer)
+	body = bytes.ReplaceAll(body, []byte{92}, nil)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1beta/models/gemini-2.5-flash:generateContent", bytes.NewReader(body))
+
+	upstream := &queuedHTTPUpstreamStub{responses: []*http.Response{{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}},
+		Body:       io.NopCloser(strings.NewReader("data: {\"response\":{\"candidates\":[{\"content\":{\"parts\":[{\"text\":\"ok\"}]},\"finishReason\":\"STOP\"}],\"usageMetadata\":{}}}\n\n")),
+	}}}
+	svc := &AntigravityGatewayService{
+		settingService: NewSettingService(&antigravitySettingRepoStub{}, &config.Config{Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize}}),
+		tokenProvider:  &AntigravityTokenProvider{},
+		httpUpstream:   upstream,
+	}
+	account := &Account{
+		ID: 103, Name: "native-gemini", Platform: PlatformAntigravity, Type: AccountTypeOAuth, Status: StatusActive, Concurrency: 1,
+		Credentials: map[string]any{"access_token": "token", "project_id": "project-103", "model_mapping": map[string]any{"gemini-2.5-flash": "gemini-2.5-flash"}},
+	}
+
+	result, err := svc.ForwardGemini(context.Background(), c, account, "gemini-2.5-flash", "generateContent", false, body, false)
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.Len(t, upstream.requestBodies, 1)
+
+	var wrapped map[string]any
+	require.NoError(t, json.Unmarshal(upstream.requestBodies[0], &wrapped))
+	request := wrapped["request"].(map[string]any)
+	toolConfig := request["toolConfig"].(map[string]any)
+	require.Equal(t, true, toolConfig["includeServerSideToolInvocations"])
+	require.NotContains(t, toolConfig, "include_server_side_tool_invocations")
+}
+
 func TestAntigravityGatewayService_ForwardGemini_MissingProjectReturnsLocalError(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	writer := httptest.NewRecorder()

From 76b70b1685e45e6c722a71887b7ff0e1dd5681ad Mon Sep 17 00:00:00 2001
From: lyy0709 <65712338+lyy0709@users.noreply.github.com>
Date: Mon, 17 Aug 2026 13:34:41 +0800
Subject: [PATCH 33/68] fix(openai): validate bulk account settings

---
 .../account_long_context_billing_test.go      |  12 +-
 backend/internal/service/admin_account.go     |  38 ++-
 backend/internal/service/admin_service.go     |  11 +-
 .../service/admin_service_bulk_update_test.go | 290 +++++++++++++++++-
 .../service/openai_bulk_account_settings.go   | 225 ++++++++++++++
 5 files changed, 545 insertions(+), 31 deletions(-)
 create mode 100644 backend/internal/service/openai_bulk_account_settings.go

diff --git a/backend/internal/service/account_long_context_billing_test.go b/backend/internal/service/account_long_context_billing_test.go
index 709559d932aa..08bd3f0b3f05 100644
--- a/backend/internal/service/account_long_context_billing_test.go
+++ b/backend/internal/service/account_long_context_billing_test.go
@@ -258,18 +258,20 @@ func TestAdminServiceBulkUpdateAccountsRejectsMalformedOpenAILongContextBillingV
 	require.Zero(t, repo.bulkUpdateCalls)
 }
 
-func TestAdminServiceBulkUpdateAccountsAllowsProviderOwnedValueForNonOpenAIAccounts(t *testing.T) {
+func TestAdminServiceBulkUpdateAccountsRejectsOpenAILongContextKeyForNonOpenAIAccounts(t *testing.T) {
 	repo := &longContextBillingRepoStub{account: &Account{ID: 1, Platform: PlatformGrok}}
 	svc := &adminServiceImpl{accountRepo: repo}
 
 	result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
 		AccountIDs: []int64{1},
-		Extra:      map[string]any{openAILongContextBillingEnabledKey: []string{"provider-owned"}},
+		Extra:      map[string]any{openAILongContextBillingEnabledKey: true},
 	})
 
-	require.NoError(t, err)
-	require.NotNil(t, result)
-	require.Equal(t, 1, repo.bulkUpdateCalls)
+	require.Nil(t, result)
+	var appErr *infraerrors.ApplicationError
+	require.ErrorAs(t, err, &appErr)
+	require.Equal(t, "OPENAI_BULK_TARGET_INVALID", appErr.Reason)
+	require.Zero(t, repo.bulkUpdateCalls)
 }
 
 func TestAdminServiceBulkUpdateAccountsRejectsMalformedValueForMixedTargetsIncludingOpenAI(t *testing.T) {
diff --git a/backend/internal/service/admin_account.go b/backend/internal/service/admin_account.go
index 9958846658ae..77d9f353282a 100644
--- a/backend/internal/service/admin_account.go
+++ b/backend/internal/service/admin_account.go
@@ -906,26 +906,36 @@ func (s *adminServiceImpl) BulkUpdateAccounts(ctx context.Context, input *BulkUp
 			return nil, err
 		}
 	}
+	openAISettings, err := normalizeBulkOpenAISettings(input)
+	if err != nil {
+		return nil, err
+	}
 
 	needMixedChannelCheck := input.GroupIDs != nil && !input.SkipMixedChannelCheck
-	_, hasLongContextBillingUpdate := input.Extra[openAILongContextBillingEnabledKey]
 
 	// 预取所有目标账号，供凭据守卫/代理守卫/混合渠道检查共用，避免多次 DB 查询。
 	var cachedTargets []*Account
-	if len(input.Credentials) > 0 || input.ProxyID != nil || needMixedChannelCheck || hasLongContextBillingUpdate || input.ProbeEnabled != nil || input.RateMultiplier != nil {
+	if len(input.Credentials) > 0 || input.ProxyID != nil || needMixedChannelCheck || openAISettings.any() || input.ProbeEnabled != nil || input.RateMultiplier != nil {
 		loaded, err := s.accountRepo.GetByIDs(ctx, input.AccountIDs)
 		if err != nil {
 			return nil, err
 		}
 		cachedTargets = loaded
 	}
-	if input.ProbeEnabled != nil {
-		targetsByID := make(map[int64]*Account, len(cachedTargets))
-		for _, account := range cachedTargets {
-			if account != nil {
-				targetsByID[account.ID] = account
-			}
+	targetsByID := make(map[int64]*Account, len(cachedTargets))
+	for _, account := range cachedTargets {
+		if account != nil {
+			targetsByID[account.ID] = account
+		}
+	}
+	if openAISettings.any() {
+		inheritedCount, err := validateBulkOpenAISettingsTargets(input, openAISettings, targetsByID)
+		if err != nil {
+			return nil, err
 		}
+		result.LongContextInheritedCount = inheritedCount
+	}
+	if input.ProbeEnabled != nil {
 		for _, accountID := range input.AccountIDs {
 			account, ok := targetsByID[accountID]
 			if !ok {
@@ -936,18 +946,6 @@ func (s *adminServiceImpl) BulkUpdateAccounts(ctx context.Context, input *BulkUp
 			}
 		}
 	}
-	if hasLongContextBillingUpdate {
-		for _, account := range cachedTargets {
-			if account == nil || account.Platform != PlatformOpenAI {
-				continue
-			}
-			if err := ValidateOpenAILongContextBillingExtra(account.Platform, input.Extra); err != nil {
-				return nil, err
-			}
-			break
-		}
-	}
-
 	// 影子账号绝不持有凭据:批量更新携带凭据时,目标中不得含影子(外审 G5,与单账号
 	// UpdateAccount 守卫对齐)。覆盖显式 IDs 与 filter 解析出的 IDs(此处 AccountIDs 已解析完成)。
 	if len(input.Credentials) > 0 {
diff --git a/backend/internal/service/admin_service.go b/backend/internal/service/admin_service.go
index 3030b945b968..be56680764fc 100644
--- a/backend/internal/service/admin_service.go
+++ b/backend/internal/service/admin_service.go
@@ -477,11 +477,12 @@ type UserGroupRPMStatus struct {
 
 // BulkUpdateAccountsResult is the aggregated response for bulk updates.
 type BulkUpdateAccountsResult struct {
-	Success    int                       `json:"success"`
-	Failed     int                       `json:"failed"`
-	SuccessIDs []int64                   `json:"success_ids"`
-	FailedIDs  []int64                   `json:"failed_ids"`
-	Results    []BulkUpdateAccountResult `json:"results"`
+	Success                   int                       `json:"success"`
+	Failed                    int                       `json:"failed"`
+	SuccessIDs                []int64                   `json:"success_ids"`
+	FailedIDs                 []int64                   `json:"failed_ids"`
+	Results                   []BulkUpdateAccountResult `json:"results"`
+	LongContextInheritedCount int                       `json:"long_context_inherited_count,omitempty"`
 }
 
 type CreateProxyInput struct {
diff --git a/backend/internal/service/admin_service_bulk_update_test.go b/backend/internal/service/admin_service_bulk_update_test.go
index 206efb21f1b5..e935beaa9877 100644
--- a/backend/internal/service/admin_service_bulk_update_test.go
+++ b/backend/internal/service/admin_service_bulk_update_test.go
@@ -18,6 +18,8 @@ type accountRepoStubForBulkUpdate struct {
 	accountRepoStub
 	bulkUpdateErr       error
 	bulkUpdateIDs       []int64
+	bulkUpdateCalls     int
+	lastBulkUpdate      AccountBulkUpdate
 	bindGroupErrByID    map[int64]error
 	bindGroupsCalls     []int64
 	bindGroupsByAccount map[int64][]int64
@@ -50,14 +52,23 @@ type accountRepoStubForBulkUpdate struct {
 	}
 }
 
-func (s *accountRepoStubForBulkUpdate) BulkUpdate(_ context.Context, ids []int64, _ AccountBulkUpdate) (int64, error) {
+func (s *accountRepoStubForBulkUpdate) BulkUpdate(_ context.Context, ids []int64, updates AccountBulkUpdate) (int64, error) {
+	s.bulkUpdateCalls++
 	s.bulkUpdateIDs = append([]int64{}, ids...)
+	s.lastBulkUpdate = updates
 	if s.bulkUpdateErr != nil {
 		return 0, s.bulkUpdateErr
 	}
 	return int64(len(ids)), nil
 }
 
+func requireApplicationErrorReason(t *testing.T, err error, reason string) {
+	t.Helper()
+	var appErr *infraerrors.ApplicationError
+	require.ErrorAs(t, err, &appErr)
+	require.Equal(t, reason, appErr.Reason)
+}
+
 func (s *accountRepoStubForBulkUpdate) Create(_ context.Context, account *Account) error {
 	s.createAccount = account
 	if s.createID > 0 {
@@ -307,3 +318,280 @@ func TestAdminServiceBulkUpdateAccounts_ResolvesIDsFromFilters(t *testing.T) {
 	require.Equal(t, 0, result.Failed)
 	require.Equal(t, []int64{7, 11}, result.SuccessIDs)
 }
+
+func TestAdminServiceBulkUpdateAccounts_NormalizesOpenAISettings(t *testing.T) {
+	repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: []*Account{
+		{ID: 1, Platform: PlatformOpenAI, Type: AccountTypeAPIKey},
+		{ID: 2, Platform: PlatformOpenAI, Type: AccountTypeAPIKey},
+	}}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		AccountIDs: []int64{1, 2},
+		Credentials: map[string]any{
+			openAIEndpointCapabilitiesCredentialKey: []any{"chat_completions", "embeddings"},
+		},
+		Extra: map[string]any{
+			openAILongContextBillingEnabledKey: true,
+			"openai_responses_mode":            "auto",
+		},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 2, result.Success)
+	require.Zero(t, result.LongContextInheritedCount)
+	require.Equal(t, 1, repo.bulkUpdateCalls)
+	require.Contains(t, repo.lastBulkUpdate.Credentials, openAIEndpointCapabilitiesCredentialKey)
+	require.Nil(t, repo.lastBulkUpdate.Credentials[openAIEndpointCapabilitiesCredentialKey])
+	require.Equal(t, true, repo.lastBulkUpdate.Extra[openAILongContextBillingEnabledKey])
+	require.Contains(t, repo.lastBulkUpdate.Extra, "openai_responses_mode")
+	require.Nil(t, repo.lastBulkUpdate.Extra["openai_responses_mode"])
+}
+
+func TestAdminServiceBulkUpdateAccounts_AcceptsLongContextAccountTypes(t *testing.T) {
+	for _, accountType := range []string{AccountTypeOAuth, AccountTypeSetupToken, AccountTypeAPIKey} {
+		t.Run(accountType, func(t *testing.T) {
+			repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: []*Account{{
+				ID: 1, Platform: PlatformOpenAI, Type: accountType,
+			}}}
+			svc := &adminServiceImpl{accountRepo: repo}
+
+			result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+				AccountIDs: []int64{1},
+				Extra:      map[string]any{openAILongContextBillingEnabledKey: false},
+			})
+
+			require.NoError(t, err)
+			require.Equal(t, 1, result.Success)
+			require.Equal(t, 1, repo.bulkUpdateCalls)
+		})
+	}
+}
+
+func TestAdminServiceBulkUpdateAccounts_EmbeddingsOnlyResetsResponsesMode(t *testing.T) {
+	repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: []*Account{
+		{ID: 1, Platform: PlatformOpenAI, Type: AccountTypeAPIKey},
+	}}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	_, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		AccountIDs: []int64{1},
+		Credentials: map[string]any{
+			openAIEndpointCapabilitiesCredentialKey: []string{"embeddings"},
+		},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, []string{"embeddings"}, repo.lastBulkUpdate.Credentials[openAIEndpointCapabilitiesCredentialKey])
+	require.Contains(t, repo.lastBulkUpdate.Extra, "openai_responses_mode")
+	require.Nil(t, repo.lastBulkUpdate.Extra["openai_responses_mode"])
+}
+
+func TestAdminServiceBulkUpdateAccounts_RejectsInvalidOpenAISettingValuesBeforeWrite(t *testing.T) {
+	tests := []struct {
+		name        string
+		credentials map[string]any
+		extra       map[string]any
+		reason      string
+	}{
+		{name: "long context type", extra: map[string]any{openAILongContextBillingEnabledKey: "true"}, reason: "OPENAI_LONG_CONTEXT_BILLING_INVALID"},
+		{name: "empty capabilities", credentials: map[string]any{openAIEndpointCapabilitiesCredentialKey: []any{}}, reason: "OPENAI_ENDPOINT_CAPABILITIES_INVALID"},
+		{name: "unknown capability", credentials: map[string]any{openAIEndpointCapabilitiesCredentialKey: []any{"responses"}}, reason: "OPENAI_ENDPOINT_CAPABILITIES_INVALID"},
+		{name: "capabilities type", credentials: map[string]any{openAIEndpointCapabilitiesCredentialKey: "chat_completions"}, reason: "OPENAI_ENDPOINT_CAPABILITIES_INVALID"},
+		{name: "responses mode", extra: map[string]any{"openai_responses_mode": "sometimes"}, reason: "OPENAI_RESPONSES_MODE_INVALID"},
+		{name: "responses type", extra: map[string]any{"openai_responses_mode": true}, reason: "OPENAI_RESPONSES_MODE_INVALID"},
+		{
+			name:        "embeddings conflict",
+			credentials: map[string]any{openAIEndpointCapabilitiesCredentialKey: []any{"embeddings"}},
+			extra:       map[string]any{"openai_responses_mode": "force_responses"},
+			reason:      "OPENAI_RESPONSES_MODE_INVALID",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			repo := &accountRepoStubForBulkUpdate{}
+			svc := &adminServiceImpl{accountRepo: repo}
+			result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+				AccountIDs:  []int64{1},
+				Credentials: tt.credentials,
+				Extra:       tt.extra,
+			})
+			require.Nil(t, result)
+			requireApplicationErrorReason(t, err, tt.reason)
+			require.Zero(t, repo.bulkUpdateCalls)
+		})
+	}
+}
+
+func TestAdminServiceBulkUpdateAccounts_RejectsInvalidOpenAITargetsBeforeWrite(t *testing.T) {
+	tests := []struct {
+		name     string
+		accounts []*Account
+		input    *BulkUpdateAccountsInput
+	}{
+		{
+			name:     "missing account",
+			accounts: []*Account{{ID: 1, Platform: PlatformOpenAI, Type: AccountTypeOAuth}},
+			input: &BulkUpdateAccountsInput{
+				AccountIDs: []int64{1, 2},
+				Extra:      map[string]any{openAILongContextBillingEnabledKey: true},
+			},
+		},
+		{
+			name:     "mixed platform long context",
+			accounts: []*Account{{ID: 1, Platform: PlatformAnthropic, Type: AccountTypeOAuth}},
+			input: &BulkUpdateAccountsInput{
+				AccountIDs: []int64{1},
+				Extra:      map[string]any{openAILongContextBillingEnabledKey: true},
+			},
+		},
+		{
+			name:     "oauth endpoint capabilities",
+			accounts: []*Account{{ID: 1, Platform: PlatformOpenAI, Type: AccountTypeOAuth}},
+			input: &BulkUpdateAccountsInput{
+				AccountIDs:  []int64{1},
+				Credentials: map[string]any{openAIEndpointCapabilitiesCredentialKey: nil},
+			},
+		},
+		{
+			name:     "unsupported OpenAI long context account type",
+			accounts: []*Account{{ID: 1, Platform: PlatformOpenAI, Type: AccountTypeServiceAccount}},
+			input: &BulkUpdateAccountsInput{
+				AccountIDs: []int64{1},
+				Extra:      map[string]any{openAILongContextBillingEnabledKey: true},
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: tt.accounts}
+			svc := &adminServiceImpl{accountRepo: repo}
+			result, err := svc.BulkUpdateAccounts(context.Background(), tt.input)
+			require.Nil(t, result)
+			requireApplicationErrorReason(t, err, "OPENAI_BULK_TARGET_INVALID")
+			require.Zero(t, repo.bulkUpdateCalls)
+		})
+	}
+}
+
+func TestAdminServiceBulkUpdateAccounts_ForcedResponsesRequiresChatCapability(t *testing.T) {
+	repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: []*Account{{
+		ID:       1,
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeAPIKey,
+		Credentials: map[string]any{
+			openAIEndpointCapabilitiesCredentialKey: []any{"embeddings"},
+		},
+	}}}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		AccountIDs: []int64{1},
+		Extra:      map[string]any{"openai_responses_mode": "force_chat_completions"},
+	})
+
+	require.Nil(t, result)
+	requireApplicationErrorReason(t, err, "OPENAI_BULK_TARGET_INVALID")
+	require.Zero(t, repo.bulkUpdateCalls)
+}
+
+func TestAdminServiceBulkUpdateAccounts_ForcedResponsesAcceptsChatCapabilityUpdate(t *testing.T) {
+	repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: []*Account{{
+		ID:       1,
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeAPIKey,
+		Credentials: map[string]any{
+			openAIEndpointCapabilitiesCredentialKey: []any{"embeddings"},
+		},
+	}}}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	_, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		AccountIDs: []int64{1},
+		Credentials: map[string]any{
+			openAIEndpointCapabilitiesCredentialKey: []any{"chat_completions"},
+		},
+		Extra: map[string]any{"openai_responses_mode": "force_responses"},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 1, repo.bulkUpdateCalls)
+}
+
+func TestAdminServiceBulkUpdateAccounts_ReportsLongContextShadowInheritance(t *testing.T) {
+	parentID := int64(1)
+	repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: []*Account{
+		{ID: parentID, Platform: PlatformOpenAI, Type: AccountTypeOAuth},
+		{ID: 2, Platform: PlatformOpenAI, Type: AccountTypeOAuth, ParentAccountID: &parentID},
+	}}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		AccountIDs: []int64{parentID, 2},
+		Extra:      map[string]any{openAILongContextBillingEnabledKey: true},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 1, result.LongContextInheritedCount)
+	require.Equal(t, 1, repo.bulkUpdateCalls)
+}
+
+func TestAdminServiceBulkUpdateAccounts_RequiresParentForShadowOnlyLongContextUpdate(t *testing.T) {
+	parentID := int64(10)
+	repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: []*Account{
+		{ID: 1, Platform: PlatformOpenAI, Type: AccountTypeOAuth, ParentAccountID: &parentID},
+		{ID: 2, Platform: PlatformOpenAI, Type: AccountTypeOAuth, ParentAccountID: &parentID},
+	}}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		AccountIDs: []int64{1, 2},
+		Extra:      map[string]any{openAILongContextBillingEnabledKey: true},
+	})
+
+	require.Nil(t, result)
+	requireApplicationErrorReason(t, err, "OPENAI_LONG_CONTEXT_PARENT_REQUIRED")
+	require.Zero(t, repo.bulkUpdateCalls)
+}
+
+func TestAdminServiceBulkUpdateAccounts_ShadowLongContextAllowsOtherUpdates(t *testing.T) {
+	parentID := int64(10)
+	repo := &accountRepoStubForBulkUpdate{getByIDsAccounts: []*Account{{
+		ID: 1, Platform: PlatformOpenAI, Type: AccountTypeOAuth, ParentAccountID: &parentID,
+	}}}
+	svc := &adminServiceImpl{accountRepo: repo}
+	status := StatusDisabled
+
+	result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		AccountIDs: []int64{1},
+		Status:     status,
+		Extra:      map[string]any{openAILongContextBillingEnabledKey: false},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 1, result.LongContextInheritedCount)
+	require.Equal(t, 1, repo.bulkUpdateCalls)
+	require.NotNil(t, repo.lastBulkUpdate.Status)
+	require.Equal(t, status, *repo.lastBulkUpdate.Status)
+}
+
+func TestAdminServiceBulkUpdateAccounts_ValidatesFilterResolvedOpenAITargets(t *testing.T) {
+	repo := &accountRepoStubForBulkUpdate{
+		listData:         []Account{{ID: 7}},
+		listResult:       &pagination.PaginationResult{Total: 1},
+		getByIDsAccounts: []*Account{{ID: 7, Platform: PlatformAnthropic, Type: AccountTypeOAuth}},
+	}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	result, err := svc.BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		Filters: &BulkUpdateAccountFilters{Platform: PlatformOpenAI},
+		Extra:   map[string]any{openAILongContextBillingEnabledKey: true},
+	})
+
+	require.Nil(t, result)
+	requireApplicationErrorReason(t, err, "OPENAI_BULK_TARGET_INVALID")
+	require.Equal(t, []int64{7}, repo.getByIDsIDs)
+	require.Zero(t, repo.bulkUpdateCalls)
+}
diff --git a/backend/internal/service/openai_bulk_account_settings.go b/backend/internal/service/openai_bulk_account_settings.go
new file mode 100644
index 000000000000..f418e575f42f
--- /dev/null
+++ b/backend/internal/service/openai_bulk_account_settings.go
@@ -0,0 +1,225 @@
+package service
+
+import (
+	"fmt"
+	"strconv"
+
+	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai_compat"
+)
+
+type bulkOpenAISettings struct {
+	longContextBilling      bool
+	endpointCapabilities    bool
+	responsesMode           bool
+	capabilitiesIncludeChat bool
+	forcedResponsesMode     bool
+}
+
+func (s bulkOpenAISettings) any() bool {
+	return s.longContextBilling || s.endpointCapabilities || s.responsesMode
+}
+
+func normalizeBulkOpenAISettings(input *BulkUpdateAccountsInput) (bulkOpenAISettings, error) {
+	var settings bulkOpenAISettings
+	if input == nil {
+		return settings, nil
+	}
+
+	if _, exists := input.Extra[openAILongContextBillingEnabledKey]; exists {
+		settings.longContextBilling = true
+		if err := ValidateOpenAILongContextBillingExtra(PlatformOpenAI, input.Extra); err != nil {
+			return settings, err
+		}
+	}
+
+	if raw, exists := input.Credentials[openAIEndpointCapabilitiesCredentialKey]; exists {
+		settings.endpointCapabilities = true
+		capabilities, includeChat, err := normalizeBulkOpenAIEndpointCapabilities(raw)
+		if err != nil {
+			return settings, err
+		}
+		settings.capabilitiesIncludeChat = includeChat
+		input.Credentials[openAIEndpointCapabilitiesCredentialKey] = capabilities
+	}
+
+	if raw, exists := input.Extra[openai_compat.ExtraKeyResponsesMode]; exists {
+		settings.responsesMode = true
+		mode, forced, err := normalizeBulkOpenAIResponsesMode(raw)
+		if err != nil {
+			return settings, err
+		}
+		settings.forcedResponsesMode = forced
+		input.Extra[openai_compat.ExtraKeyResponsesMode] = mode
+	}
+
+	if settings.endpointCapabilities && !settings.capabilitiesIncludeChat {
+		if settings.forcedResponsesMode {
+			return settings, infraerrors.BadRequest(
+				"OPENAI_RESPONSES_MODE_INVALID",
+				"a forced Responses route requires the chat_completions endpoint capability",
+			)
+		}
+		if input.Extra == nil {
+			input.Extra = make(map[string]any, 1)
+		}
+		input.Extra[openai_compat.ExtraKeyResponsesMode] = nil
+		settings.responsesMode = true
+	}
+
+	return settings, nil
+}
+
+func normalizeBulkOpenAIEndpointCapabilities(raw any) (any, bool, error) {
+	if raw == nil {
+		return nil, true, nil
+	}
+
+	values := make([]string, 0, 2)
+	switch typed := raw.(type) {
+	case []any:
+		for _, item := range typed {
+			value, ok := item.(string)
+			if !ok {
+				return nil, false, invalidBulkOpenAIEndpointCapabilities()
+			}
+			values = append(values, value)
+		}
+	case []string:
+		values = append(values, typed...)
+	default:
+		return nil, false, invalidBulkOpenAIEndpointCapabilities()
+	}
+
+	selected := make(map[string]bool, 2)
+	for _, value := range values {
+		switch OpenAIEndpointCapability(value) {
+		case OpenAIEndpointCapabilityChatCompletions, OpenAIEndpointCapabilityEmbeddings:
+			selected[value] = true
+		default:
+			return nil, false, invalidBulkOpenAIEndpointCapabilities()
+		}
+	}
+	if len(selected) == 0 {
+		return nil, false, invalidBulkOpenAIEndpointCapabilities()
+	}
+
+	includeChat := selected[string(OpenAIEndpointCapabilityChatCompletions)]
+	if includeChat && selected[string(OpenAIEndpointCapabilityEmbeddings)] {
+		return nil, true, nil
+	}
+	if includeChat {
+		return []string{string(OpenAIEndpointCapabilityChatCompletions)}, true, nil
+	}
+	return []string{string(OpenAIEndpointCapabilityEmbeddings)}, false, nil
+}
+
+func invalidBulkOpenAIEndpointCapabilities() error {
+	return infraerrors.BadRequest(
+		"OPENAI_ENDPOINT_CAPABILITIES_INVALID",
+		"openai_capabilities must contain chat_completions, embeddings, or both",
+	)
+}
+
+func normalizeBulkOpenAIResponsesMode(raw any) (any, bool, error) {
+	if raw == nil {
+		return nil, false, nil
+	}
+	mode, ok := raw.(string)
+	if !ok {
+		return nil, false, invalidBulkOpenAIResponsesMode()
+	}
+	switch openai_compat.ResponsesSupportMode(mode) {
+	case openai_compat.ResponsesSupportModeAuto:
+		return nil, false, nil
+	case openai_compat.ResponsesSupportModeForceResponses,
+		openai_compat.ResponsesSupportModeForceChatCompletions:
+		return mode, true, nil
+	default:
+		return nil, false, invalidBulkOpenAIResponsesMode()
+	}
+}
+
+func invalidBulkOpenAIResponsesMode() error {
+	return infraerrors.BadRequest(
+		"OPENAI_RESPONSES_MODE_INVALID",
+		"openai_responses_mode must be auto, force_responses, force_chat_completions, or null",
+	)
+}
+
+func validateBulkOpenAISettingsTargets(
+	input *BulkUpdateAccountsInput,
+	settings bulkOpenAISettings,
+	targetsByID map[int64]*Account,
+) (int, error) {
+	if input == nil || !settings.any() {
+		return 0, nil
+	}
+
+	inheritedCount := 0
+	for _, accountID := range input.AccountIDs {
+		account, ok := targetsByID[accountID]
+		if !ok || account == nil {
+			return 0, invalidBulkOpenAITarget(accountID, "account does not exist")
+		}
+
+		if settings.longContextBilling {
+			if account.Platform != PlatformOpenAI || !supportsOpenAILongContextBilling(account.Type) {
+				return 0, invalidBulkOpenAITarget(accountID, "long-context billing requires an OpenAI OAuth, setup-token, or API-key account")
+			}
+			if account.IsShadow() {
+				inheritedCount++
+			}
+		}
+
+		if settings.endpointCapabilities || settings.responsesMode {
+			if account.Platform != PlatformOpenAI || account.Type != AccountTypeAPIKey {
+				return 0, invalidBulkOpenAITarget(accountID, "endpoint capabilities and Responses routing require an OpenAI API-key account")
+			}
+		}
+
+		if settings.forcedResponsesMode && !settings.capabilitiesIncludeChat &&
+			!settings.endpointCapabilities &&
+			!account.SupportsOpenAIEndpointCapability(OpenAIEndpointCapabilityChatCompletions) {
+			return 0, invalidBulkOpenAITarget(accountID, "a forced Responses route requires the chat_completions endpoint capability")
+		}
+	}
+
+	if settings.longContextBilling && inheritedCount == len(input.AccountIDs) && bulkUpdateOnlyChangesLongContext(input) {
+		return 0, infraerrors.BadRequest(
+			"OPENAI_LONG_CONTEXT_PARENT_REQUIRED",
+			"long-context billing is owned by parent accounts; select at least one parent account",
+		)
+	}
+	return inheritedCount, nil
+}
+
+func supportsOpenAILongContextBilling(accountType string) bool {
+	switch accountType {
+	case AccountTypeOAuth, AccountTypeSetupToken, AccountTypeAPIKey:
+		return true
+	default:
+		return false
+	}
+}
+
+func invalidBulkOpenAITarget(accountID int64, message string) error {
+	return infraerrors.BadRequest(
+		"OPENAI_BULK_TARGET_INVALID",
+		fmt.Sprintf("account %d: %s", accountID, message),
+	).WithMetadata(map[string]string{"account_id": strconv.FormatInt(accountID, 10)})
+}
+
+func bulkUpdateOnlyChangesLongContext(input *BulkUpdateAccountsInput) bool {
+	if input == nil || input.Name != "" || input.ProxyID != nil || input.Concurrency != nil ||
+		input.Priority != nil || input.RateMultiplier != nil || input.LoadFactor != nil ||
+		input.Status != "" || input.Schedulable != nil || input.GroupIDs != nil ||
+		len(input.Credentials) != 0 || input.ProbeEnabled != nil {
+		return false
+	}
+	if len(input.Extra) != 1 {
+		return false
+	}
+	_, ok := input.Extra[openAILongContextBillingEnabledKey]
+	return ok
+}

From b8642ef674ba80ce827cbb71df6edd091db8c3d9 Mon Sep 17 00:00:00 2001
From: Tamseno <tamseno@users.noreply.github.com>
Date: Mon, 17 Aug 2026 13:49:46 +0800
Subject: [PATCH 34/68] fix(auth): make invitation code consumption atomic with
 user creation

RegisterWithVerification checked CanUse() and then marked the code used in
two separate, non-transactional steps; the second step's failure was
swallowed ("invitation code mark failure does not affect registration").
Concurrent registrations with the same invitation code could all pass the
check and each create an account, turning a one-time invitation code into
an unlimited account factory (TOCTOU race).

Fix:
- AuthService: create user and claim the invitation code inside one DB
  transaction (createUserAndClaimInvitation). The claim reuses
  redeemRepo.Use's conditional UPDATE (WHERE status='unused'); losers are
  rejected with INVITATION_CODE_INVALID and their transaction (including
  the user insert) is rolled back. No-code registration path unchanged.
- userRepository.create: explicitly join an outer ent transaction via
  TxFromContext instead of relying on Client.Tx returning ErrTxStarted
  (ent's Tx never inspects the context, so the old reuse branch was dead
  code and user inserts always committed in their own transaction,
  leaving orphan users behind when the outer transaction rolled back).

Regression tests:
- unit: concurrent register with one invitation code must succeed exactly
  once (8 goroutines -> 1 success, 7 x INVITATION_CODE_INVALID)
- integration: outer-tx rollback removes user and releases the claim;
  commit persists both atomically
---
 backend/internal/repository/user_repo.go      |  33 +--
 .../user_repo_invitation_claim_test.go        | 113 +++++++++
 backend/internal/service/auth_service.go      |  71 +++++-
 .../auth_service_invitation_race_test.go      | 236 ++++++++++++++++++
 4 files changed, 429 insertions(+), 24 deletions(-)
 create mode 100644 backend/internal/repository/user_repo_invitation_claim_test.go
 create mode 100644 backend/internal/service/auth_service_invitation_race_test.go

diff --git a/backend/internal/repository/user_repo.go b/backend/internal/repository/user_repo.go
index def7462c12e1..4b8841bb6fac 100644
--- a/backend/internal/repository/user_repo.go
+++ b/backend/internal/repository/user_repo.go
@@ -70,24 +70,25 @@ func (r *userRepository) create(ctx context.Context, userIn *service.User, guard
 
 	// 统一使用 ent 的事务：保证用户与允许分组的更新原子化，
 	// 并避免基于 *sql.Tx 手动构造 ent client 导致的 ExecQuerier 断言错误。
-	tx, err := r.client.Tx(ctx)
-	if err != nil && !errors.Is(err, dbent.ErrTxStarted) {
-		return err
-	}
-
+	//
+	// 注意：ent 的 Client.Tx 不感知上下文中的事务（只检查 driver 类型），
+	// 因此必须显式检查 TxFromContext：当调用方已开启外部事务（如注册时的
+	// “建用户 + 占用邀请码”原子事务），直接复用其 client，由调用方统一提交/回滚，
+	// 否则用户写入会落入独立事务并自行提交，导致外层事务无法回滚（孤儿用户）。
 	var txClient *dbent.Client
 	txCtx := ctx
-	if err == nil {
-		defer func() { _ = tx.Rollback() }()
-		txClient = tx.Client()
-		txCtx = dbent.NewTxContext(ctx, tx)
+	var ownedTx *dbent.Tx
+	if existingTx := dbent.TxFromContext(ctx); existingTx != nil {
+		txClient = existingTx.Client()
 	} else {
-		// 已处于外部事务中（ErrTxStarted），复用当前事务 client 并由调用方负责提交/回滚。
-		if existingTx := dbent.TxFromContext(ctx); existingTx != nil {
-			txClient = existingTx.Client()
-		} else {
-			txClient = r.client
+		tx, err := r.client.Tx(ctx)
+		if err != nil {
+			return err
 		}
+		ownedTx = tx
+		defer func() { _ = ownedTx.Rollback() }()
+		txClient = tx.Client()
+		txCtx = dbent.NewTxContext(ctx, tx)
 	}
 
 	lockKeys := []string{normalizedEmailUniquenessLockKey(userIn.Email)}
@@ -158,8 +159,8 @@ func (r *userRepository) create(ctx context.Context, userIn *service.User, guard
 		return err
 	}
 
-	if tx != nil {
-		if err := tx.Commit(); err != nil {
+	if ownedTx != nil {
+		if err := ownedTx.Commit(); err != nil {
 			return err
 		}
 	}
diff --git a/backend/internal/repository/user_repo_invitation_claim_test.go b/backend/internal/repository/user_repo_invitation_claim_test.go
new file mode 100644
index 000000000000..ba24c365b992
--- /dev/null
+++ b/backend/internal/repository/user_repo_invitation_claim_test.go
@@ -0,0 +1,113 @@
+//go:build integration
+
+package repository
+
+import (
+	"context"
+	"testing"
+
+	dbent "github.com/Wei-Shaw/sub2api/ent"
+	"github.com/Wei-Shaw/sub2api/ent/redeemcode"
+	"github.com/Wei-Shaw/sub2api/ent/user"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/stretchr/testify/require"
+)
+
+// TestCreateWithEmailAliasGuardJoinsOuterTransaction 验证用户创建会加入调用方开启的
+// 外部 ent 事务（注册流程“建用户 + 占用邀请码”原子性的基础）：
+//   - 外层事务回滚后，用户与邀请码占用必须一并撤销（不得残留孤儿账号）；
+//   - 外层事务提交后，用户与邀请码占用同时生效。
+//
+// 回归背景：此前 create() 通过 r.client.Tx(ctx) 自开事务且自行 Commit —— ent 的
+// Client.Tx 不感知上下文事务（只检查 driver 类型），ErrTxStarted 分支实际是死代码，
+// 导致外层事务包不住用户写入；并发注册同一邀请码时，败者用户的写入已自行提交，
+// 即使注册被拒绝也会留下可登录的孤儿账号（1 个邀请码仍可生成任意数量账号）。
+func TestCreateWithEmailAliasGuardJoinsOuterTransaction(t *testing.T) {
+	client := testEntClient(t)
+	userRepo := NewUserRepository(client, integrationDB)
+	redeemRepo := NewRedeemCodeRepository(client)
+
+	ctx := context.Background()
+
+	// 清理：本测试会真实提交少量数据，确保不影响同包其它集成测试。
+	var committedUserEmails []string
+	var seededCodeIDs []int64
+	t.Cleanup(func() {
+		if len(committedUserEmails) > 0 {
+			_, _ = client.User.Delete().Where(user.EmailIn(committedUserEmails...)).Exec(ctx)
+		}
+		if len(seededCodeIDs) > 0 {
+			_, _ = client.RedeemCode.Delete().Where(redeemcode.IDIn(seededCodeIDs...)).Exec(ctx)
+		}
+	})
+
+	seedCode := func(code string) int64 {
+		_, err := client.RedeemCode.Create().
+			SetCode(code).
+			SetType(service.RedeemTypeInvitation).
+			SetStatus(service.StatusUnused).
+			SetValue(0).
+			Save(ctx)
+		require.NoError(t, err, "seed redeem code")
+		c, err := client.RedeemCode.Query().Where(redeemcode.CodeEQ(code)).Only(ctx)
+		require.NoError(t, err)
+		seededCodeIDs = append(seededCodeIDs, c.ID)
+		return c.ID
+	}
+
+	t.Run("rollback removes user and releases claim", func(t *testing.T) {
+		codeID := seedCode("ITX-RACE-ROLLBACK-001")
+		tx, err := client.Tx(ctx)
+		require.NoError(t, err)
+		txCtx := dbent.NewTxContext(ctx, tx)
+
+		u := &service.User{
+			Email:        "itx-rollback@example.com",
+			PasswordHash: "test-password-hash",
+			Role:         service.RoleUser,
+			Status:       service.StatusActive,
+			Balance:      0,
+			Concurrency:  1,
+		}
+		require.NoError(t, userRepo.CreateWithEmailAliasGuard(txCtx, u))
+		require.Greater(t, u.ID, int64(0), "create 应回填用户 ID")
+		require.NoError(t, redeemRepo.Use(txCtx, codeID, u.ID))
+		require.NoError(t, tx.Rollback())
+
+		exists, err := userRepo.ExistsByEmail(ctx, "itx-rollback@example.com")
+		require.NoError(t, err)
+		require.False(t, exists, "回滚后不得残留孤儿用户")
+
+		after, err := client.RedeemCode.Get(ctx, codeID)
+		require.NoError(t, err)
+		require.Equal(t, service.StatusUnused, after.Status, "回滚后邀请码应保持 unused")
+	})
+
+	t.Run("commit persists user and claim together", func(t *testing.T) {
+		codeID := seedCode("ITX-RACE-COMMIT-001")
+		tx, err := client.Tx(ctx)
+		require.NoError(t, err)
+		txCtx := dbent.NewTxContext(ctx, tx)
+
+		u := &service.User{
+			Email:        "itx-commit@example.com",
+			PasswordHash: "test-password-hash",
+			Role:         service.RoleUser,
+			Status:       service.StatusActive,
+			Balance:      0,
+			Concurrency:  1,
+		}
+		require.NoError(t, userRepo.CreateWithEmailAliasGuard(txCtx, u))
+		require.NoError(t, redeemRepo.Use(txCtx, codeID, u.ID))
+		require.NoError(t, tx.Commit())
+		committedUserEmails = append(committedUserEmails, u.Email)
+
+		exists, err := userRepo.ExistsByEmail(ctx, "itx-commit@example.com")
+		require.NoError(t, err)
+		require.True(t, exists, "提交后用户应存在")
+
+		after, err := client.RedeemCode.Get(ctx, codeID)
+		require.NoError(t, err)
+		require.Equal(t, service.StatusUsed, after.Status, "提交后邀请码应为 used")
+	})
+}
diff --git a/backend/internal/service/auth_service.go b/backend/internal/service/auth_service.go
index 2beba0d402c3..72adbe890232 100644
--- a/backend/internal/service/auth_service.go
+++ b/backend/internal/service/auth_service.go
@@ -245,13 +245,15 @@ func (s *AuthService) RegisterWithVerification(ctx context.Context, email, passw
 		Status:       StatusActive,
 	}
 
-	if err := s.createUserWithRegistrationEmailGuard(ctx, user); err != nil {
+	if err := s.createUserAndClaimInvitation(ctx, user, invitationRedeemCode); err != nil {
 		// 优先检查邮箱冲突错误（竞态条件下可能发生）
 		switch {
 		case errors.Is(err, ErrEmailExists):
 			return "", nil, ErrEmailExists
 		case errors.Is(err, ErrEmailDomainRegistrationLimit):
 			return "", nil, ErrEmailDomainRegistrationLimit
+		case errors.Is(err, ErrInvitationCodeInvalid):
+			return "", nil, ErrInvitationCodeInvalid
 		default:
 			logger.LegacyPrintf("service.auth", "[Auth] Database error creating user: %v", err)
 			return "", nil, ErrServiceUnavailable
@@ -273,13 +275,8 @@ func (s *AuthService) RegisterWithVerification(ctx context.Context, email, passw
 		}
 	}
 
-	// 标记邀请码为已使用（如果使用了邀请码）
-	if invitationRedeemCode != nil {
-		if err := s.redeemRepo.Use(ctx, invitationRedeemCode.ID, user.ID); err != nil {
-			// 邀请码标记失败不影响注册，只记录日志
-			logger.LegacyPrintf("service.auth", "[Auth] Failed to mark invitation code as used for user %d: %v", user.ID, err)
-		}
-	}
+	// 邀请码占用已由 createUserAndClaimInvitation 在“用户创建 + 邀请码占用”的
+	// 同一个数据库事务内原子完成（一次性约束，见函数注释），此处不再单独标记。
 	// 应用优惠码（如果提供且功能已启用）
 	if promoCode != "" && s.promoService != nil && s.settingService != nil && s.settingService.IsPromoCodeEnabled(ctx) {
 		if err := s.promoService.ApplyPromoCode(ctx, user.ID, promoCode); err != nil {
@@ -1274,6 +1271,64 @@ func (s *AuthService) createUserWithRegistrationEmailGuard(ctx context.Context,
 	return quotaRepo.CreateWithEmailAliasGuardAndDomainLimit(ctx, user, domain)
 }
 
+// createUserAndClaimInvitation 原子化完成“用户创建 + 邀请码占用”。
+//
+// 背景：邀请码属于一次性凭证，必须保证“一个邀请码最多注册一个账号”。旧实现先检查
+// CanUse()、再创建用户、最后才 redeemRepo.Use()（且失败仅记日志），检查与消耗分离且
+// 不在同一事务，并发注册可在同一邀请码上同时通过检查并各自创建账号（TOCTOU 竞态）。
+//
+// 本实现把两者放入同一个数据库事务：
+//   - 占用走 redeemRepo.Use 的条件更新（WHERE status='unused'，乐观锁）；
+//   - 并发下只有一个事务能占用成功，其余事务回滚——既不产生多余账号，也不让码被烧掉；
+//   - 事务回滚同时撤销用户创建，避免“账号已建、码被占用”的中间态。
+//
+// 无邀请码时保持原单次创建路径（不开事务）；entClient 缺失的异常配置下退化为顺序执行，
+// 并发正确性仍由 Use 的条件更新兜底（可能产生孤儿用户，但不会放行第二个注册）。
+func (s *AuthService) createUserAndClaimInvitation(ctx context.Context, user *User, invitation *RedeemCode) error {
+	commitUser := func(execCtx context.Context) error {
+		if err := s.createUserWithRegistrationEmailGuard(execCtx, user); err != nil {
+			return err
+		}
+		if invitation == nil {
+			return nil
+		}
+		// createUserWithRegistrationEmailGuard 会回填 user.ID（applyUserEntityToService），
+		// 直接以其原子占用邀请码；占用失败即整体回滚（含用户创建，见 user_repo.create
+		// 对外部事务的复用）。
+		if err := s.redeemRepo.Use(execCtx, invitation.ID, user.ID); err != nil {
+			// 并发下唯一的合法失败路径：另一个注册已占用该码
+			logger.LegacyPrintf("service.auth",
+				"[Auth] Rejected registration: invitation code %s already claimed (user_id=%d err=%v)",
+				invitation.Code, user.ID, err)
+			return ErrInvitationCodeInvalid
+		}
+		return nil
+	}
+
+	if invitation == nil {
+		return commitUser(ctx)
+	}
+	if s.entClient == nil {
+		return commitUser(ctx)
+	}
+
+	tx, err := s.entClient.Tx(ctx)
+	if err != nil {
+		logger.LegacyPrintf("service.auth", "[Auth] Failed to start registration transaction: %v", err)
+		return ErrServiceUnavailable
+	}
+	execCtx := dbent.NewTxContext(ctx, tx)
+	if err := commitUser(execCtx); err != nil {
+		_ = tx.Rollback()
+		return err
+	}
+	if err := tx.Commit(); err != nil {
+		logger.LegacyPrintf("service.auth", "[Auth] Failed to commit registration transaction: %v", err)
+		return ErrServiceUnavailable
+	}
+	return nil
+}
+
 func buildEmailSuffixNotAllowedError(whitelist []string) error {
 	if len(whitelist) == 0 {
 		return ErrEmailSuffixNotAllowed
diff --git a/backend/internal/service/auth_service_invitation_race_test.go b/backend/internal/service/auth_service_invitation_race_test.go
new file mode 100644
index 000000000000..e1e7c82aef6f
--- /dev/null
+++ b/backend/internal/service/auth_service_invitation_race_test.go
@@ -0,0 +1,236 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+)
+
+// raceSafeUserRepo 是仅覆盖注册路径的并发安全用户仓储桩。
+// 未用到的方法走嵌入接口（调用即 panic，注册路径不会触发）。
+type raceSafeUserRepo struct {
+	UserRepository
+
+	mu      sync.Mutex
+	nextID  int64
+	byEmail map[string]*User
+	byID    map[int64]*User
+}
+
+func newRaceSafeUserRepo() *raceSafeUserRepo {
+	return &raceSafeUserRepo{nextID: 1, byEmail: map[string]*User{}, byID: map[int64]*User{}}
+}
+
+func (s *raceSafeUserRepo) ExistsByEmail(_ context.Context, email string) (bool, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	_, ok := s.byEmail[email]
+	return ok, nil
+}
+
+func (s *raceSafeUserRepo) ExistsByEmailAlias(ctx context.Context, email string) (bool, error) {
+	return s.ExistsByEmail(ctx, email)
+}
+
+func (s *raceSafeUserRepo) CreateWithEmailAliasGuard(_ context.Context, user *User) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if _, ok := s.byEmail[user.Email]; ok {
+		return ErrEmailExists
+	}
+	user.ID = s.nextID
+	s.nextID++
+	clone := *user
+	s.byEmail[user.Email] = &clone
+	s.byID[user.ID] = &clone
+	return nil
+}
+
+func (s *raceSafeUserRepo) GetByEmail(_ context.Context, email string) (*User, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	u, ok := s.byEmail[email]
+	if !ok {
+		return nil, ErrUserNotFound
+	}
+	clone := *u
+	return &clone, nil
+}
+
+func (s *raceSafeUserRepo) GetByID(_ context.Context, id int64) (*User, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	u, ok := s.byID[id]
+	if !ok {
+		return nil, ErrUserNotFound
+	}
+	clone := *u
+	return &clone, nil
+}
+
+func (s *raceSafeUserRepo) Update(context.Context, *User, UserUpdateFields) error {
+	return nil
+}
+
+// raceSafeRedeemRepo 是并发安全的兑换码仓储桩：Use 以互斥锁 + 状态条件
+// 模拟数据库的条件更新（WHERE status='unused'），语义与线上实现一致。
+type raceSafeRedeemRepo struct {
+	RedeemCodeRepository
+
+	mu    sync.Mutex
+	codes map[string]*RedeemCode
+}
+
+func (s *raceSafeRedeemRepo) GetByCode(_ context.Context, code string) (*RedeemCode, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	c, ok := s.codes[code]
+	if !ok {
+		return nil, ErrRedeemCodeNotFound
+	}
+	clone := *c
+	return &clone, nil
+}
+
+func (s *raceSafeRedeemRepo) Use(_ context.Context, id, userID int64) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	for _, c := range s.codes {
+		if c.ID != id {
+			continue
+		}
+		if c.Status != StatusUnused {
+			return ErrRedeemCodeUsed
+		}
+		now := time.Now().UTC()
+		c.Status = StatusUsed
+		c.UsedBy = &userID
+		c.UsedAt = &now
+		return nil
+	}
+	return ErrRedeemCodeNotFound
+}
+
+// TestAuthService_Register_InvitationCodeSingleUseUnderConcurrency 回归测试：
+// 同一邀请码并发注册必须恰好成功 1 次，其余请求以 INVITATION_CODE_INVALID 拒绝。
+//
+// 修复前：邀请码“检查(CanUse) 与 标记已用(Use)”分离且不在同一事务，Use 失败被吞，
+// 并发请求全部注册成功（一个邀请码可创建任意数量账号）。此测试在该实现下必然失败。
+// 修复后：用户创建与邀请码占用在同一事务内原子完成（或退化路径下由 Use 条件更新
+// 兜底），并发下仅最先占码的注册成功。
+func TestAuthService_Register_InvitationCodeSingleUseUnderConcurrency(t *testing.T) {
+	const code = "INV-RACE-001"
+	userRepo := newRaceSafeUserRepo()
+	redeemRepo := &raceSafeRedeemRepo{codes: map[string]*RedeemCode{
+		code: {ID: 1, Code: code, Type: RedeemTypeInvitation, Status: StatusUnused},
+	}}
+	settings := map[string]string{
+		"registration_enabled":    "true",
+		"invitation_code_enabled": "true",
+	}
+	svc := newOAuthEmailFlowAuthService(
+		userRepo,
+		redeemRepo,
+		&refreshTokenCacheStub{},
+		settings,
+		nil, // emailCache：注册不要求邮箱验证，保持关闭
+		&userPlatformQuotaRepoStub{},
+	)
+
+	const n = 8
+	ctx := context.Background()
+	start := make(chan struct{})
+	results := make(chan error, n)
+	var wg sync.WaitGroup
+	for i := 0; i < n; i++ {
+		wg.Add(1)
+		go func(i int) {
+			defer wg.Done()
+			<-start
+			email := fmt.Sprintf("race-%d@example.com", i)
+			_, _, err := svc.RegisterWithVerification(ctx, email, "Password123!", "", "", code, "")
+			results <- err
+		}(i)
+	}
+	close(start)
+	wg.Wait()
+	close(results)
+
+	successes := 0
+	rejected := 0
+	for err := range results {
+		switch {
+		case err == nil:
+			successes++
+		case errors.Is(err, ErrInvitationCodeInvalid):
+			rejected++
+		default:
+			t.Fatalf("unexpected registration error: %v", err)
+		}
+	}
+	require.Equal(t, 1, successes, "同一邀请码并发注册必须恰好成功 1 次")
+	require.Equal(t, n-1, rejected, "其余并发请求必须以 INVITATION_CODE_INVALID 拒绝")
+
+	claimed, err := redeemRepo.GetByCode(ctx, code)
+	require.NoError(t, err)
+	require.Equal(t, StatusUsed, claimed.Status, "邀请码最终必须处于 used 状态")
+	require.NotNil(t, claimed.UsedBy, "used_by 必须记录实际注册用户")
+}
+
+// TestAuthService_Register_InvitationCodeRejectedWhenAlreadyUsed 顺序路径回归：
+// 已使用过的邀请码再次注册（即便换邮箱）必须被拒绝。
+func TestAuthService_Register_InvitationCodeRejectedWhenAlreadyUsed(t *testing.T) {
+	const code = "INV-RACE-002"
+	userRepo := newRaceSafeUserRepo()
+	redeemRepo := &raceSafeRedeemRepo{codes: map[string]*RedeemCode{
+		code: {ID: 2, Code: code, Type: RedeemTypeInvitation, Status: StatusUsed},
+	}}
+	settings := map[string]string{
+		"registration_enabled":    "true",
+		"invitation_code_enabled": "true",
+	}
+	svc := newOAuthEmailFlowAuthService(
+		userRepo,
+		redeemRepo,
+		&refreshTokenCacheStub{},
+		settings,
+		nil,
+		&userPlatformQuotaRepoStub{},
+	)
+
+	_, _, err := svc.RegisterWithVerification(context.Background(), "later@example.com", "Password123!", "", "", code, "")
+	require.ErrorIs(t, err, ErrInvitationCodeInvalid)
+}
+
+// TestAuthService_Register_InvitationCodeMissingWhenEnabled 门控回归：
+// 邀请码开启时，不带邀请码的注册必须被拒绝（不产生用户）。
+func TestAuthService_Register_InvitationCodeMissingWhenEnabled(t *testing.T) {
+	userRepo := newRaceSafeUserRepo()
+	redeemRepo := &raceSafeRedeemRepo{codes: map[string]*RedeemCode{}}
+	settings := map[string]string{
+		"registration_enabled":    "true",
+		"invitation_code_enabled": "true",
+	}
+	svc := newOAuthEmailFlowAuthService(
+		userRepo,
+		redeemRepo,
+		&refreshTokenCacheStub{},
+		settings,
+		nil,
+		&userPlatformQuotaRepoStub{},
+	)
+
+	_, _, err := svc.RegisterWithVerification(context.Background(), "no-invite@example.com", "Password123!", "", "", "", "")
+	require.ErrorIs(t, err, ErrInvitationCodeRequired)
+
+	ok, err := userRepo.ExistsByEmail(context.Background(), "no-invite@example.com")
+	require.NoError(t, err)
+	require.False(t, ok, "被拒绝的注册不应产生用户")
+}

From 8d82bb069c89c72d96b4d986f00d3bf561b70b36 Mon Sep 17 00:00:00 2001
From: lyy0709 <65712338+lyy0709@users.noreply.github.com>
Date: Mon, 17 Aug 2026 13:56:16 +0800
Subject: [PATCH 35/68] fix(openai): expose bulk account settings

---
 frontend/src/api/admin/accounts.ts            |   2 +
 .../account/BulkEditAccountModal.vue          | 288 +++++++++++++++-
 .../__tests__/BulkEditAccountModal.spec.ts    | 309 +++++++++++++++++-
 .../src/i18n/locales/en/admin/accounts.ts     |   4 +
 .../src/i18n/locales/zh/admin/accounts.ts     |   4 +
 5 files changed, 600 insertions(+), 7 deletions(-)

diff --git a/frontend/src/api/admin/accounts.ts b/frontend/src/api/admin/accounts.ts
index 21ce9f092f20..73bfd29af393 100644
--- a/frontend/src/api/admin/accounts.ts
+++ b/frontend/src/api/admin/accounts.ts
@@ -470,6 +470,7 @@ export async function bulkUpdate(
   failed: number
   success_ids?: number[]
   failed_ids?: number[]
+  long_context_inherited_count?: number
   results: Array<{ account_id: number; success: boolean; error?: string }>
   }> {
   const payload = Array.isArray(accountIdsOrPayload)
@@ -483,6 +484,7 @@ export async function bulkUpdate(
     failed: number
     success_ids?: number[]
     failed_ids?: number[]
+    long_context_inherited_count?: number
     results: Array<{ account_id: number; success: boolean; error?: string }>
   }>('/admin/accounts/bulk-update', payload)
   return data
diff --git a/frontend/src/components/account/BulkEditAccountModal.vue b/frontend/src/components/account/BulkEditAccountModal.vue
index 86379b6bb3a0..e8159c7bfb36 100644
--- a/frontend/src/components/account/BulkEditAccountModal.vue
+++ b/frontend/src/components/account/BulkEditAccountModal.vue
@@ -133,6 +133,66 @@
         </div>
       </div>
 
+      <!-- OpenAI API long-context billing -->
+      <div
+        v-if="allOpenAIPassthroughCapable"
+        class="border-t border-gray-200 pt-4 dark:border-dark-600"
+      >
+        <div class="mb-3 flex items-center justify-between gap-4">
+          <div class="flex-1">
+            <label
+              id="bulk-edit-openai-long-context-billing-label"
+              class="input-label mb-0"
+              for="bulk-edit-openai-long-context-billing-enabled"
+            >
+              {{ t('admin.accounts.openai.longContextBilling') }}
+            </label>
+            <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+              {{ t('admin.accounts.openai.longContextBillingDesc') }}
+            </p>
+          </div>
+          <input
+            v-model="enableOpenAILongContextBilling"
+            id="bulk-edit-openai-long-context-billing-enabled"
+            type="checkbox"
+            aria-controls="bulk-edit-openai-long-context-billing-body"
+            class="rounded border-gray-300 text-primary-600 focus:ring-primary-500"
+          />
+        </div>
+        <div
+          id="bulk-edit-openai-long-context-billing-body"
+          :class="!enableOpenAILongContextBilling && 'pointer-events-none opacity-50'"
+          role="group"
+          aria-labelledby="bulk-edit-openai-long-context-billing-label"
+        >
+          <button
+            type="button"
+            data-testid="bulk-edit-openai-long-context-billing-toggle"
+            role="switch"
+            :disabled="!enableOpenAILongContextBilling"
+            :aria-checked="openAILongContextBillingEnabled"
+            :class="[
+              'relative inline-flex h-6 w-11 flex-shrink-0 cursor-pointer rounded-full border-2 border-transparent transition-colors duration-200 ease-in-out focus:outline-none focus:ring-2 focus:ring-primary-500 focus:ring-offset-2',
+              openAILongContextBillingEnabled ? 'bg-primary-600' : 'bg-gray-200 dark:bg-dark-600'
+            ]"
+            @click="openAILongContextBillingEnabled = !openAILongContextBillingEnabled"
+          >
+            <span
+              :class="[
+                'pointer-events-none inline-block h-5 w-5 transform rounded-full bg-white shadow ring-0 transition duration-200 ease-in-out',
+                openAILongContextBillingEnabled ? 'translate-x-5' : 'translate-x-0'
+              ]"
+            />
+          </button>
+        </div>
+        <p
+          class="mt-3 rounded-lg bg-amber-50 px-3 py-2 text-xs text-amber-700 dark:bg-amber-900/20 dark:text-amber-300"
+          data-testid="bulk-edit-openai-long-context-shadow-hint"
+        >
+          {{ t('admin.accounts.bulkEdit.longContextShadowHint') }}
+        </p>
+      </div>
+
       <!-- Base URL (API Key only) -->
       <div class="border-t border-gray-200 pt-4 dark:border-dark-600">
         <div class="mb-3 flex items-center justify-between">
@@ -970,6 +1030,101 @@
         </div>
       </div>
 
+      <!-- OpenAI API Key endpoint capabilities -->
+      <div v-if="allOpenAIAPIKey" class="border-t border-gray-200 pt-4 dark:border-dark-600">
+        <div class="mb-3 flex items-center justify-between gap-4">
+          <div class="flex-1">
+            <label
+              id="bulk-edit-openai-endpoint-capabilities-label"
+              class="input-label mb-0"
+              for="bulk-edit-openai-endpoint-capabilities-enabled"
+            >
+              {{ t('admin.accounts.openai.endpointCapabilities') }}
+            </label>
+            <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+              {{ t('admin.accounts.openai.endpointCapabilitiesDesc') }}
+            </p>
+          </div>
+          <input
+            v-model="enableOpenAIEndpointCapabilities"
+            id="bulk-edit-openai-endpoint-capabilities-enabled"
+            type="checkbox"
+            aria-controls="bulk-edit-openai-endpoint-capabilities-body"
+            class="rounded border-gray-300 text-primary-600 focus:ring-primary-500"
+          />
+        </div>
+        <div
+          id="bulk-edit-openai-endpoint-capabilities-body"
+          :class="!enableOpenAIEndpointCapabilities && 'pointer-events-none opacity-50'"
+          role="group"
+          aria-labelledby="bulk-edit-openai-endpoint-capabilities-label"
+        >
+          <div class="grid grid-cols-1 gap-2 sm:grid-cols-2">
+            <label
+              v-for="option in openAIEndpointCapabilityOptions"
+              :key="option.value"
+              class="flex cursor-pointer items-center gap-2 rounded-lg border border-gray-200 px-3 py-2 text-sm dark:border-dark-600"
+            >
+              <input
+                type="checkbox"
+                :disabled="!enableOpenAIEndpointCapabilities"
+                class="rounded border-gray-300 text-primary-600 focus:ring-primary-500 dark:border-dark-500"
+                :data-testid="`bulk-edit-openai-endpoint-capability-${option.value}`"
+                :checked="openAIEndpointCapabilities.includes(option.value)"
+                @change="toggleOpenAIEndpointCapability(option.value, $event)"
+              />
+              <span class="text-gray-700 dark:text-gray-200">{{ option.label }}</span>
+            </label>
+          </div>
+        </div>
+      </div>
+
+      <!-- OpenAI API Key Responses route -->
+      <div v-if="allOpenAIAPIKey" class="border-t border-gray-200 pt-4 dark:border-dark-600">
+        <div class="mb-3 flex items-center justify-between gap-4">
+          <div class="flex-1">
+            <label
+              id="bulk-edit-openai-responses-mode-label"
+              class="input-label mb-0"
+              for="bulk-edit-openai-responses-mode-enabled"
+            >
+              {{ t('admin.accounts.openai.responsesMode') }}
+            </label>
+            <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+              {{ t('admin.accounts.openai.responsesModeDesc') }}
+            </p>
+          </div>
+          <input
+            v-model="enableOpenAIResponsesMode"
+            id="bulk-edit-openai-responses-mode-enabled"
+            type="checkbox"
+            aria-controls="bulk-edit-openai-responses-mode-body"
+            class="rounded border-gray-300 text-primary-600 focus:ring-primary-500"
+          />
+        </div>
+        <div
+          id="bulk-edit-openai-responses-mode-body"
+          :class="!enableOpenAIResponsesMode && 'pointer-events-none opacity-50'"
+          role="group"
+          aria-labelledby="bulk-edit-openai-responses-mode-label"
+        >
+          <Select
+            v-model="openAIResponsesMode"
+            :disabled="!enableOpenAIResponsesMode || !openAIResponsesModeApplicable"
+            data-testid="bulk-edit-openai-responses-mode-select"
+            :options="openAIResponsesModeOptions"
+            aria-labelledby="bulk-edit-openai-responses-mode-label"
+          />
+          <p
+            v-if="enableOpenAIEndpointCapabilities && !openAITextGenerationCapabilityEnabled"
+            class="mt-2 rounded-lg bg-amber-50 px-3 py-2 text-xs text-amber-700 dark:bg-amber-900/20 dark:text-amber-300"
+            data-testid="bulk-edit-openai-responses-mode-not-applicable"
+          >
+            {{ t('admin.accounts.openai.responsesModeTextDisabledHint') }}
+          </p>
+        </div>
+      </div>
+
       <!-- OpenAI API Key WS mode -->
       <div v-if="allOpenAIAPIKey" class="border-t border-gray-200 pt-4 dark:border-dark-600">
         <div class="mb-3 flex items-center justify-between">
@@ -1321,7 +1476,15 @@ import { ref, watch, computed } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { useAppStore } from '@/stores/app'
 import { adminAPI } from '@/api/admin'
-import type { Proxy as ProxyConfig, AdminGroup, AccountPlatform, AccountType, OpenAICompactMode } from '@/types'
+import type {
+  Proxy as ProxyConfig,
+  AdminGroup,
+  AccountPlatform,
+  AccountType,
+  OpenAICompactMode,
+  OpenAIEndpointCapability,
+  OpenAIResponsesMode
+} from '@/types'
 import BaseDialog from '@/components/common/BaseDialog.vue'
 import ConfirmDialog from '@/components/common/ConfirmDialog.vue'
 import Select from '@/components/common/Select.vue'
@@ -1495,6 +1658,9 @@ const enableStatus = ref(false)
 const enableGroups = ref(false)
 const enableOpenAIPassthrough = ref(false)
 const enableOpenAIFlattenNamespaces = ref(false)
+const enableOpenAILongContextBilling = ref(false)
+const enableOpenAIEndpointCapabilities = ref(false)
+const enableOpenAIResponsesMode = ref(false)
 const enableOpenAIWSMode = ref(false)
 const enableOpenAIAPIKeyWSMode = ref(false)
 const enableUpstreamBillingAutoProbe = ref(false)
@@ -1528,6 +1694,12 @@ const groupIds = ref<number[]>([])
 const openaiPassthroughEnabled = ref(false)
 // Codex namespace 工具摊平兼容开关（仅 OAuth），缺省关闭即原样保留
 const openaiFlattenNamespacesEnabled = ref(false)
+const openAILongContextBillingEnabled = ref(false)
+const openAIEndpointCapabilities = ref<OpenAIEndpointCapability[]>([
+  'chat_completions',
+  'embeddings'
+])
+const openAIResponsesMode = ref<OpenAIResponsesMode>('auto')
 const openaiOAuthResponsesWebSocketV2Mode = ref<OpenAIWSMode>(OPENAI_WS_MODE_OFF)
 const openaiAPIKeyResponsesWebSocketV2Mode = ref<OpenAIWSMode>(OPENAI_WS_MODE_OFF)
 const upstreamBillingAutoProbeMode = ref<'enabled' | 'disabled'>('enabled')
@@ -1592,6 +1764,65 @@ const openAICompactModeOptions = computed(() => [
   { value: 'force_on', label: t('admin.accounts.openai.compactModeForceOn') },
   { value: 'force_off', label: t('admin.accounts.openai.compactModeForceOff') }
 ])
+const openAIResponsesModeOptions = computed(() => [
+  { value: 'auto', label: t('admin.accounts.openai.responsesModeAuto') },
+  { value: 'force_responses', label: t('admin.accounts.openai.responsesModeForceResponses') },
+  {
+    value: 'force_chat_completions',
+    label: t('admin.accounts.openai.responsesModeForceChatCompletions')
+  }
+])
+const openAITextEndpointCapabilityLabel = computed(() => {
+  if (openAIResponsesMode.value === 'force_responses') {
+    return t('admin.accounts.openai.capabilityResponses')
+  }
+  if (openAIResponsesMode.value === 'force_chat_completions') {
+    return t('admin.accounts.openai.capabilityChatCompletions')
+  }
+  return t('admin.accounts.openai.capabilityTextAuto')
+})
+const openAIEndpointCapabilityOptions = computed<
+  Array<{ value: OpenAIEndpointCapability; label: string }>
+>(() => [
+  { value: 'chat_completions', label: openAITextEndpointCapabilityLabel.value },
+  { value: 'embeddings', label: t('admin.accounts.openai.capabilityEmbeddings') }
+])
+const openAITextGenerationCapabilityEnabled = computed(() =>
+  openAIEndpointCapabilities.value.includes('chat_completions')
+)
+const openAIResponsesModeApplicable = computed(
+  () => !enableOpenAIEndpointCapabilities.value || openAITextGenerationCapabilityEnabled.value
+)
+
+const normalizeOpenAIEndpointCapabilities = (values: OpenAIEndpointCapability[]) => {
+  const allowed: OpenAIEndpointCapability[] = ['chat_completions', 'embeddings']
+  const selected = allowed.filter((value) => values.includes(value))
+  return selected.length > 0 ? selected : allowed
+}
+
+const toggleOpenAIEndpointCapability = (
+  capability: OpenAIEndpointCapability,
+  event?: Event
+) => {
+  if (openAIEndpointCapabilities.value.includes(capability)) {
+    if (openAIEndpointCapabilities.value.length <= 1) {
+      const input = event?.target as HTMLInputElement | null
+      if (input) input.checked = true
+      return
+    }
+    openAIEndpointCapabilities.value = openAIEndpointCapabilities.value.filter(
+      (value) => value !== capability
+    )
+    if (!openAITextGenerationCapabilityEnabled.value) {
+      openAIResponsesMode.value = 'auto'
+    }
+    return
+  }
+  openAIEndpointCapabilities.value = normalizeOpenAIEndpointCapabilities([
+    ...openAIEndpointCapabilities.value,
+    capability
+  ])
+}
 const openAIWSModeConcurrencyHintKey = computed(() =>
   resolveOpenAIWSModeConcurrencyHintKey(openaiOAuthResponsesWebSocketV2Mode.value)
 )
@@ -1692,6 +1923,11 @@ const buildUpdatePayload = (): Record<string, unknown> | null => {
   const updates: Record<string, unknown> = {}
   const credentials: Record<string, unknown> = {}
   let credentialsChanged = false
+  const applyOpenAILongContextBilling =
+    enableOpenAILongContextBilling.value && allOpenAIPassthroughCapable.value
+  const applyOpenAIEndpointCapabilities =
+    enableOpenAIEndpointCapabilities.value && allOpenAIAPIKey.value
+  const applyOpenAIResponsesMode = enableOpenAIResponsesMode.value && allOpenAIAPIKey.value
   const ensureExtra = (): Record<string, unknown> => {
     if (!updates.extra) {
       updates.extra = {}
@@ -1752,6 +1988,30 @@ const buildUpdatePayload = (): Record<string, unknown> | null => {
     extra.openai_responses_flatten_namespaces = openaiFlattenNamespacesEnabled.value
   }
 
+  if (applyOpenAILongContextBilling) {
+    const extra = ensureExtra()
+    extra.openai_long_context_billing_enabled = openAILongContextBillingEnabled.value
+  }
+
+  if (applyOpenAIEndpointCapabilities) {
+    credentials.openai_capabilities =
+      openAIEndpointCapabilities.value.length === 2
+        ? null
+        : [...openAIEndpointCapabilities.value]
+    credentialsChanged = true
+  }
+
+  if (
+    applyOpenAIResponsesMode ||
+    (applyOpenAIEndpointCapabilities && !openAITextGenerationCapabilityEnabled.value)
+  ) {
+    const extra = ensureExtra()
+    extra.openai_responses_mode =
+      !openAIResponsesModeApplicable.value || openAIResponsesMode.value === 'auto'
+        ? null
+        : openAIResponsesMode.value
+  }
+
   if (enableModelRestriction.value && !isOpenAIModelRestrictionDisabled.value) {
     // 统一使用 model_mapping 字段
     if (modelRestrictionMode.value === 'whitelist') {
@@ -1931,6 +2191,9 @@ const handleSubmit = async () => {
     enableBaseUrl.value ||
     enableOpenAIPassthrough.value ||
     enableOpenAIFlattenNamespaces.value ||
+    (enableOpenAILongContextBilling.value && allOpenAIPassthroughCapable.value) ||
+    (enableOpenAIEndpointCapabilities.value && allOpenAIAPIKey.value) ||
+    (enableOpenAIResponsesMode.value && allOpenAIAPIKey.value) ||
     enableModelRestriction.value ||
     enableCustomErrorCodes.value ||
     enableInterceptWarmup.value ||
@@ -2011,11 +2274,22 @@ const submitBulkUpdate = async (baseUpdates: Record<string, unknown>) => {
       : await adminAPI.accounts.bulkUpdate(props.accountIds, updates)
     const success = res.success || 0
     const failed = res.failed || 0
+    const inherited = res.long_context_inherited_count || 0
 
     if (success > 0 && failed === 0) {
-      appStore.showSuccess(t('admin.accounts.bulkEdit.success', { count: success }))
+      if (inherited > 0) {
+        appStore.showSuccess(t('admin.accounts.bulkEdit.successWithInherited', {
+          count: success,
+          inherited
+        }))
+      } else {
+        appStore.showSuccess(t('admin.accounts.bulkEdit.success', { count: success }))
+      }
     } else if (success > 0) {
-      appStore.showError(t('admin.accounts.bulkEdit.partialSuccess', { success, failed }))
+      const key = inherited > 0
+        ? 'admin.accounts.bulkEdit.partialSuccessWithInherited'
+        : 'admin.accounts.bulkEdit.partialSuccess'
+      appStore.showError(t(key, { success, failed, inherited }))
     } else {
       appStore.showError(t('admin.accounts.bulkEdit.failed'))
     }
@@ -2035,6 +2309,8 @@ const submitBulkUpdate = async (baseUpdates: Record<string, unknown>) => {
       appStore.showError(t('admin.accounts.bulkEdit.rateSyncConflict', {
         count: error.metadata?.count ?? 1
       }))
+    } else if (error.reason === 'OPENAI_LONG_CONTEXT_PARENT_REQUIRED') {
+      appStore.showError(t('admin.accounts.bulkEdit.longContextParentRequired'))
     } else {
       appStore.showError(error.message || t('admin.accounts.bulkEdit.failed'))
       console.error('Error bulk updating accounts:', error)
@@ -2077,6 +2353,9 @@ watch(
       enableGroups.value = false
       enableOpenAIPassthrough.value = false
       enableOpenAIFlattenNamespaces.value = false
+      enableOpenAILongContextBilling.value = false
+      enableOpenAIEndpointCapabilities.value = false
+      enableOpenAIResponsesMode.value = false
       enableOpenAIWSMode.value = false
       enableOpenAIAPIKeyWSMode.value = false
       enableUpstreamBillingAutoProbe.value = false
@@ -2092,6 +2371,9 @@ watch(
       baseUrl.value = ''
       openaiPassthroughEnabled.value = false
       openaiFlattenNamespacesEnabled.value = false
+      openAILongContextBillingEnabled.value = false
+      openAIEndpointCapabilities.value = ['chat_completions', 'embeddings']
+      openAIResponsesMode.value = 'auto'
       modelRestrictionMode.value = 'whitelist'
       allowedModels.value = []
       modelMappings.value = []
diff --git a/frontend/src/components/account/__tests__/BulkEditAccountModal.spec.ts b/frontend/src/components/account/__tests__/BulkEditAccountModal.spec.ts
index 45b8473cbf4f..dec8be054a92 100644
--- a/frontend/src/components/account/__tests__/BulkEditAccountModal.spec.ts
+++ b/frontend/src/components/account/__tests__/BulkEditAccountModal.spec.ts
@@ -1,17 +1,20 @@
 import { describe, expect, it, vi, beforeEach } from 'vitest'
 import { flushPromises, mount } from '@vue/test-utils'
+import { nextTick } from 'vue'
 import BulkEditAccountModal from '../BulkEditAccountModal.vue'
 import ModelWhitelistSelector from '../ModelWhitelistSelector.vue'
 import { adminAPI } from '@/api/admin'
 
-const { showError } = vi.hoisted(() => ({
-  showError: vi.fn()
+const { showError, showSuccess, translate } = vi.hoisted(() => ({
+  showError: vi.fn(),
+  showSuccess: vi.fn(),
+  translate: vi.fn((key: string) => key)
 }))
 
 vi.mock('@/stores/app', () => ({
   useAppStore: () => ({
     showError,
-    showSuccess: vi.fn(),
+    showSuccess,
     showInfo: vi.fn()
   })
 }))
@@ -34,7 +37,7 @@ vi.mock('vue-i18n', async () => {
   return {
     ...actual,
     useI18n: () => ({
-      t: (key: string) => key
+      t: translate
     })
   }
 })
@@ -82,6 +85,8 @@ describe('BulkEditAccountModal', () => {
     vi.mocked(adminAPI.accounts.bulkUpdate).mockReset()
     vi.mocked(adminAPI.accounts.checkMixedChannelRisk).mockReset()
     showError.mockReset()
+    showSuccess.mockReset()
+    translate.mockClear()
 
     vi.mocked(adminAPI.accounts.bulkUpdate).mockResolvedValue({
       success: 2,
@@ -403,6 +408,302 @@ describe('BulkEditAccountModal', () => {
     })
   })
 
+  it('OpenAI 支持类型展示长上下文设置，混合平台隐藏全部新增设置', () => {
+    for (const selectedTypes of [['oauth'], ['setup-token'], ['apikey'], ['oauth', 'setup-token', 'apikey']]) {
+      const wrapper = mountModal({
+        selectedPlatforms: ['openai'],
+        selectedTypes
+      })
+      expect(wrapper.find('#bulk-edit-openai-long-context-billing-enabled').exists()).toBe(true)
+      wrapper.unmount()
+    }
+
+    const mixed = mountModal({
+      selectedPlatforms: ['openai', 'anthropic'],
+      selectedTypes: ['apikey']
+    })
+    expect(mixed.find('#bulk-edit-openai-long-context-billing-enabled').exists()).toBe(false)
+    expect(mixed.find('#bulk-edit-openai-endpoint-capabilities-enabled').exists()).toBe(false)
+    expect(mixed.find('#bulk-edit-openai-responses-mode-enabled').exists()).toBe(false)
+  })
+
+  it('端点能力与 Responses 路由仅对全部 OpenAI API Key 目标展示', () => {
+    const apiKey = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+    expect(apiKey.find('#bulk-edit-openai-endpoint-capabilities-enabled').exists()).toBe(true)
+    expect(apiKey.find('#bulk-edit-openai-responses-mode-enabled').exists()).toBe(true)
+    apiKey.unmount()
+
+    const oauth = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['oauth']
+    })
+    expect(oauth.find('#bulk-edit-openai-endpoint-capabilities-enabled').exists()).toBe(false)
+    expect(oauth.find('#bulk-edit-openai-responses-mode-enabled').exists()).toBe(false)
+  })
+
+  it('长上下文设置独立启用并提交布尔值', async () => {
+    const enabledWrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['oauth']
+    })
+
+    await enabledWrapper.get('#bulk-edit-openai-long-context-billing-enabled').setValue(true)
+    await enabledWrapper.get('[data-testid="bulk-edit-openai-long-context-billing-toggle"]').trigger('click')
+    await enabledWrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenLastCalledWith([1, 2], {
+      extra: { openai_long_context_billing_enabled: true }
+    })
+    enabledWrapper.unmount()
+
+    vi.mocked(adminAPI.accounts.bulkUpdate).mockClear()
+    const disabledWrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['setup-token']
+    })
+    await disabledWrapper.get('#bulk-edit-openai-long-context-billing-enabled').setValue(true)
+    await disabledWrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledWith([1, 2], {
+      extra: { openai_long_context_billing_enabled: false }
+    })
+  })
+
+  it('端点能力默认值提交 null，表示恢复两个默认端点', async () => {
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+
+    await wrapper.get('#bulk-edit-openai-endpoint-capabilities-enabled').setValue(true)
+    await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledWith([1, 2], {
+      credentials: { openai_capabilities: null }
+    })
+  })
+
+  it('Responses 路由独立启用，auto 提交 null，强制模式提交明确值', async () => {
+    const autoWrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+    await autoWrapper.get('#bulk-edit-openai-responses-mode-enabled').setValue(true)
+    await autoWrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenLastCalledWith([1, 2], {
+      extra: { openai_responses_mode: null }
+    })
+    autoWrapper.unmount()
+
+    vi.mocked(adminAPI.accounts.bulkUpdate).mockClear()
+    const forcedWrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+    await forcedWrapper.get('#bulk-edit-openai-responses-mode-enabled').setValue(true)
+    await forcedWrapper.get('[data-testid="bulk-edit-openai-responses-mode-select"]').setValue('force_responses')
+    await forcedWrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledWith([1, 2], {
+      extra: { openai_responses_mode: 'force_responses' }
+    })
+  })
+
+  it('仅启用 Embeddings 时恢复 Responses 自动模式并精确提交联动字段', async () => {
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+
+    await wrapper.get('#bulk-edit-openai-endpoint-capabilities-enabled').setValue(true)
+    await wrapper.get('#bulk-edit-openai-responses-mode-enabled').setValue(true)
+    await wrapper.get('[data-testid="bulk-edit-openai-responses-mode-select"]').setValue('force_chat_completions')
+    await wrapper.get('[data-testid="bulk-edit-openai-endpoint-capability-chat_completions"]').setValue(false)
+
+    expect((wrapper.get('[data-testid="bulk-edit-openai-responses-mode-select"]').element as HTMLSelectElement).value)
+      .toBe('auto')
+    expect(wrapper.find('[data-testid="bulk-edit-openai-responses-mode-not-applicable"]').exists()).toBe(true)
+
+    await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledWith([1, 2], {
+      credentials: { openai_capabilities: ['embeddings'] },
+      extra: { openai_responses_mode: null }
+    })
+  })
+
+  it('关闭端点能力修改后 Responses 路由恢复独立可编辑', async () => {
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+
+    await wrapper.get('#bulk-edit-openai-endpoint-capabilities-enabled').setValue(true)
+    await wrapper.get('[data-testid="bulk-edit-openai-endpoint-capability-chat_completions"]').setValue(false)
+    await wrapper.get('#bulk-edit-openai-endpoint-capabilities-enabled').setValue(false)
+    await wrapper.get('#bulk-edit-openai-responses-mode-enabled').setValue(true)
+
+    const select = wrapper.get('[data-testid="bulk-edit-openai-responses-mode-select"]')
+    expect(select.attributes('disabled')).toBeUndefined()
+    await select.setValue('force_responses')
+    await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledWith([1, 2], {
+      extra: { openai_responses_mode: 'force_responses' }
+    })
+  })
+
+  it('目标变化后不提交已经隐藏的 OpenAI 设置', async () => {
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+
+    await wrapper.get('#bulk-edit-openai-long-context-billing-enabled').setValue(true)
+    await wrapper.get('#bulk-edit-openai-endpoint-capabilities-enabled').setValue(true)
+    await wrapper.get('#bulk-edit-openai-responses-mode-enabled').setValue(true)
+    await wrapper.setProps({ selectedPlatforms: ['anthropic'], selectedTypes: ['apikey'] })
+    await wrapper.get('#bulk-edit-status-enabled').setValue(true)
+    await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledWith([1, 2], {
+      status: 'active'
+    })
+  })
+
+  it('至少保留一个端点能力', async () => {
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+    await wrapper.get('#bulk-edit-openai-endpoint-capabilities-enabled').setValue(true)
+    await wrapper.get('[data-testid="bulk-edit-openai-endpoint-capability-chat_completions"]').setValue(false)
+    await wrapper.get('[data-testid="bulk-edit-openai-endpoint-capability-embeddings"]').setValue(false)
+
+    expect((wrapper.get('[data-testid="bulk-edit-openai-endpoint-capability-embeddings"]').element as HTMLInputElement).checked)
+      .toBe(true)
+  })
+
+  it('关闭弹窗后重置新增设置的启用状态和值', async () => {
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['apikey']
+    })
+    await wrapper.get('#bulk-edit-openai-long-context-billing-enabled').setValue(true)
+    await wrapper.get('[data-testid="bulk-edit-openai-long-context-billing-toggle"]').trigger('click')
+    await wrapper.get('#bulk-edit-openai-endpoint-capabilities-enabled').setValue(true)
+    await wrapper.get('[data-testid="bulk-edit-openai-endpoint-capability-chat_completions"]').setValue(false)
+    await wrapper.get('#bulk-edit-openai-responses-mode-enabled').setValue(true)
+
+    await wrapper.setProps({ show: false })
+    await nextTick()
+
+    expect((wrapper.get('#bulk-edit-openai-long-context-billing-enabled').element as HTMLInputElement).checked).toBe(false)
+    expect(wrapper.get('[data-testid="bulk-edit-openai-long-context-billing-toggle"]').attributes('aria-checked')).toBe('false')
+    expect((wrapper.get('#bulk-edit-openai-endpoint-capabilities-enabled').element as HTMLInputElement).checked).toBe(false)
+    expect((wrapper.get('[data-testid="bulk-edit-openai-endpoint-capability-chat_completions"]').element as HTMLInputElement).checked).toBe(true)
+    expect((wrapper.get('[data-testid="bulk-edit-openai-responses-mode-select"]').element as HTMLSelectElement).value).toBe('auto')
+  })
+
+  it('筛选全量模式固定展示影子继承说明并按 filters 提交', async () => {
+    const wrapper = mountModal({
+      accountIds: [],
+      selectedPlatforms: [],
+      selectedTypes: [],
+      target: {
+        mode: 'filtered',
+        filters: { platform: 'openai', type: 'oauth', status: 'active' },
+        previewCount: 20,
+        selectedPlatforms: ['openai'],
+        selectedTypes: ['oauth']
+      }
+    })
+
+    expect(wrapper.get('[data-testid="bulk-edit-openai-long-context-shadow-hint"]').text())
+      .toContain('admin.accounts.bulkEdit.longContextShadowHint')
+    await wrapper.get('#bulk-edit-openai-long-context-billing-enabled').setValue(true)
+    await wrapper.get('[data-testid="bulk-edit-openai-long-context-billing-toggle"]').trigger('click')
+    await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledWith({
+      filters: { platform: 'openai', type: 'oauth', status: 'active' },
+      extra: { openai_long_context_billing_enabled: true }
+    })
+  })
+
+  it('成功响应包含影子继承数量时展示专用提示', async () => {
+    vi.mocked(adminAPI.accounts.bulkUpdate).mockResolvedValueOnce({
+      success: 2,
+      failed: 0,
+      long_context_inherited_count: 1,
+      results: []
+    } as any)
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['oauth']
+    })
+    await wrapper.get('#bulk-edit-openai-long-context-billing-enabled').setValue(true)
+    await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(showSuccess).toHaveBeenCalledWith('admin.accounts.bulkEdit.successWithInherited')
+    expect(translate).toHaveBeenCalledWith('admin.accounts.bulkEdit.successWithInherited', {
+      count: 2,
+      inherited: 1
+    })
+  })
+
+  it('部分成功且包含影子继承数量时展示组合提示', async () => {
+    vi.mocked(adminAPI.accounts.bulkUpdate).mockResolvedValueOnce({
+      success: 1,
+      failed: 1,
+      long_context_inherited_count: 1,
+      results: []
+    } as any)
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['oauth']
+    })
+    await wrapper.get('#bulk-edit-openai-long-context-billing-enabled').setValue(true)
+    await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(showError).toHaveBeenCalledWith('admin.accounts.bulkEdit.partialSuccessWithInherited')
+    expect(translate).toHaveBeenCalledWith('admin.accounts.bulkEdit.partialSuccessWithInherited', {
+      success: 1,
+      failed: 1,
+      inherited: 1
+    })
+  })
+
+  it('全影子长上下文错误使用专用提示并保持弹窗打开', async () => {
+    vi.mocked(adminAPI.accounts.bulkUpdate).mockRejectedValueOnce({
+      status: 400,
+      reason: 'OPENAI_LONG_CONTEXT_PARENT_REQUIRED',
+      message: 'select parent'
+    })
+    const wrapper = mountModal({
+      selectedPlatforms: ['openai'],
+      selectedTypes: ['oauth']
+    })
+    await wrapper.get('#bulk-edit-openai-long-context-billing-enabled').setValue(true)
+    await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
+    await flushPromises()
+
+    expect(showError).toHaveBeenCalledWith('admin.accounts.bulkEdit.longContextParentRequired')
+    expect(wrapper.emitted('close')).toBeUndefined()
+  })
+
   it('OpenAI API Key 批量编辑可统一开启上游倍率自动探测', async () => {
     const wrapper = mountModal({
       selectedPlatforms: ['openai'],
diff --git a/frontend/src/i18n/locales/en/admin/accounts.ts b/frontend/src/i18n/locales/en/admin/accounts.ts
index 0eb245033f2f..03002333457a 100644
--- a/frontend/src/i18n/locales/en/admin/accounts.ts
+++ b/frontend/src/i18n/locales/en/admin/accounts.ts
@@ -431,12 +431,16 @@ export default {
         submit: 'Update Accounts',
         updating: 'Updating...',
         success: 'Updated {count} account(s)',
+        successWithInherited: 'Updated {count} account(s). {inherited} selected shadow account(s) still follow their parent account.',
         partialSuccess: 'Partially updated: {success} succeeded, {failed} failed',
+        partialSuccessWithInherited: 'Partially updated: {success} succeeded, {failed} failed. {inherited} selected shadow account(s) still follow their parent account.',
         failed: 'Bulk update failed',
         noSelection: 'Please select accounts to edit',
         noFieldsSelected: 'Select at least one field to update',
         rateSyncWarning: 'Accounts with upstream rate sync enabled cannot be changed in bulk. Disable sync in the account editor first.',
         rateSyncConflict: 'Cannot change account rates: {count} target account(s) have upstream rate sync enabled.',
+        longContextShadowHint: 'Long-context billing belongs to the parent account. Selected shadow accounts keep following their parent, including when targets come from a filter.',
+        longContextParentRequired: 'All selected accounts are shadows. Select the parent account to change long-context billing.',
         mixedPlatformWarning: 'Selected accounts span multiple platforms ({platforms}). Model mapping presets shown are combined — ensure mappings are appropriate for each platform.'
       },
       bulkDeleteTitle: 'Bulk Delete Accounts',
diff --git a/frontend/src/i18n/locales/zh/admin/accounts.ts b/frontend/src/i18n/locales/zh/admin/accounts.ts
index 1c6d000c3c75..05340b101afa 100644
--- a/frontend/src/i18n/locales/zh/admin/accounts.ts
+++ b/frontend/src/i18n/locales/zh/admin/accounts.ts
@@ -509,12 +509,16 @@ export default {
         submit: '批量更新',
         updating: '更新中...',
         success: '成功更新 {count} 个账号',
+        successWithInherited: '成功更新 {count} 个账号；其中 {inherited} 个影子账号仍跟随母账号。',
         partialSuccess: '部分更新成功：成功 {success} 个，失败 {failed} 个',
+        partialSuccessWithInherited: '部分更新成功：成功 {success} 个，失败 {failed} 个；其中 {inherited} 个影子账号仍跟随母账号。',
         failed: '批量更新失败',
         noSelection: '请选择要编辑的账号',
         noFieldsSelected: '请至少选择一个要更新的字段',
         rateSyncWarning: '已开启上游倍率同步的账号不能批量手工修改倍率，请先在账号编辑页关闭同步。',
         rateSyncConflict: '无法修改账号倍率：{count} 个目标账号已开启上游倍率同步。',
+        longContextShadowHint: '长上下文计费归母账号所有。选中的影子账号仍跟随母账号，筛选全量目标时同样如此。',
+        longContextParentRequired: '选中的账号全部是影子账号，请选择母账号修改长上下文计费。',
         mixedPlatformWarning: '所选账号跨越多个平台（{platforms}）。显示的模型映射预设为合并结果——请确保映射对每个平台都适用。'
       },
       bulkDeleteTitle: '批量删除账号',

From 3c3bb2fa1949c7e5d0dd4e9b4cb07cab61595cd3 Mon Sep 17 00:00:00 2001
From: Tamseno <tamseno@users.noreply.github.com>
Date: Mon, 17 Aug 2026 15:23:49 +0800
Subject: [PATCH 36/68] fix(gemini): support includeServerSideToolInvocations
 in GeminiToolConfig

- Add IncludeServerSideToolInvocations field to GeminiToolConfig to prevent dropping client tool settings.
- Fix HTTP 400 error when mixing built-in tools (e.g. Google Search) with function calling on Gemini 3.6/3.7 models.
- Add serialization/deserialization unit test TestGeminiToolConfig_IncludeServerSideToolInvocations.

Fixes #5709
---
 .../internal/pkg/antigravity/gemini_types.go  |  3 ++-
 .../antigravity/request_transformer_test.go   | 24 +++++++++++++++++++
 2 files changed, 26 insertions(+), 1 deletion(-)

diff --git a/backend/internal/pkg/antigravity/gemini_types.go b/backend/internal/pkg/antigravity/gemini_types.go
index 033dccbd5d3d..4fd7882adc1b 100644
--- a/backend/internal/pkg/antigravity/gemini_types.go
+++ b/backend/internal/pkg/antigravity/gemini_types.go
@@ -112,7 +112,8 @@ type GeminiImageSearch struct {
 
 // GeminiToolConfig Gemini 工具配置
 type GeminiToolConfig struct {
-	FunctionCallingConfig *GeminiFunctionCallingConfig `json:"functionCallingConfig,omitempty"`
+	FunctionCallingConfig            *GeminiFunctionCallingConfig `json:"functionCallingConfig,omitempty"`
+	IncludeServerSideToolInvocations *bool                        `json:"includeServerSideToolInvocations,omitempty"`
 }
 
 // GeminiFunctionCallingConfig 函数调用配置
diff --git a/backend/internal/pkg/antigravity/request_transformer_test.go b/backend/internal/pkg/antigravity/request_transformer_test.go
index ea95cc7ea656..008e54724a93 100644
--- a/backend/internal/pkg/antigravity/request_transformer_test.go
+++ b/backend/internal/pkg/antigravity/request_transformer_test.go
@@ -565,3 +565,27 @@ func TestTransformClaudeToGeminiWithOptions_PreservesWebSearchAlongsideFunctions
 	require.Equal(t, "get_weather", req.Request.Tools[0].FunctionDeclarations[0].Name)
 	require.NotNil(t, req.Request.Tools[1].GoogleSearch)
 }
+
+func TestGeminiToolConfig_IncludeServerSideToolInvocations(t *testing.T) {
+	t.Run("serialize and deserialize toolConfig with includeServerSideToolInvocations", func(t *testing.T) {
+		trueVal := true
+		cfg := GeminiToolConfig{
+			FunctionCallingConfig: &GeminiFunctionCallingConfig{
+				Mode: "VALIDATED",
+			},
+			IncludeServerSideToolInvocations: &trueVal,
+		}
+
+		data, err := json.Marshal(cfg)
+		require.NoError(t, err)
+		require.Contains(t, string(data), `"includeServerSideToolInvocations":true`)
+
+		var decoded GeminiToolConfig
+		err = json.Unmarshal(data, &decoded)
+		require.NoError(t, err)
+		require.NotNil(t, decoded.IncludeServerSideToolInvocations)
+		require.True(t, *decoded.IncludeServerSideToolInvocations)
+		require.Equal(t, "VALIDATED", decoded.FunctionCallingConfig.Mode)
+	})
+}
+

From 971544570d6da41c6486d0e92f225bf28dfe2d7a Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Mon, 17 Aug 2026 17:11:33 +0800
Subject: [PATCH 37/68] test(antigravity): check tool config assertions

---
 backend/internal/service/antigravity_gateway_compat_test.go | 3 ++-
 .../internal/service/antigravity_gateway_service_test.go    | 6 ++++--
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/backend/internal/service/antigravity_gateway_compat_test.go b/backend/internal/service/antigravity_gateway_compat_test.go
index 3d2176a00135..3df0b0081a7a 100644
--- a/backend/internal/service/antigravity_gateway_compat_test.go
+++ b/backend/internal/service/antigravity_gateway_compat_test.go
@@ -254,7 +254,8 @@ func TestBuildAntigravityCompatGeminiBody_ConfiguresMixedToolInvocations(t *test
 
 			var wrapped map[string]any
 			require.NoError(t, json.Unmarshal(body, &wrapped))
-			request := wrapped["request"].(map[string]any)
+			request, ok := wrapped["request"].(map[string]any)
+			require.True(t, ok)
 			toolConfig, exists := request["toolConfig"].(map[string]any)
 			if !tt.wantField {
 				require.False(t, exists)
diff --git a/backend/internal/service/antigravity_gateway_service_test.go b/backend/internal/service/antigravity_gateway_service_test.go
index 1a1bfc998b2c..d1c389c7a525 100644
--- a/backend/internal/service/antigravity_gateway_service_test.go
+++ b/backend/internal/service/antigravity_gateway_service_test.go
@@ -370,8 +370,10 @@ func TestAntigravityGatewayService_ForwardGemini_PreservesServerSideToolInvocati
 
 	var wrapped map[string]any
 	require.NoError(t, json.Unmarshal(upstream.requestBodies[0], &wrapped))
-	request := wrapped["request"].(map[string]any)
-	toolConfig := request["toolConfig"].(map[string]any)
+	request, ok := wrapped["request"].(map[string]any)
+	require.True(t, ok)
+	toolConfig, ok := request["toolConfig"].(map[string]any)
+	require.True(t, ok)
 	require.Equal(t, true, toolConfig["includeServerSideToolInvocations"])
 	require.NotContains(t, toolConfig, "include_server_side_tool_invocations")
 }

From 7cdca9e4958045dab38a77769b2b82402c83a365 Mon Sep 17 00:00:00 2001
From: shaw <dev@sub2api.org>
Date: Mon, 17 Aug 2026 17:28:51 +0800
Subject: [PATCH 38/68] =?UTF-8?q?feat(groups):=20=E6=94=BE=E8=A1=8C=20kimi?=
 =?UTF-8?q?/zhipu/deepseek=20=E5=B9=B3=E5=8F=B0=E5=88=86=E7=BB=84=E5=88=9B?=
 =?UTF-8?q?=E5=BB=BA=E5=85=A5=E5=8F=A3?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

PR #5666 引入 CN 平台后,路由/调度/前端类型均已支持 CN 平台分组,但分组
创建入口两头缺失:后端 Create/UpdateGroupRequest 的 platform oneof 白名单
与前端 GroupsView 平台选项都没有三平台,导致 CN 账号「无可用分组」、整条
流量链路不通(composite 不能作为替代:CN 不可为 composite 路由目标)。

- group_handler.go: 两处 oneof 加 kimi/zhipu/deepseek;composite 路由目标
  白名单有意不动(DetectModelPlatform/isConcreteRequestPlatform 均无 CN 分支)
- GroupsView: platformOptions/platformFilterOptions 补三项;两处徽章配色链
  按 platformColors.ts 色系补 CN 分支
- i18n: admin.groups.platforms 补 kimi/zhipu/deepseek 键(zh/en),缺键时
  分组徽章/GroupRPM/RateMultipliers 弹窗/ChannelsView 会渲染原始 key
- GroupBadge: badgeClass/labelClass 补 CN 配色
- 新增表驱动测试:9 平台 Create/Update 全放行、非法值(别名/大小写/空格)
  全拒绝、composite target 对 CN 保持拒绝的守卫
---
 .../internal/handler/admin/group_handler.go   |  4 +-
 .../admin/group_handler_platform_test.go      | 83 +++++++++++++++++++
 frontend/src/components/common/GroupBadge.vue | 24 ++++++
 .../src/i18n/locales/en/admin/overview.ts     |  3 +
 .../src/i18n/locales/zh/admin/overview.ts     |  3 +
 frontend/src/views/admin/GroupsView.vue       | 22 ++++-
 6 files changed, 135 insertions(+), 4 deletions(-)
 create mode 100644 backend/internal/handler/admin/group_handler_platform_test.go

diff --git a/backend/internal/handler/admin/group_handler.go b/backend/internal/handler/admin/group_handler.go
index dfbd4eba6f42..1c3ecb07e13b 100644
--- a/backend/internal/handler/admin/group_handler.go
+++ b/backend/internal/handler/admin/group_handler.go
@@ -98,7 +98,7 @@ func NewGroupHandler(adminService service.AdminService, dashboardService *servic
 type CreateGroupRequest struct {
 	Name                      string                        `json:"name" binding:"required"`
 	Description               string                        `json:"description"`
-	Platform                  string                        `json:"platform" binding:"omitempty,oneof=anthropic openai gemini antigravity grok composite"`
+	Platform                  string                        `json:"platform" binding:"omitempty,oneof=anthropic openai gemini antigravity grok kimi zhipu deepseek composite"`
 	RateMultiplier            float64                       `json:"rate_multiplier"`
 	IsExclusive               bool                          `json:"is_exclusive"`
 	SubscriptionType          string                        `json:"subscription_type" binding:"omitempty,oneof=standard subscription"`
@@ -166,7 +166,7 @@ type CreateGroupRequest struct {
 type UpdateGroupRequest struct {
 	Name                      string                         `json:"name"`
 	Description               *string                        `json:"description"`
-	Platform                  string                         `json:"platform" binding:"omitempty,oneof=anthropic openai gemini antigravity grok composite"`
+	Platform                  string                         `json:"platform" binding:"omitempty,oneof=anthropic openai gemini antigravity grok kimi zhipu deepseek composite"`
 	RateMultiplier            *float64                       `json:"rate_multiplier"`
 	IsExclusive               *bool                          `json:"is_exclusive"`
 	Status                    string                         `json:"status" binding:"omitempty,oneof=active inactive"`
diff --git a/backend/internal/handler/admin/group_handler_platform_test.go b/backend/internal/handler/admin/group_handler_platform_test.go
new file mode 100644
index 000000000000..ca1703180cba
--- /dev/null
+++ b/backend/internal/handler/admin/group_handler_platform_test.go
@@ -0,0 +1,83 @@
+//go:build unit
+
+package admin
+
+import (
+	"bytes"
+	"fmt"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+// 回归分组平台枚举:kimi/zhipu/deepseek 必须能通过 Create/Update 的 binding 校验
+// （历史 bug:调度/路由链路已支持 CN 平台分组,但 oneof 白名单漏加三平台,导致
+// 平台分组无法创建、CN 账号"无可用分组"）;非法值仍须被拒。
+func bindGroupPlatformJSON(t *testing.T, target any, body string) error {
+	t.Helper()
+	gin.SetMode(gin.TestMode)
+	c, _ := gin.CreateTestContext(httptest.NewRecorder())
+	c.Request = httptest.NewRequest("POST", "/", bytes.NewBufferString(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+	return c.ShouldBindJSON(target)
+}
+
+func TestGroupPlatformBinding_AllowedPlatforms(t *testing.T) {
+	allowed := []string{
+		"anthropic", "openai", "gemini", "antigravity", "grok",
+		"kimi", "zhipu", "deepseek", "composite",
+	}
+	for _, platform := range allowed {
+		t.Run("create_"+platform, func(t *testing.T) {
+			var req CreateGroupRequest
+			body := fmt.Sprintf(`{"name":"g","platform":%q}`, platform)
+			require.NoError(t, bindGroupPlatformJSON(t, &req, body),
+				"platform %q 应通过 CreateGroupRequest 校验", platform)
+			require.Equal(t, platform, req.Platform)
+		})
+		t.Run("update_"+platform, func(t *testing.T) {
+			var req UpdateGroupRequest
+			body := fmt.Sprintf(`{"platform":%q}`, platform)
+			require.NoError(t, bindGroupPlatformJSON(t, &req, body),
+				"platform %q 应通过 UpdateGroupRequest 校验", platform)
+			require.Equal(t, platform, req.Platform)
+		})
+	}
+}
+
+func TestGroupPlatformBinding_RejectsInvalidPlatforms(t *testing.T) {
+	invalid := []string{
+		"moonshot", // 厂商别名,不是平台标识
+		"Kimi",     // 大小写敏感
+		"openai ",  // 尾随空格
+		"glm",
+		"bogus",
+	}
+	for _, platform := range invalid {
+		t.Run("create_"+platform, func(t *testing.T) {
+			var req CreateGroupRequest
+			body := fmt.Sprintf(`{"name":"g","platform":%q}`, platform)
+			require.Error(t, bindGroupPlatformJSON(t, &req, body),
+				"platform %q 应被 CreateGroupRequest 拒绝", platform)
+		})
+		t.Run("update_"+platform, func(t *testing.T) {
+			var req UpdateGroupRequest
+			body := fmt.Sprintf(`{"platform":%q}`, platform)
+			require.Error(t, bindGroupPlatformJSON(t, &req, body),
+				"platform %q 应被 UpdateGroupRequest 拒绝", platform)
+		})
+	}
+}
+
+// 守住 composite 路由目标不放行 CN:CN 平台不可作为 composite 路由目标
+// （DetectModelPlatform/isConcreteRequestPlatform 均无 CN 分支,放行即打开半实现路径）。
+func TestCompositeRouteTargetPlatform_StillExcludesCNProviders(t *testing.T) {
+	for _, platform := range []string{"kimi", "zhipu", "deepseek"} {
+		var req CompositeRouteRequest
+		body := fmt.Sprintf(`{"public_model":"m","target_platform":%q}`, platform)
+		require.Error(t, bindGroupPlatformJSON(t, &req, body),
+			"composite target_platform %q 应保持被拒", platform)
+	}
+}
diff --git a/frontend/src/components/common/GroupBadge.vue b/frontend/src/components/common/GroupBadge.vue
index ee4b4b91540f..8bb615781b46 100644
--- a/frontend/src/components/common/GroupBadge.vue
+++ b/frontend/src/components/common/GroupBadge.vue
@@ -162,6 +162,15 @@ const labelClass = computed(() => {
   if (props.platform === 'grok') {
     return `${base} bg-zinc-300/70 text-zinc-800 dark:bg-zinc-700/60 dark:text-zinc-200`
   }
+  if (props.platform === 'kimi') {
+    return `${base} bg-pink-200/60 text-pink-800 dark:bg-pink-800/40 dark:text-pink-300`
+  }
+  if (props.platform === 'zhipu') {
+    return `${base} bg-indigo-200/60 text-indigo-800 dark:bg-indigo-800/40 dark:text-indigo-300`
+  }
+  if (props.platform === 'deepseek') {
+    return `${base} bg-teal-200/60 text-teal-800 dark:bg-teal-800/40 dark:text-teal-300`
+  }
   if (props.platform === 'composite') {
     return `${base} bg-cyan-200/70 text-cyan-900 dark:bg-cyan-900/50 dark:text-cyan-300`
   }
@@ -200,6 +209,21 @@ const badgeClass = computed(() => {
       ? 'bg-zinc-200 text-zinc-800 dark:bg-zinc-700 dark:text-zinc-100'
       : 'bg-zinc-100 text-zinc-700 dark:bg-zinc-800 dark:text-zinc-200'
   }
+  if (props.platform === 'kimi') {
+    return isSubscription.value
+      ? 'bg-pink-100 text-pink-700 dark:bg-pink-900/30 dark:text-pink-400'
+      : 'bg-pink-50 text-pink-700 dark:bg-pink-900/20 dark:text-pink-400'
+  }
+  if (props.platform === 'zhipu') {
+    return isSubscription.value
+      ? 'bg-indigo-100 text-indigo-700 dark:bg-indigo-900/30 dark:text-indigo-400'
+      : 'bg-indigo-50 text-indigo-700 dark:bg-indigo-900/20 dark:text-indigo-400'
+  }
+  if (props.platform === 'deepseek') {
+    return isSubscription.value
+      ? 'bg-teal-100 text-teal-700 dark:bg-teal-900/30 dark:text-teal-400'
+      : 'bg-teal-50 text-teal-700 dark:bg-teal-900/20 dark:text-teal-400'
+  }
   if (props.platform === 'composite') {
     return isSubscription.value
       ? 'bg-cyan-100 text-cyan-800 dark:bg-cyan-900/30 dark:text-cyan-300'
diff --git a/frontend/src/i18n/locales/en/admin/overview.ts b/frontend/src/i18n/locales/en/admin/overview.ts
index 225865db9da5..b10dc0d7d13b 100644
--- a/frontend/src/i18n/locales/en/admin/overview.ts
+++ b/frontend/src/i18n/locales/en/admin/overview.ts
@@ -950,6 +950,9 @@ export default {
         gemini: 'Gemini',
         antigravity: 'Antigravity',
         grok: 'Grok',
+        kimi: 'Kimi',
+        zhipu: 'Zhipu GLM',
+        deepseek: 'DeepSeek',
         composite: 'Composite',
       },
       deleteConfirm:
diff --git a/frontend/src/i18n/locales/zh/admin/overview.ts b/frontend/src/i18n/locales/zh/admin/overview.ts
index 200bb6c547e5..e32cadc6419c 100644
--- a/frontend/src/i18n/locales/zh/admin/overview.ts
+++ b/frontend/src/i18n/locales/zh/admin/overview.ts
@@ -883,6 +883,9 @@ export default {
         gemini: 'Gemini',
         antigravity: 'Antigravity',
         grok: 'Grok',
+        kimi: 'Kimi',
+        zhipu: 'Zhipu GLM',
+        deepseek: 'DeepSeek',
         composite: 'Composite',
       },
       saving: '保存中...',
diff --git a/frontend/src/views/admin/GroupsView.vue b/frontend/src/views/admin/GroupsView.vue
index 904b8926d4f3..9c3c36558e0e 100644
--- a/frontend/src/views/admin/GroupsView.vue
+++ b/frontend/src/views/admin/GroupsView.vue
@@ -146,7 +146,13 @@
                       ? 'bg-purple-100 text-purple-700 dark:bg-purple-900/30 dark:text-purple-400'
                       : value === 'grok'
                         ? 'bg-zinc-200 text-zinc-800 dark:bg-zinc-700 dark:text-zinc-100'
-                        : 'bg-blue-100 text-blue-700 dark:bg-blue-900/30 dark:text-blue-400',
+                        : value === 'kimi'
+                          ? 'bg-pink-100 text-pink-700 dark:bg-pink-900/30 dark:text-pink-400'
+                          : value === 'zhipu'
+                            ? 'bg-indigo-100 text-indigo-700 dark:bg-indigo-900/30 dark:text-indigo-400'
+                            : value === 'deepseek'
+                              ? 'bg-teal-100 text-teal-700 dark:bg-teal-900/30 dark:text-teal-400'
+                              : 'bg-blue-100 text-blue-700 dark:bg-blue-900/30 dark:text-blue-400',
               ]"
             >
               <PlatformIcon :platform="value" size="xs" />
@@ -3965,7 +3971,13 @@
                           ? 'bg-purple-100 text-purple-700 dark:bg-purple-900/30 dark:text-purple-400'
                           : group.platform === 'grok'
                             ? 'bg-zinc-200 text-zinc-800 dark:bg-zinc-700 dark:text-zinc-100'
-                            : 'bg-blue-100 text-blue-700 dark:bg-blue-900/30 dark:text-blue-400',
+                            : group.platform === 'kimi'
+                              ? 'bg-pink-100 text-pink-700 dark:bg-pink-900/30 dark:text-pink-400'
+                              : group.platform === 'zhipu'
+                                ? 'bg-indigo-100 text-indigo-700 dark:bg-indigo-900/30 dark:text-indigo-400'
+                                : group.platform === 'deepseek'
+                                  ? 'bg-teal-100 text-teal-700 dark:bg-teal-900/30 dark:text-teal-400'
+                                  : 'bg-blue-100 text-blue-700 dark:bg-blue-900/30 dark:text-blue-400',
                   ]"
                 >
                   {{ t("admin.groups.platforms." + group.platform) }}
@@ -4727,6 +4739,9 @@ const platformOptions = computed(() => [
   { value: "gemini", label: "Gemini" },
   { value: "antigravity", label: "Antigravity" },
   { value: "grok", label: "Grok" },
+  { value: "kimi", label: "Kimi" },
+  { value: "zhipu", label: "Zhipu GLM" },
+  { value: "deepseek", label: "DeepSeek" },
   { value: "composite", label: "Composite" },
 ]);
 
@@ -4737,6 +4752,9 @@ const platformFilterOptions = computed(() => [
   { value: "gemini", label: "Gemini" },
   { value: "antigravity", label: "Antigravity" },
   { value: "grok", label: "Grok" },
+  { value: "kimi", label: "Kimi" },
+  { value: "zhipu", label: "Zhipu GLM" },
+  { value: "deepseek", label: "DeepSeek" },
   { value: "composite", label: "Composite" },
 ]);
 

From 10c8b70203feac8fbd744d386af6600aa87c3837 Mon Sep 17 00:00:00 2001
From: shaw <dev@sub2api.org>
Date: Mon, 17 Aug 2026 17:28:53 +0800
Subject: [PATCH 39/68] =?UTF-8?q?fix(cn-providers):=20=E4=BF=AE=E5=A4=8D?=
 =?UTF-8?q?=20CN=20=E5=88=86=E7=BB=84=E4=BA=94=E9=A1=B9=E5=8A=9F=E8=83=BD?=
 =?UTF-8?q?=E7=BC=BA=E9=99=B7(=E8=B0=83=E5=BA=A6=E9=97=B8=E9=97=A8/?=
 =?UTF-8?q?=E8=AE=A1=E8=B4=B9/=E6=96=AD=E5=BC=80=E6=BC=8F=E8=AE=B0/count?=
 =?UTF-8?q?=5Ftokens/403)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

对已合并 PR #5666 + 分组入口放行后的全量功能审计发现的 P0/P1 修复,
全部先经代码与厂商文档实证再实施:

1. /v1/messages 调度闸门(P0):sanitizeGroupMessagesDispatchFields 对非
   openai 平台恒置 AllowMessagesDispatch=false,而闸门豁免名单只有 grok,
   CN 分组经正常途径创建后恒 403——原生 Anthropic 直通(Claude Code 主用例)
   完全不可达。修复:闸门对 CN 与 grok 同语义豁免;count_tokens 处的内联
   裸检查统一走同一 helper;ResolveMessagesDispatchModel 对 CN 早退,避免
   openai 专属的 gpt-5.x 默认映射发给 CN 上游。

2. 计费候选链(P0):候选链兜底含客户端原始模型名,配合 getFallbackPricing
   的 claude→Sonnet 统一兜底,映射的 CN 模型无价时 CN 流量会按 Claude 原价
   (数倍~数十倍)静默误计,且 usage 日志显示 claude-* 名无从察觉。修复:
   CN 账号的 claude-* 候选仅在显式分组/渠道定价时放行;候选全滤空时按
   ErrModelPricingUnavailable 走零成本+告警落账(顺带修复原空候选错误会
   丢弃整条 usage 记录的次生问题)。

3. 断开/中断漏记(P0,#5148 对齐,惠及 openai 平台):messages/responses/
   chat_completions 三个 handler 的错误路径此前在 err!=nil 时丢弃携带的
   部分 result——客户端断开排水后的完整 usage 被丢,payg 上游照常计费而
   平台漏记(anthropic 网关早有同修复,openai 网关缺失)。修复:错误路径
   result 非空时照常提交 usage;failover 错误恒 result=nil 无重复计费。
   Responses×anthropic 流式转换器同时改为断开后继续排水至流自然结束
   (末尾 message_delta 的 output_tokens 不再丢),finalize 帧补工具名反转
   与客户端工具还原、仅在客户端仍连接时写出。

4. count_tokens(P1,证据修正):经实证三家 Anthropic 兼容层均无
   /v1/messages/count_tokens(DeepSeek 官方文档无此端点且注明
   anthropic-version 被忽略;OpenModel 标注该端点 Anthropic only),
   anthropic 协议转发上游=常态 404,且错误处置缺模型上下文会把不计费的
   探测放大成整账号停调。修复:CN 全协议一律本地 tiktoken 估算(与 Grok
   同方案),删除上游转发死代码。

5. 403 处置(P1):CN 此前落入通用 handleAuthError,单次 HTML 403(CDN/
   代理拦截页)即永久禁用,且 403 在 failover 集里会逐账号重放连环禁用
   整组。修复:CN 与 openai 同口径——HTML 豁免 + 3 次累计 + 临时冷却。

新增回归测试 8 项:闸门豁免(含 openai 仍受控断言)、CN 调度映射空返回、
候选过滤三态、空候选零成本落账、断开排水 usage 完整性、HTML-403 零处罚、
结构化 403 首次临时停调。handler/service 全包测试通过。
---
 .../handler/openai_chat_completions.go        |  80 ++++----
 .../openai_gateway_cn_dispatch_test.go        |  29 +++
 .../handler/openai_gateway_count_tokens.go    |   2 +-
 .../handler/openai_gateway_handler.go         | 180 ++++++++++--------
 .../service/openai_gateway_cn_fixes_test.go   | 144 ++++++++++++++
 .../service/openai_gateway_count_tokens.go    |  16 +-
 ...penai_gateway_messages_anthropic_native.go |  89 ---------
 ...enai_gateway_responses_anthropic_native.go |  46 +++--
 .../internal/service/openai_gateway_usage.go  |  40 +++-
 .../service/openai_messages_dispatch.go       |   7 +
 backend/internal/service/ratelimit_service.go |   5 +-
 11 files changed, 411 insertions(+), 227 deletions(-)
 create mode 100644 backend/internal/handler/openai_gateway_cn_dispatch_test.go
 create mode 100644 backend/internal/service/openai_gateway_cn_fixes_test.go

diff --git a/backend/internal/handler/openai_chat_completions.go b/backend/internal/handler/openai_chat_completions.go
index 42dc13112a79..79709cabb7b7 100644
--- a/backend/internal/handler/openai_chat_completions.go
+++ b/backend/internal/handler/openai_chat_completions.go
@@ -255,6 +255,48 @@ func (h *OpenAIGatewayHandler) ChatCompletions(c *gin.Context) {
 		if err == nil && result != nil && result.FirstTokenMs != nil {
 			service.SetOpsLatencyMs(c, service.OpsTimeToFirstTokenMsKey, int64(*result.FirstTokenMs))
 		}
+		// #5148 对齐：错误返回携带的部分 result（流中断前上游已计量的 usage）照常
+		// 入账；failover 错误恒定 result=nil，不会重复计费。
+		submitChatUsage := func(res *service.OpenAIForwardResult) {
+			if res == nil {
+				return
+			}
+			userAgent := c.GetHeader("User-Agent")
+			clientIP := ip.GetClientIP(c)
+			inboundEndpoint := GetInboundEndpoint(c)
+			upstreamEndpoint := resolveOpenAIUpstreamEndpoint(c, account, res)
+			quotaPlatform := service.QuotaPlatform(c.Request.Context(), apiKey)
+			sessionID := service.ExtractClientSessionID(c)
+			cyberBlocked := service.GetOpsCyberPolicy(c) != nil
+			h.submitOpenAIUsageRecordTask(c.Request.Context(), res, func(ctx context.Context) {
+				if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
+					Result:             res,
+					APIKey:             apiKey,
+					User:               apiKey.User,
+					Account:            account,
+					Subscription:       subscription,
+					InboundEndpoint:    inboundEndpoint,
+					UpstreamEndpoint:   upstreamEndpoint,
+					UserAgent:          userAgent,
+					IPAddress:          clientIP,
+					APIKeyService:      h.apiKeyService,
+					QuotaPlatform:      quotaPlatform,
+					SessionID:          sessionID,
+					ChannelUsageFields: clientRequestedUsageFields(c, channelMapping, reqModel, res.UpstreamModel),
+					PricingAt:          pricingAt,
+					CyberBlocked:       cyberBlocked,
+				}); err != nil {
+					logger.L().With(
+						zap.String("component", "handler.openai_gateway.chat_completions"),
+						zap.Int64("user_id", subject.UserID),
+						zap.Int64("api_key_id", apiKey.ID),
+						zap.Any("group_id", apiKey.GroupID),
+						zap.String("model", reqModel),
+						zap.Int64("account_id", account.ID),
+					).Error("openai_chat_completions.record_usage_failed", zap.Error(err))
+				}
+			})
+		}
 		if err != nil {
 			if result != nil && result.ImageCount > 0 {
 				reqLog.Warn("openai_chat_completions.forward_partial_error_with_image_result",
@@ -339,6 +381,7 @@ func (h *OpenAIGatewayHandler) ChatCompletions(c *gin.Context) {
 					zap.Bool("upstream_error_response_already_written", upstreamErrorAlreadyCommunicated),
 					zap.Error(err),
 				)
+				submitChatUsage(result)
 				return
 			}
 		}
@@ -348,42 +391,7 @@ func (h *OpenAIGatewayHandler) ChatCompletions(c *gin.Context) {
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, account.GetMappedModel(reqModel), true, nil)
 		}
 
-		userAgent := c.GetHeader("User-Agent")
-		clientIP := ip.GetClientIP(c)
-		inboundEndpoint := GetInboundEndpoint(c)
-		upstreamEndpoint := resolveOpenAIUpstreamEndpoint(c, account, result)
-		quotaPlatform := service.QuotaPlatform(c.Request.Context(), apiKey)
-		sessionID := service.ExtractClientSessionID(c)
-
-		cyberBlocked := service.GetOpsCyberPolicy(c) != nil
-		h.submitOpenAIUsageRecordTask(c.Request.Context(), result, func(ctx context.Context) {
-			if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
-				Result:             result,
-				APIKey:             apiKey,
-				User:               apiKey.User,
-				Account:            account,
-				Subscription:       subscription,
-				InboundEndpoint:    inboundEndpoint,
-				UpstreamEndpoint:   upstreamEndpoint,
-				UserAgent:          userAgent,
-				IPAddress:          clientIP,
-				APIKeyService:      h.apiKeyService,
-				QuotaPlatform:      quotaPlatform,
-				SessionID:          sessionID,
-				ChannelUsageFields: clientRequestedUsageFields(c, channelMapping, reqModel, result.UpstreamModel),
-				PricingAt:          pricingAt,
-				CyberBlocked:       cyberBlocked,
-			}); err != nil {
-				logger.L().With(
-					zap.String("component", "handler.openai_gateway.chat_completions"),
-					zap.Int64("user_id", subject.UserID),
-					zap.Int64("api_key_id", apiKey.ID),
-					zap.Any("group_id", apiKey.GroupID),
-					zap.String("model", reqModel),
-					zap.Int64("account_id", account.ID),
-				).Error("openai_chat_completions.record_usage_failed", zap.Error(err))
-			}
-		})
+		submitChatUsage(result)
 		reqLog.Debug("openai_chat_completions.request_completed",
 			zap.Int64("account_id", account.ID),
 			zap.Int("switch_count", switchCount),
diff --git a/backend/internal/handler/openai_gateway_cn_dispatch_test.go b/backend/internal/handler/openai_gateway_cn_dispatch_test.go
new file mode 100644
index 000000000000..b7f11c5462e2
--- /dev/null
+++ b/backend/internal/handler/openai_gateway_cn_dispatch_test.go
@@ -0,0 +1,29 @@
+package handler
+
+// CN 分组 /v1/messages 调度闸门回归（修复:正常途径创建的 CN 分组曾恒 403）：
+// sanitizeGroupMessagesDispatchFields 对非 openai 平台强制 AllowMessagesDispatch
+// =false，故 CN 分组必须与 grok 一样在闸门处豁免，否则原生 Anthropic 直通
+//（Claude Code 主用例）永远不可达。
+
+import (
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/stretchr/testify/require"
+)
+
+func TestAllowOpenAICompatibleMessagesDispatch_CNProvidersExempt(t *testing.T) {
+	require.True(t, allowOpenAICompatibleMessagesDispatch(nil), "无 key 保持放行")
+
+	for _, platform := range []string{service.PlatformKimi, service.PlatformZhipu, service.PlatformDeepseek, service.PlatformGrok} {
+		apiKey := &service.APIKey{Group: &service.Group{Platform: platform, AllowMessagesDispatch: false}}
+		require.True(t, allowOpenAICompatibleMessagesDispatch(apiKey),
+			"%s 分组必须豁免 allow_messages_dispatch 闸门", platform)
+	}
+
+	// 非回归：openai 分组仍受开关控制。
+	openaiOff := &service.APIKey{Group: &service.Group{Platform: service.PlatformOpenAI, AllowMessagesDispatch: false}}
+	require.False(t, allowOpenAICompatibleMessagesDispatch(openaiOff))
+	openaiOn := &service.APIKey{Group: &service.Group{Platform: service.PlatformOpenAI, AllowMessagesDispatch: true}}
+	require.True(t, allowOpenAICompatibleMessagesDispatch(openaiOn))
+}
diff --git a/backend/internal/handler/openai_gateway_count_tokens.go b/backend/internal/handler/openai_gateway_count_tokens.go
index ac6deda4aa0f..70e61e24f1ba 100644
--- a/backend/internal/handler/openai_gateway_count_tokens.go
+++ b/backend/internal/handler/openai_gateway_count_tokens.go
@@ -78,7 +78,7 @@ func (h *OpenAIGatewayHandler) CountTokens(c *gin.Context) {
 		zap.Any("group_id", apiKey.GroupID),
 	)
 
-	if apiKey.Group != nil && !apiKey.Group.AllowMessagesDispatch {
+	if !allowOpenAICompatibleMessagesDispatch(apiKey) {
 		h.anthropicErrorResponse(c, http.StatusForbidden, "permission_error",
 			"This group does not allow /v1/messages dispatch")
 		return
diff --git a/backend/internal/handler/openai_gateway_handler.go b/backend/internal/handler/openai_gateway_handler.go
index 28515d98f0d7..43893958004c 100644
--- a/backend/internal/handler/openai_gateway_handler.go
+++ b/backend/internal/handler/openai_gateway_handler.go
@@ -197,6 +197,13 @@ func allowOpenAICompatibleMessagesDispatch(apiKey *service.APIKey) bool {
 	if apiKey.Group.Platform == service.PlatformGrok {
 		return true
 	}
+	// 国产供应商分组与 grok 同语义:/v1/messages 就是其主要服务形态(anthropic
+	// 协议账号原生直通 Claude Code),无需 allow_messages_dispatch 开关授权——
+	// 该开关对非 openai 平台恒被 sanitizeGroupMessagesDispatchFields 置 false,
+	// 若不豁免,CN 分组将永远 403。
+	if service.IsCNProvider(apiKey.Group.Platform) {
+		return true
+	}
 	return apiKey.Group.AllowMessagesDispatch
 }
 
@@ -597,6 +604,50 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 		if err == nil && result != nil && result.FirstTokenMs != nil {
 			service.SetOpsLatencyMs(c, service.OpsTimeToFirstTokenMsKey, int64(*result.FirstTokenMs))
 		}
+		// #5148 对齐：错误返回携带的部分 result（流中断前上游已计量的 usage）照常
+		// 入账；failover 错误恒定 result=nil，不会重复计费。
+		submitResponsesUsage := func(res *service.OpenAIForwardResult) {
+			if res == nil {
+				return
+			}
+			userAgent := c.GetHeader("User-Agent")
+			clientIP := ip.GetClientIP(c)
+			requestPayloadHash := service.HashUsageRequestPayload(body)
+			inboundEndpoint := GetInboundEndpoint(c)
+			upstreamEndpoint := resolveOpenAIUpstreamEndpoint(c, account, res)
+			quotaPlatform := service.QuotaPlatform(c.Request.Context(), apiKey)
+			sessionID := service.ExtractClientSessionID(c)
+			cyberBlocked := service.GetOpsCyberPolicy(c) != nil
+			h.submitOpenAIUsageRecordTask(c.Request.Context(), res, func(ctx context.Context) {
+				if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
+					Result:             res,
+					APIKey:             apiKey,
+					User:               apiKey.User,
+					Account:            account,
+					Subscription:       subscription,
+					InboundEndpoint:    inboundEndpoint,
+					UpstreamEndpoint:   upstreamEndpoint,
+					UserAgent:          userAgent,
+					IPAddress:          clientIP,
+					RequestPayloadHash: requestPayloadHash,
+					APIKeyService:      h.apiKeyService,
+					QuotaPlatform:      quotaPlatform,
+					SessionID:          sessionID,
+					ChannelUsageFields: clientRequestedUsageFields(c, channelMapping, reqModel, res.UpstreamModel),
+					PricingAt:          pricingAt,
+					CyberBlocked:       cyberBlocked,
+				}); err != nil {
+					logger.L().With(
+						zap.String("component", "handler.openai_gateway.responses"),
+						zap.Int64("user_id", subject.UserID),
+						zap.Int64("api_key_id", apiKey.ID),
+						zap.Any("group_id", apiKey.GroupID),
+						zap.String("model", reqModel),
+						zap.Int64("account_id", account.ID),
+					).Error("openai.record_usage_failed", zap.Error(err))
+				}
+			})
+		}
 		if err != nil {
 			if result != nil && result.ImageCount > 0 {
 				reqLog.Warn("openai.forward_partial_error_with_image_result",
@@ -696,6 +747,7 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 					zap.Bool("upstream_error_response_already_written", upstreamErrorAlreadyCommunicated),
 					zap.Error(err),
 				}
+				submitResponsesUsage(result)
 				if shouldLogOpenAIForwardFailureAsWarn(c, wroteFallback) {
 					reqLog.Warn("openai.forward_failed", fields...)
 					return
@@ -714,46 +766,8 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, account.GetMappedModel(reqModel), openAIForwardSucceededForScheduling(result), nil)
 		}
 
-		// 捕获请求信息（用于异步记录，避免在 goroutine 中访问 gin.Context）
-		userAgent := c.GetHeader("User-Agent")
-		clientIP := ip.GetClientIP(c)
-		requestPayloadHash := service.HashUsageRequestPayload(body)
-		inboundEndpoint := GetInboundEndpoint(c)
-		upstreamEndpoint := resolveOpenAIUpstreamEndpoint(c, account, result)
-		quotaPlatform := service.QuotaPlatform(c.Request.Context(), apiKey)
-		sessionID := service.ExtractClientSessionID(c)
-
 		// 使用量记录通过有界 worker 池提交，避免请求热路径创建无界 goroutine。
-		cyberBlocked := service.GetOpsCyberPolicy(c) != nil
-		h.submitOpenAIUsageRecordTask(c.Request.Context(), result, func(ctx context.Context) {
-			if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
-				Result:             result,
-				APIKey:             apiKey,
-				User:               apiKey.User,
-				Account:            account,
-				Subscription:       subscription,
-				InboundEndpoint:    inboundEndpoint,
-				UpstreamEndpoint:   upstreamEndpoint,
-				UserAgent:          userAgent,
-				IPAddress:          clientIP,
-				RequestPayloadHash: requestPayloadHash,
-				APIKeyService:      h.apiKeyService,
-				QuotaPlatform:      quotaPlatform,
-				SessionID:          sessionID,
-				ChannelUsageFields: clientRequestedUsageFields(c, channelMapping, reqModel, result.UpstreamModel),
-				PricingAt:          pricingAt,
-				CyberBlocked:       cyberBlocked,
-			}); err != nil {
-				logger.L().With(
-					zap.String("component", "handler.openai_gateway.responses"),
-					zap.Int64("user_id", subject.UserID),
-					zap.Int64("api_key_id", apiKey.ID),
-					zap.Any("group_id", apiKey.GroupID),
-					zap.String("model", reqModel),
-					zap.Int64("account_id", account.ID),
-				).Error("openai.record_usage_failed", zap.Error(err))
-			}
-		})
+		submitResponsesUsage(result)
 		reqLog.Debug("openai.request_completed",
 			zap.Int64("account_id", account.ID),
 			zap.Int("switch_count", switchCount),
@@ -1145,6 +1159,51 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
 		if err == nil && result != nil && result.FirstTokenMs != nil {
 			service.SetOpsLatencyMs(c, service.OpsTimeToFirstTokenMsKey, int64(*result.FirstTokenMs))
 		}
+		// Forward 与错误一起返回的部分结果：流中断/客户端断开排水前上游已计量的
+		// usage 照常入账，避免上游已产生消耗的请求完全漏记（#5148，对齐 anthropic
+		// 网关同名修复）。failover 错误恒定 result=nil，不会重复计费。
+		submitMessagesUsage := func(res *service.OpenAIForwardResult) {
+			if res == nil {
+				return
+			}
+			userAgent := c.GetHeader("User-Agent")
+			clientIP := ip.GetClientIP(c)
+			requestPayloadHash := service.HashUsageRequestPayload(body)
+			inboundEndpoint := GetInboundEndpoint(c)
+			upstreamEndpoint := resolveOpenAIUpstreamEndpoint(c, account, res)
+			quotaPlatform := service.QuotaPlatform(c.Request.Context(), apiKey)
+			sessionID := service.ExtractClientSessionID(c)
+			cyberBlocked := service.GetOpsCyberPolicy(c) != nil
+			h.submitOpenAIUsageRecordTask(c.Request.Context(), res, func(ctx context.Context) {
+				if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
+					Result:             res,
+					APIKey:             apiKey,
+					User:               apiKey.User,
+					Account:            account,
+					Subscription:       subscription,
+					InboundEndpoint:    inboundEndpoint,
+					UpstreamEndpoint:   upstreamEndpoint,
+					UserAgent:          userAgent,
+					IPAddress:          clientIP,
+					RequestPayloadHash: requestPayloadHash,
+					APIKeyService:      h.apiKeyService,
+					QuotaPlatform:      quotaPlatform,
+					SessionID:          sessionID,
+					ChannelUsageFields: clientRequestedUsageFields(c, channelMappingMsg, reqModel, res.UpstreamModel),
+					PricingAt:          pricingAt,
+					CyberBlocked:       cyberBlocked,
+				}); err != nil {
+					logger.L().With(
+						zap.String("component", "handler.openai_gateway.messages"),
+						zap.Int64("user_id", subject.UserID),
+						zap.Int64("api_key_id", apiKey.ID),
+						zap.Any("group_id", apiKey.GroupID),
+						zap.String("model", reqModel),
+						zap.Int64("account_id", account.ID),
+					).Error("openai_messages.record_usage_failed", zap.Error(err))
+				}
+			})
+		}
 		if err != nil {
 			if result != nil && result.ImageCount > 0 {
 				reqLog.Warn("openai_messages.forward_partial_error_with_image_result",
@@ -1219,6 +1278,9 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
 						zap.Int64("account_id", account.ID),
 						zap.Error(err),
 					)
+					// 断开排水期间上游已计量的 usage 必须入账（此前直接 return 丢弃，
+					// payg 上游照常计费而平台漏记）。
+					submitMessagesUsage(result)
 					return
 				}
 				h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, account.GetMappedModel(currentRoutingModel), false, nil)
@@ -1228,6 +1290,7 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
 					zap.Bool("fallback_error_response_written", wroteFallback),
 					zap.Error(err),
 				)
+				submitMessagesUsage(result)
 				return
 			}
 		}
@@ -1237,44 +1300,7 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, account.GetMappedModel(currentRoutingModel), true, nil)
 		}
 
-		userAgent := c.GetHeader("User-Agent")
-		clientIP := ip.GetClientIP(c)
-		requestPayloadHash := service.HashUsageRequestPayload(body)
-		inboundEndpoint := GetInboundEndpoint(c)
-		upstreamEndpoint := resolveOpenAIUpstreamEndpoint(c, account, result)
-		quotaPlatform := service.QuotaPlatform(c.Request.Context(), apiKey)
-		sessionID := service.ExtractClientSessionID(c)
-
-		cyberBlocked := service.GetOpsCyberPolicy(c) != nil
-		h.submitOpenAIUsageRecordTask(c.Request.Context(), result, func(ctx context.Context) {
-			if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
-				Result:             result,
-				APIKey:             apiKey,
-				User:               apiKey.User,
-				Account:            account,
-				Subscription:       subscription,
-				InboundEndpoint:    inboundEndpoint,
-				UpstreamEndpoint:   upstreamEndpoint,
-				UserAgent:          userAgent,
-				IPAddress:          clientIP,
-				RequestPayloadHash: requestPayloadHash,
-				APIKeyService:      h.apiKeyService,
-				QuotaPlatform:      quotaPlatform,
-				SessionID:          sessionID,
-				ChannelUsageFields: clientRequestedUsageFields(c, channelMappingMsg, reqModel, result.UpstreamModel),
-				PricingAt:          pricingAt,
-				CyberBlocked:       cyberBlocked,
-			}); err != nil {
-				logger.L().With(
-					zap.String("component", "handler.openai_gateway.messages"),
-					zap.Int64("user_id", subject.UserID),
-					zap.Int64("api_key_id", apiKey.ID),
-					zap.Any("group_id", apiKey.GroupID),
-					zap.String("model", reqModel),
-					zap.Int64("account_id", account.ID),
-				).Error("openai_messages.record_usage_failed", zap.Error(err))
-			}
-		})
+		submitMessagesUsage(result)
 		reqLog.Debug("openai_messages.request_completed",
 			zap.Int64("account_id", account.ID),
 			zap.Int("switch_count", switchCount),
diff --git a/backend/internal/service/openai_gateway_cn_fixes_test.go b/backend/internal/service/openai_gateway_cn_fixes_test.go
new file mode 100644
index 000000000000..ce80a1aca7fc
--- /dev/null
+++ b/backend/internal/service/openai_gateway_cn_fixes_test.go
@@ -0,0 +1,144 @@
+//go:build unit
+
+package service
+
+// 国产供应商功能修复回归测试：
+//  1. CN 分组不适用 /v1/messages 调度级模型映射（openai 的 gpt-5.x 默认值发给
+//     CN 上游必错）；
+//  2. 计费候选链对 CN 账号过滤 claude-* 兜底候选（防按 Claude 原价误计 CN 流量）；
+//  3. 空候选按 ErrModelPricingUnavailable 处理（零成本落账而非丢弃 usage 记录）；
+//  4. Responses×anthropic 流式转换器客户端断开后继续排水、usage 汇总完整。
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+func TestResolveMessagesDispatchModel_CNProvidersNoDispatchMapping(t *testing.T) {
+	for _, platform := range []string{PlatformKimi, PlatformZhipu, PlatformDeepseek} {
+		g := &Group{Platform: platform}
+		require.Empty(t, g.ResolveMessagesDispatchModel("claude-sonnet-4-5"),
+			"CN 分组(%s)不得返回调度级映射模型（openai 默认值会发给 CN 上游）", platform)
+		require.Empty(t, g.ResolveMessagesDispatchModel("claude-opus-4-1"), platform)
+	}
+	// 非回归：openai 分组保持原有默认映射行为。
+	openaiGroup := &Group{Platform: PlatformOpenAI}
+	require.NotEmpty(t, openaiGroup.ResolveMessagesDispatchModel("claude-sonnet-4-5"),
+		"openai 分组的调度默认映射不应受 CN 修复影响")
+}
+
+func TestFilterCNProviderBillingModelCandidates(t *testing.T) {
+	svc := &OpenAIGatewayService{} // resolver 为 nil → 无显式分组/渠道定价
+	apiKey := &APIKey{Group: &Group{ID: 1, Platform: PlatformKimi}}
+
+	cnAccount := &Account{ID: 1, Platform: PlatformKimi}
+	filtered := svc.filterCNProviderBillingModelCandidates(context.Background(), cnAccount, apiKey,
+		[]string{"kimi-k2-0905-preview", "claude-sonnet-4-5", "moonshot-v1-8k"})
+	require.Equal(t, []string{"kimi-k2-0905-preview", "moonshot-v1-8k"}, filtered,
+		"无显式定价时 claude-* 候选必须被过滤")
+
+	allClaude := svc.filterCNProviderBillingModelCandidates(context.Background(), cnAccount, apiKey,
+		[]string{"claude-sonnet-4-5", "claude-sonnet-4-5"})
+	require.Empty(t, allClaude, "全 claude 候选应被清空（上层走零成本+告警落账）")
+
+	// 非 CN 账号完全不受影响。
+	openaiAccount := &Account{ID: 2, Platform: PlatformOpenAI}
+	passthrough := svc.filterCNProviderBillingModelCandidates(context.Background(), openaiAccount, apiKey,
+		[]string{"claude-sonnet-4-5", "gpt-5.4"})
+	require.Equal(t, []string{"claude-sonnet-4-5", "gpt-5.4"}, passthrough)
+
+	require.Nil(t, svc.filterCNProviderBillingModelCandidates(context.Background(), nil, apiKey, nil))
+}
+
+func TestCalculateOpenAIRecordUsageCost_EmptyCandidatesIsPricingUnavailable(t *testing.T) {
+	svc := &OpenAIGatewayService{}
+	apiKey := &APIKey{Group: &Group{ID: 1, Platform: PlatformKimi}}
+
+	_, err := svc.calculateOpenAIRecordUsageCost(
+		context.Background(), nil, apiKey, nil,
+		1.0, 1.0, 1.0, 1.0, UsageTokens{InputTokens: 100}, "", nil,
+	)
+	require.Error(t, err)
+	require.True(t, isUsagePricingUnavailableError(err),
+		"空候选必须按无价可循处理（上层零成本落账），而不是丢弃整条 usage 记录: %v", err)
+}
+
+func TestResponsesStreamingFromNativeAnthropic_ClientDisconnectDrainsUsage(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newNativeAnthropicHangTestService(5)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+	// failAfter=0：首次写出即失败，模拟客户端断开（复用测试包既有 failingGinWriter）。
+	failWriter := &failingGinWriter{ResponseWriter: c.Writer, failAfter: 0}
+	c.Writer = failWriter
+
+	resp, pr, pw := newHangingUpstreamResponse()
+	go func() {
+		// 首事件触发客户端写失败后，末尾 message_delta 才携带最终 output_tokens：
+		// 断开即弃会把整段生成记成 1 token。
+		_, _ = pw.Write([]byte(miniAnthropicSSEStream()))
+		_ = pw.Close()
+	}()
+	defer func() { _ = pr.Close() }()
+
+	res, err := svc.handleResponsesStreamingFromNativeAnthropic(
+		resp, c, "glm-4.7", "glm-4.7", "glm-4.7", nil, time.Now(), apicompat.ResponsesClientToolMapping{})
+
+	require.NoError(t, err, "断开排水至上游自然结束应返回 nil error（usage 走成功路径落账）")
+	require.NotNil(t, res)
+	require.True(t, res.ClientDisconnect)
+	require.Equal(t, 10, res.Usage.InputTokens, "input_tokens 应来自 message_start")
+	require.Equal(t, 5, res.Usage.OutputTokens,
+		"output_tokens 必须来自排水读到的末尾 message_delta（断开即弃时会是 1）")
+}
+
+func TestHandle403_CNProviderHTMLBodySkipsAccountPenalty(t *testing.T) {
+	for _, platform := range []string{PlatformKimi, PlatformZhipu, PlatformDeepseek} {
+		repo := &rateLimitAccountRepoStub{}
+		service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+		account := &Account{ID: 401, Platform: platform, Type: AccountTypeAPIKey}
+
+		shouldDisable := service.HandleUpstreamError(
+			context.Background(),
+			account,
+			http.StatusForbidden,
+			http.Header{},
+			[]byte("<html><body>Access denied by CDN</body></html>"),
+		)
+
+		require.False(t, shouldDisable, "%s: HTML 403（CDN/代理拦截页）不得作为账号失效证据", platform)
+		require.Equal(t, 0, repo.setErrorCalls, "%s: 不得永久禁用账号", platform)
+		require.Equal(t, 0, repo.tempCalls, "%s: 不得临时停调账号", platform)
+	}
+}
+
+func TestHandle403_CNProviderStructured403TempUnschedulableFirstHit(t *testing.T) {
+	repo := &rateLimitAccountRepoStub{}
+	counter := &openAI403CounterCacheStub{counts: []int64{1}}
+	service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	service.SetOpenAI403CounterCache(counter)
+	account := &Account{ID: 402, Platform: PlatformKimi, Type: AccountTypeAPIKey}
+
+	shouldDisable := service.HandleUpstreamError(
+		context.Background(),
+		account,
+		http.StatusForbidden,
+		http.Header{},
+		[]byte(`{"error":{"message":"forbidden"}}`),
+	)
+
+	require.True(t, shouldDisable)
+	require.Equal(t, 0, repo.setErrorCalls, "首次结构化 403 应临时停调而非永久禁用")
+	require.Equal(t, 1, repo.tempCalls)
+	require.Contains(t, repo.lastTempReason, "(1/3)")
+}
diff --git a/backend/internal/service/openai_gateway_count_tokens.go b/backend/internal/service/openai_gateway_count_tokens.go
index f20b45b5269c..bcdaa457b824 100644
--- a/backend/internal/service/openai_gateway_count_tokens.go
+++ b/backend/internal/service/openai_gateway_count_tokens.go
@@ -92,15 +92,13 @@ func (s *OpenAIGatewayService) ForwardCountTokensAsAnthropic(
 		return fmt.Errorf("count_tokens: missing account")
 	}
 
-	// 国产供应商 Anthropic 协议：上游有原生 /v1/messages/count_tokens 端点，
-	// 直接透传（仅模型名映射），不走 /v1/responses/input_tokens 估算。
-	if account.IsAnthropicProtocol() {
-		return s.forwardCountTokensViaNativeAnthropic(ctx, c, account, body, defaultMappedModel)
-	}
-
-	// 国产供应商其余协议（chat_completions / responses）：三家上游均无
-	// OpenAI 兼容的 /v1/responses/input_tokens 端点，与 Grok 一样本地估算，
-	// 不发上游请求（Claude Code 客户端会高频调用 count_tokens）。
+	// 国产供应商（全部协议，含 anthropic）：一律本地估算，不发上游请求。
+	// 依据（2026-08 核实）：三家的 Anthropic 兼容层均未提供
+	// /v1/messages/count_tokens——DeepSeek 官方 anthropic_api 文档无此端点
+	// （且注明 anthropic-version 头被忽略），聚合网关 OpenModel 明确标注
+	// count_tokens 为 "Anthropic only"，Kimi/智谱亦无任何文档承诺。转发上游
+	// 只会常态 404，且错误还会流入账号处置逻辑误伤整账号调度；Claude Code
+	// 高频调用此端点，本地 tiktoken 估算是与 Grok 一致的既有方案。
 	if account.IsCNProvider() {
 		estimated, err := estimateAnthropicCountTokensLocally(body)
 		if err != nil {
diff --git a/backend/internal/service/openai_gateway_messages_anthropic_native.go b/backend/internal/service/openai_gateway_messages_anthropic_native.go
index e785f3755fda..497cfe09ab10 100644
--- a/backend/internal/service/openai_gateway_messages_anthropic_native.go
+++ b/backend/internal/service/openai_gateway_messages_anthropic_native.go
@@ -515,95 +515,6 @@ func (s *OpenAIGatewayService) nativeAnthropicStreamResult(
 	}
 }
 
-// forwardCountTokensViaNativeAnthropic 把 Anthropic count_tokens 请求透传到
-// 国产供应商原生 Anthropic 端点（{base}/v1/messages/count_tokens），
-// 仅做模型名映射，不做协议转换。
-func (s *OpenAIGatewayService) forwardCountTokensViaNativeAnthropic(
-	ctx context.Context,
-	c *gin.Context,
-	account *Account,
-	body []byte,
-	defaultMappedModel string,
-) error {
-	originalModel := strings.TrimSpace(gjson.GetBytes(body, "model").String())
-	if originalModel == "" {
-		writeAnthropicCountTokensError(c, http.StatusBadRequest, "invalid_request_error", "model is required")
-		return fmt.Errorf("count_tokens: missing model in request")
-	}
-	billingModel := resolveOpenAIForwardModel(account, originalModel, strings.TrimSpace(defaultMappedModel))
-	upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
-	if upstreamModel != originalModel {
-		rewritten, err := sjson.SetBytes(body, "model", upstreamModel)
-		if err != nil {
-			return fmt.Errorf("count_tokens: rewrite model: %w", err)
-		}
-		body = rewritten
-	}
-
-	apiKey := strings.TrimSpace(account.GetOpenAIProtocolAPIKey())
-	if apiKey == "" {
-		writeAnthropicCountTokensError(c, http.StatusBadGateway, "upstream_error", "Account api_key is missing")
-		return fmt.Errorf("count_tokens: account %d missing api_key", account.ID)
-	}
-	targetURL, err := s.nativeAnthropicTargetURL(account)
-	if err != nil {
-		return fmt.Errorf("count_tokens: %w", err)
-	}
-	targetURL = strings.TrimSuffix(targetURL, "/v1/messages") + "/v1/messages/count_tokens"
-
-	upstreamReq, err := http.NewRequestWithContext(ctx, http.MethodPost, targetURL, bytes.NewReader(body))
-	if err != nil {
-		writeAnthropicCountTokensError(c, http.StatusInternalServerError, "api_error", "Failed to build request")
-		return fmt.Errorf("count_tokens: build request: %w", err)
-	}
-	reqHeader := upstreamReq.Header
-	reqHeader.Del("authorization")
-	reqHeader.Del("x-api-key")
-	setAnthropicAPIKeyAuthHeader(reqHeader, account, apiKey)
-	reqHeader.Set("content-type", "application/json")
-	reqHeader.Set("accept", "application/json")
-	account.ApplyHeaderOverrides(reqHeader)
-
-	proxyURL := ""
-	if account.Proxy != nil {
-		proxyURL = account.Proxy.URL()
-	}
-	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
-	if err != nil {
-		safeErr := sanitizeUpstreamErrorMessage(err.Error())
-		setOpsUpstreamError(c, 0, safeErr, "")
-		writeAnthropicCountTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream request failed")
-		return fmt.Errorf("count_tokens: upstream request failed: %s", safeErr)
-	}
-	defer func() { _ = resp.Body.Close() }()
-
-	// count_tokens 响应体极小；与其他探测路径一致加 256KB 上限防异常上游放大内存。
-	respBody, err := io.ReadAll(io.LimitReader(resp.Body, cnQuotaMaxBodyBytes))
-	if err != nil {
-		writeAnthropicCountTokensError(c, http.StatusBadGateway, "upstream_error", "Failed to read response")
-		return fmt.Errorf("count_tokens: read response: %w", err)
-	}
-	if resp.StatusCode >= 400 {
-		if s.rateLimitService != nil {
-			s.rateLimitService.HandleUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)
-		}
-		upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(respBody)))
-		setOpsUpstreamError(c, resp.StatusCode, upstreamMsg, "")
-		writeAnthropicCountTokensError(c, resp.StatusCode, "upstream_error", "Upstream request failed")
-		return fmt.Errorf("count_tokens: upstream error: %d", resp.StatusCode)
-	}
-
-	inputTokens := gjson.GetBytes(respBody, "input_tokens")
-	if !inputTokens.Exists() {
-		writeAnthropicCountTokensError(c, http.StatusBadGateway, "upstream_error", "Upstream response missing input_tokens")
-		return fmt.Errorf("count_tokens: response missing input_tokens field")
-	}
-	c.JSON(http.StatusOK, gin.H{
-		"input_tokens": int(inputTokens.Int()),
-	})
-	return nil
-}
-
 // claudeUsageToOpenAIUsage 把 Anthropic 格式 usage 映射到 OpenAI 网关统一的
 // 用量结构（字段一一对应）。
 func claudeUsageToOpenAIUsage(u *ClaudeUsage) OpenAIUsage {
diff --git a/backend/internal/service/openai_gateway_responses_anthropic_native.go b/backend/internal/service/openai_gateway_responses_anthropic_native.go
index b72067fc4d77..04cd7fc27484 100644
--- a/backend/internal/service/openai_gateway_responses_anthropic_native.go
+++ b/backend/internal/service/openai_gateway_responses_anthropic_native.go
@@ -377,7 +377,11 @@ func (s *OpenAIGatewayService) handleResponsesStreamingFromNativeAnthropic(
 		return resultWithUsage(), fmt.Errorf("stream data interval timeout")
 	}
 
-	processAnthropicEvent := func(event *apicompat.AnthropicStreamEvent) bool {
+	// 与 CC 姊妹路径（handleCCStreamingFromNativeAnthropic.writeChunk）同语义：
+	// 客户端断开后不再写出，但继续排水上游至流自然结束——Anthropic 的最终
+	// output_tokens 只在末尾 message_delta 携带，提前退出会把整段生成记成 ~1
+	// token，payg 上游照常计费而平台漏记。状态机照常推进以保证 finalize 一致。
+	processAnthropicEvent := func(event *apicompat.AnthropicStreamEvent) {
 		if firstChunk {
 			firstChunk = false
 			ms := int(time.Since(startTime).Milliseconds())
@@ -392,6 +396,9 @@ func (s *OpenAIGatewayService) handleResponsesStreamingFromNativeAnthropic(
 		}
 
 		events := apicompat.AnthropicEventToResponsesEvents(event, state)
+		if clientDisconnected {
+			return
+		}
 		for _, evt := range events {
 			payload, err := json.Marshal(evt)
 			if err != nil {
@@ -406,14 +413,13 @@ func (s *OpenAIGatewayService) handleResponsesStreamingFromNativeAnthropic(
 				eventType := gjson.GetBytes(restored, "type").String()
 				if _, err := fmt.Fprintf(c.Writer, "event: %s\ndata: %s\n\n", eventType, restored); err != nil {
 					clientDisconnected = true
-					return true
+					return
 				}
 			}
 		}
 		if len(events) > 0 {
 			c.Writer.Flush()
 		}
-		return false
 	}
 
 	for {
@@ -447,21 +453,39 @@ func (s *OpenAIGatewayService) handleResponsesStreamingFromNativeAnthropic(
 			continue
 		}
 
-		if processAnthropicEvent(&event) {
-			return resultWithUsage(), nil
-		}
+		processAnthropicEvent(&event)
 	}
 
-	// Finalize state machine（客户端已断开时仍执行，保证 usage 汇总完整）。
-	if finalEvents := apicompat.FinalizeAnthropicResponsesStream(state); len(finalEvents) > 0 {
+	// Finalize state machine（客户端已断开时仍推进，保证 usage 汇总完整；仅在
+	// 客户端仍连接时写出）。终态帧与逐事件路径一致过工具名反转与客户端工具还原，
+	// 避免流截断时终态帧携带改写后的工具名。
+	if finalEvents := apicompat.FinalizeAnthropicResponsesStream(state); len(finalEvents) > 0 && !clientDisconnected {
+		wrote := false
 		for _, evt := range finalEvents {
-			sse, err := apicompat.ResponsesEventToSSE(evt)
+			payload, err := json.Marshal(evt)
 			if err != nil {
 				continue
 			}
-			fmt.Fprint(c.Writer, sse) //nolint:errcheck
+			payload = reverseToolNamesIfPresent(c, payload)
+			payloads, _, err := clientToolRestorer.RestoreEvent(payload)
+			if err != nil {
+				continue
+			}
+			for _, restored := range payloads {
+				eventType := gjson.GetBytes(restored, "type").String()
+				if _, err := fmt.Fprintf(c.Writer, "event: %s\ndata: %s\n\n", eventType, restored); err != nil {
+					clientDisconnected = true
+					break
+				}
+				wrote = true
+			}
+			if clientDisconnected {
+				break
+			}
+		}
+		if wrote {
+			c.Writer.Flush()
 		}
-		c.Writer.Flush()
 	}
 
 	return resultWithUsage(), nil
diff --git a/backend/internal/service/openai_gateway_usage.go b/backend/internal/service/openai_gateway_usage.go
index c55bcdbeac89..ab2b5997340e 100644
--- a/backend/internal/service/openai_gateway_usage.go
+++ b/backend/internal/service/openai_gateway_usage.go
@@ -200,6 +200,7 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 		result.UpstreamModel,
 		result.Model,
 	)
+	billingModels = s.filterCNProviderBillingModelCandidates(ctx, account, apiKey, billingModels)
 	serviceTier := ""
 	if result.ServiceTier != nil {
 		serviceTier = strings.TrimSpace(*result.ServiceTier)
@@ -253,7 +254,7 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 			result.AudioUsage != nil || result.SearchCount > 0,
 	); responseModel != "" && !strings.EqualFold(responseModel, baselineBillingModel) {
 		if identified, responseChannelPriced := s.hasIdentifiedOpenAIResponsePricing(ctx, responseModel, apiKey); identified {
-			responseModels := usageBillingModelCandidates(responseModel)
+			responseModels := s.filterCNProviderBillingModelCandidates(ctx, account, apiKey, usageBillingModelCandidates(responseModel))
 			responseCost, responseErr := s.calculateOpenAIRecordUsageCost(
 				ctx, result, apiKey, responseModels, multiplier, imageMultiplier,
 				videoMultiplier, baseMultiplier, tokens, serviceTier, longContextBillingGate,
@@ -585,7 +586,7 @@ func (s *OpenAIGatewayService) calculateOpenAIRecordUsageCost(
 	if tokenCost == nil {
 		if tokenBillingAttempted {
 			if lastErr == nil {
-				lastErr = errors.New("no non-empty billing model candidates")
+				lastErr = fmt.Errorf("%w: no non-empty billing model candidates", ErrModelPricingUnavailable)
 			}
 			return nil, fmt.Errorf("calculate OpenAI usage cost failed for billing models %s: %w", strings.Join(billingModels, ","), lastErr)
 		}
@@ -593,8 +594,11 @@ func (s *OpenAIGatewayService) calculateOpenAIRecordUsageCost(
 		if searchCost != nil {
 			return searchCost, nil
 		}
+		// 空候选按「无价可循」处理并携带 ErrModelPricingUnavailable：上层据此走
+		// 零成本+告警落账，而不是丢弃整条 usage 记录。CN 账号的 claude-* 候选被
+		// filterCNProviderBillingModelCandidates 全数过滤后即落到这里。
 		if lastErr == nil {
-			lastErr = errors.New("openai usage billing model is empty")
+			lastErr = fmt.Errorf("%w: openai usage billing model is empty", ErrModelPricingUnavailable)
 		}
 		return nil, fmt.Errorf("calculate OpenAI usage cost failed for billing models %s: %w", strings.Join(billingModels, ","), lastErr)
 	}
@@ -841,6 +845,36 @@ func groupMediaPricingLooksIncomplete(group *Group) bool {
 		group.VideoPrice480P == nil && group.VideoPrice720P == nil && group.VideoPrice1080P == nil
 }
 
+// filterCNProviderBillingModelCandidates 过滤国产供应商（kimi/zhipu/deepseek）
+// 账号的计费候选模型名：claude-* 候选仅在运营者显式配置了分组/渠道定价时保留。
+//
+// 背景：候选链的兜底候选含客户端请求的原始模型名。CN 上游的 Anthropic 兼容端点
+// 接受 claude-* 模型名但从不真正服务 Claude 模型；若放行，目录里的 Claude 价卡
+// 与 getFallbackPricing 的 "claude"→Sonnet 统一兜底会把 CN 流量按 Claude 原价
+// （数倍～数十倍）静默误计，且 usage 日志显示的正是 claude-* 名，无从察觉。
+// 候选全部落空时走既有的零成本+告警路径（openai_usage.pricing_missing_record_
+// zero_cost），与定价层「未知型号不回退以避免误计价」的既有设计意图一致；
+// 运营者的修复手段是配置账号级 model_mapping（映射到已定价的 CN 模型）或
+// 分组/渠道显式定价。
+func (s *OpenAIGatewayService) filterCNProviderBillingModelCandidates(ctx context.Context, account *Account, apiKey *APIKey, candidates []string) []string {
+	if account == nil || !account.IsCNProvider() {
+		return candidates
+	}
+	out := make([]string, 0, len(candidates))
+	for _, candidate := range candidates {
+		trimmed := strings.TrimSpace(candidate)
+		if trimmed == "" {
+			continue
+		}
+		if strings.Contains(strings.ToLower(trimmed), "claude") &&
+			s.resolveOpenAIChannelPricing(ctx, trimmed, apiKey) == nil {
+			continue
+		}
+		out = append(out, candidate)
+	}
+	return out
+}
+
 func (s *OpenAIGatewayService) resolveOpenAIChannelPricing(ctx context.Context, billingModel string, apiKey *APIKey) *ResolvedPricing {
 	if s.resolver == nil || apiKey == nil || apiKey.Group == nil {
 		return nil
diff --git a/backend/internal/service/openai_messages_dispatch.go b/backend/internal/service/openai_messages_dispatch.go
index aedfb1b3f7f5..562e9583c51f 100644
--- a/backend/internal/service/openai_messages_dispatch.go
+++ b/backend/internal/service/openai_messages_dispatch.go
@@ -79,6 +79,13 @@ func (g *Group) ResolveMessagesDispatchModel(requestedModel string) string {
 		return xai.ModelMappingWithOptions(opts)["claude-*"]
 	}
 
+	// 国产供应商分组:调度级模型映射不适用(其配置被 sanitize 置空,且下方的
+	// gpt-5.x 默认值是 openai 专属,发给 CN 上游必错)。模型改写完全交给账号级
+	// model_mapping;anthropic 协议上游本身接受 claude-* 模型名。
+	if IsCNProvider(g.Platform) {
+		return ""
+	}
+
 	cfg := normalizeOpenAIMessagesDispatchModelConfig(g.MessagesDispatchModelConfig)
 	if mappedModel := strings.TrimSpace(cfg.ExactModelMappings[requestedModel]); mappedModel != "" {
 		return mappedModel
diff --git a/backend/internal/service/ratelimit_service.go b/backend/internal/service/ratelimit_service.go
index 0f1449191485..b361901653fc 100644
--- a/backend/internal/service/ratelimit_service.go
+++ b/backend/internal/service/ratelimit_service.go
@@ -906,7 +906,10 @@ func (s *RateLimitService) handle403(ctx context.Context, account *Account, upst
 	if account.Platform == PlatformAntigravity {
 		return s.handleAntigravity403(ctx, account, upstreamMsg, responseBody)
 	}
-	if account.Platform == PlatformOpenAI {
+	// 国产供应商与 openai 同口径:HTML 403(CDN/代理拦截页)不构成账号失效证据,
+	// 且 403 在 failover 状态集里会被逐账号重放——直接 SetError 会让一个坏请求/
+	// 一层坏代理连环永久禁用整组账号。走 HTML 豁免 + N 次累计 + 临时冷却。
+	if account.Platform == PlatformOpenAI || IsCNProvider(account.Platform) {
 		return s.handleOpenAI403(ctx, account, upstreamMsg, responseBody)
 	}
 	// 非 Antigravity 平台：保持原有行为

From c38c5beef81227f19b27356053a6db7c7e4072f4 Mon Sep 17 00:00:00 2001
From: shaw <dev@sub2api.org>
Date: Mon, 17 Aug 2026 17:44:33 +0800
Subject: [PATCH 40/68] =?UTF-8?q?fix(i18n):=20CN=20=E4=BD=99=E9=A2=9D?=
 =?UTF-8?q?=E5=8D=95=E5=85=83=E6=A0=BC=E8=AF=AF=E5=BC=95=20grokBalance=20?=
 =?UTF-8?q?=E9=94=AE=E8=87=B4=E6=B8=B2=E6=9F=93=E5=8E=9F=E5=A7=8B=20key?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

CNProviderBalanceCell 无快照占位标签引用 admin.accounts.grokBalance,
但该键实际嵌套在 admin.accounts.usageWindow 下,未命中时 DeepSeek/Kimi
payg 账号的用量窗口列直接显示原始键名。

- cnProviders 块新增自有占位键 balance(zh: '余额 --' / en: 'Balance --'),
  单元格改引该键
- 用 esbuild 实际加载模块对 zh/en 全部 admin 语言包做键路径深度比对:
  除本处外 CN 相关键(静态 14 个 + 动态拼接 accountMode/apiProtocol 系列)
  双语全部对齐;唯一存量漂移是 claudeMaxSimulation 块(zh 嵌套于
  modelRouting 下/en 为 groups 直属),全源码零消费方属死键,不在本次
  范围内处理
---
 frontend/src/components/account/CNProviderBalanceCell.vue | 4 +++-
 frontend/src/i18n/locales/en/admin/accounts.ts            | 1 +
 frontend/src/i18n/locales/zh/admin/accounts.ts            | 1 +
 3 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/frontend/src/components/account/CNProviderBalanceCell.vue b/frontend/src/components/account/CNProviderBalanceCell.vue
index 93127cbf986e..2ad2cbf3abab 100644
--- a/frontend/src/components/account/CNProviderBalanceCell.vue
+++ b/frontend/src/components/account/CNProviderBalanceCell.vue
@@ -113,7 +113,9 @@ const formatEntry = (entry: CNProviderBalanceEntry): string => {
 
 const balanceLabel = computed(() => {
   if (currentEntries.value.length === 0) {
-    return t('admin.accounts.grokBalance')?.trim() || 'Balance'
+    // 修复:此前误引 admin.accounts.grokBalance(实际嵌套在 usageWindow 下),
+    // 未命中时渲染原始 key。CN 供应商使用自己的占位键。
+    return t('admin.accounts.cnProviders.balance')
   }
   return currentEntries.value.map(formatEntry).join(' · ')
 })
diff --git a/frontend/src/i18n/locales/en/admin/accounts.ts b/frontend/src/i18n/locales/en/admin/accounts.ts
index de85669443ff..352b9fb47488 100644
--- a/frontend/src/i18n/locales/en/admin/accounts.ts
+++ b/frontend/src/i18n/locales/en/admin/accounts.ts
@@ -125,6 +125,7 @@ export default {
           responses: 'Responses',
           responsesDesc: 'Provider’s native Responses endpoint — ideal for Codex.',
         },
+        balance: 'Balance --',
         window5h: '5-hour window',
         windowWeekly: 'Weekly window',
         probeTooltip: 'Query the provider quota endpoint for 5-hour / weekly rolling window usage',
diff --git a/frontend/src/i18n/locales/zh/admin/accounts.ts b/frontend/src/i18n/locales/zh/admin/accounts.ts
index be47ced57e87..896276418672 100644
--- a/frontend/src/i18n/locales/zh/admin/accounts.ts
+++ b/frontend/src/i18n/locales/zh/admin/accounts.ts
@@ -328,6 +328,7 @@ export default {
           responses: 'Responses',
           responsesDesc: '供应商原生 Responses 端点，适配 Codex。',
         },
+        balance: '余额 --',
         window5h: '5 小时窗口',
         windowWeekly: '每周窗口',
         probeTooltip: '请求供应商额度端点，查询 5 小时 / 每周滚动窗口用量',

From 9f24a55305c7650443d555922c083051997fe2db Mon Sep 17 00:00:00 2001
From: lyen1688 <zhangxixu@gmail.com>
Date: Mon, 17 Aug 2026 19:13:03 +0800
Subject: [PATCH 41/68] =?UTF-8?q?=E5=8A=9F=E8=83=BD=EF=BC=9A=E6=94=AF?=
 =?UTF-8?q?=E6=8C=81=E6=B8=A0=E9=81=93=E6=A8=A1=E5=9E=8B=E5=88=86=E6=97=B6?=
 =?UTF-8?q?=E5=80=8D=E7=8E=87=E5=AE=9A=E4=BB=B7?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../internal/handler/admin/channel_handler.go | 102 +++++++---
 .../handler/admin/channel_handler_test.go     |  43 ++++
 .../handler/openai_gateway_handler.go         |  45 +++--
 .../handler/openai_ws_turn_pricing_test.go    |  28 +--
 .../repository/channel_repo_pricing.go        |  54 +++++-
 .../channel_repo_pricing_time_test.go         | 181 +++++++++++++++++
 backend/internal/service/admin_group.go       |   6 +
 .../service/admin_service_group_test.go       |  57 ++++++
 backend/internal/service/billing_service.go   |  26 ++-
 .../service/billing_service_unified_test.go   | 148 ++++++++++++++
 backend/internal/service/channel.go           |  50 +++--
 backend/internal/service/channel_service.go   |  54 +++++-
 .../internal/service/channel_service_test.go  |  62 ++++++
 backend/internal/service/channel_test.go      |  16 +-
 .../service/custom_channel_time_pricing.go    | 140 ++++++++++++++
 .../custom_channel_time_pricing_test.go       | 166 ++++++++++++++++
 .../service/gateway_record_usage_test.go      |  32 +++
 .../internal/service/gateway_usage_billing.go |  13 +-
 .../openai_alpha_search_billing_test.go       |   7 +-
 .../openai_gateway_record_usage_test.go       |  83 ++++++++
 .../openai_gateway_search_surcharge_test.go   |   4 +
 .../internal/service/openai_gateway_usage.go  |  11 +-
 .../internal/service/openai_ws_forwarder.go   |   3 +
 .../service/openai_ws_forwarder_ingress.go    |   8 +-
 ...openai_ws_passthrough_turn_pricing_test.go | 141 ++++++++++++--
 .../service/openai_ws_v2/passthrough_relay.go |  56 +++++-
 .../openai_ws_v2/passthrough_relay_test.go    | 137 +++++++++++++
 .../openai_ws_v2_passthrough_adapter.go       |  32 ++-
 .../225_channel_model_time_pricing.sql        |   5 +
 frontend/src/api/admin/channels.ts            |  12 ++
 .../admin/channel/PricingEntryCard.vue        |  16 +-
 .../admin/channel/TimePricingSection.vue      | 161 +++++++++++++++
 .../PricingEntryCard.timePricing.spec.ts      |  71 +++++++
 .../__tests__/TimePricingSection.spec.ts      | 183 ++++++++++++++++++
 .../admin/channel/__tests__/types.spec.ts     |  75 ++++++-
 .../src/components/admin/channel/types.ts     | 130 ++++++++++++-
 .../src/i18n/locales/en/admin/channels.ts     |  14 ++
 .../src/i18n/locales/zh/admin/channels.ts     |  14 ++
 frontend/src/views/admin/ChannelsView.vue     |  38 +++-
 frontend/src/views/admin/GroupsView.vue       |   4 +
 40 files changed, 2297 insertions(+), 131 deletions(-)
 create mode 100644 backend/internal/repository/channel_repo_pricing_time_test.go
 create mode 100644 backend/internal/service/custom_channel_time_pricing.go
 create mode 100644 backend/internal/service/custom_channel_time_pricing_test.go
 create mode 100644 backend/migrations/225_channel_model_time_pricing.sql
 create mode 100644 frontend/src/components/admin/channel/TimePricingSection.vue
 create mode 100644 frontend/src/components/admin/channel/__tests__/PricingEntryCard.timePricing.spec.ts
 create mode 100644 frontend/src/components/admin/channel/__tests__/TimePricingSection.spec.ts

diff --git a/backend/internal/handler/admin/channel_handler.go b/backend/internal/handler/admin/channel_handler.go
index ade8f0c95cab..90c1a69b6d07 100644
--- a/backend/internal/handler/admin/channel_handler.go
+++ b/backend/internal/handler/admin/channel_handler.go
@@ -57,17 +57,29 @@ type updateChannelRequest struct {
 }
 
 type channelModelPricingRequest struct {
-	Platform         string                   `json:"platform" binding:"omitempty,max=50"`
-	Models           []string                 `json:"models" binding:"required,min=1,max=100"`
-	BillingMode      string                   `json:"billing_mode" binding:"omitempty,oneof=token per_request image"`
-	InputPrice       *float64                 `json:"input_price" binding:"omitempty,min=0"`
-	OutputPrice      *float64                 `json:"output_price" binding:"omitempty,min=0"`
-	CacheWritePrice  *float64                 `json:"cache_write_price" binding:"omitempty,min=0"`
-	CacheReadPrice   *float64                 `json:"cache_read_price" binding:"omitempty,min=0"`
-	ImageInputPrice  *float64                 `json:"image_input_price" binding:"omitempty,min=0"`
-	ImageOutputPrice *float64                 `json:"image_output_price" binding:"omitempty,min=0"`
-	PerRequestPrice  *float64                 `json:"per_request_price" binding:"omitempty,min=0"`
-	Intervals        []pricingIntervalRequest `json:"intervals"`
+	Platform         string                     `json:"platform" binding:"omitempty,max=50"`
+	Models           []string                   `json:"models" binding:"required,min=1,max=100"`
+	BillingMode      string                     `json:"billing_mode" binding:"omitempty,oneof=token per_request image"`
+	InputPrice       *float64                   `json:"input_price" binding:"omitempty,min=0"`
+	OutputPrice      *float64                   `json:"output_price" binding:"omitempty,min=0"`
+	CacheWritePrice  *float64                   `json:"cache_write_price" binding:"omitempty,min=0"`
+	CacheReadPrice   *float64                   `json:"cache_read_price" binding:"omitempty,min=0"`
+	ImageInputPrice  *float64                   `json:"image_input_price" binding:"omitempty,min=0"`
+	ImageOutputPrice *float64                   `json:"image_output_price" binding:"omitempty,min=0"`
+	PerRequestPrice  *float64                   `json:"per_request_price" binding:"omitempty,min=0"`
+	Intervals        []pricingIntervalRequest   `json:"intervals"`
+	TimePricing      *channelTimePricingRequest `json:"time_pricing"`
+}
+
+type channelTimePricingRequest struct {
+	Timezone string                            `json:"timezone"`
+	Periods  []channelTimePricingPeriodRequest `json:"periods"`
+}
+
+type channelTimePricingPeriodRequest struct {
+	StartTime  string  `json:"start_time"`
+	EndTime    string  `json:"end_time"`
+	Multiplier float64 `json:"multiplier"`
 }
 
 type pricingIntervalRequest struct {
@@ -108,18 +120,30 @@ type channelResponse struct {
 }
 
 type channelModelPricingResponse struct {
-	ID               int64                     `json:"id"`
-	Platform         string                    `json:"platform"`
-	Models           []string                  `json:"models"`
-	BillingMode      string                    `json:"billing_mode"`
-	InputPrice       *float64                  `json:"input_price"`
-	OutputPrice      *float64                  `json:"output_price"`
-	CacheWritePrice  *float64                  `json:"cache_write_price"`
-	CacheReadPrice   *float64                  `json:"cache_read_price"`
-	ImageInputPrice  *float64                  `json:"image_input_price"`
-	ImageOutputPrice *float64                  `json:"image_output_price"`
-	PerRequestPrice  *float64                  `json:"per_request_price"`
-	Intervals        []pricingIntervalResponse `json:"intervals"`
+	ID               int64                       `json:"id"`
+	Platform         string                      `json:"platform"`
+	Models           []string                    `json:"models"`
+	BillingMode      string                      `json:"billing_mode"`
+	InputPrice       *float64                    `json:"input_price"`
+	OutputPrice      *float64                    `json:"output_price"`
+	CacheWritePrice  *float64                    `json:"cache_write_price"`
+	CacheReadPrice   *float64                    `json:"cache_read_price"`
+	ImageInputPrice  *float64                    `json:"image_input_price"`
+	ImageOutputPrice *float64                    `json:"image_output_price"`
+	PerRequestPrice  *float64                    `json:"per_request_price"`
+	Intervals        []pricingIntervalResponse   `json:"intervals"`
+	TimePricing      *channelTimePricingResponse `json:"time_pricing"`
+}
+
+type channelTimePricingResponse struct {
+	Timezone string                             `json:"timezone"`
+	Periods  []channelTimePricingPeriodResponse `json:"periods"`
+}
+
+type channelTimePricingPeriodResponse struct {
+	StartTime  string  `json:"start_time"`
+	EndTime    string  `json:"end_time"`
+	Multiplier float64 `json:"multiplier"`
 }
 
 type pricingIntervalResponse struct {
@@ -228,7 +252,23 @@ func pricingToResponse(p *service.ChannelModelPricing) channelModelPricingRespon
 		ImageOutputPrice: p.ImageOutputPrice,
 		PerRequestPrice:  p.PerRequestPrice,
 		Intervals:        intervals,
+		TimePricing:      timePricingToResponse(p.TimePricing),
+	}
+}
+
+func timePricingToResponse(value *service.ChannelTimePricing) *channelTimePricingResponse {
+	if value == nil {
+		return nil
 	}
+	periods := make([]channelTimePricingPeriodResponse, 0, len(value.Periods))
+	for _, period := range value.Periods {
+		periods = append(periods, channelTimePricingPeriodResponse{
+			StartTime:  period.StartTime,
+			EndTime:    period.EndTime,
+			Multiplier: period.Multiplier,
+		})
+	}
+	return &channelTimePricingResponse{Timezone: value.Timezone, Periods: periods}
 }
 
 func intervalToResponse(iv service.PricingInterval) pricingIntervalResponse {
@@ -280,11 +320,27 @@ func pricingRequestToService(reqs []channelModelPricingRequest) []service.Channe
 			ImageOutputPrice: r.ImageOutputPrice,
 			PerRequestPrice:  r.PerRequestPrice,
 			Intervals:        intervals,
+			TimePricing:      timePricingRequestToService(r.TimePricing),
 		})
 	}
 	return result
 }
 
+func timePricingRequestToService(value *channelTimePricingRequest) *service.ChannelTimePricing {
+	if value == nil {
+		return nil
+	}
+	periods := make([]service.ChannelTimePricingPeriod, 0, len(value.Periods))
+	for _, period := range value.Periods {
+		periods = append(periods, service.ChannelTimePricingPeriod{
+			StartTime:  period.StartTime,
+			EndTime:    period.EndTime,
+			Multiplier: period.Multiplier,
+		})
+	}
+	return &service.ChannelTimePricing{Timezone: value.Timezone, Periods: periods}
+}
+
 func accountStatsPricingRuleRequestToService(r accountStatsPricingRuleRequest) service.AccountStatsPricingRule {
 	return service.AccountStatsPricingRule{
 		Name:       r.Name,
diff --git a/backend/internal/handler/admin/channel_handler_test.go b/backend/internal/handler/admin/channel_handler_test.go
index d05a1a6a3b21..7f9c050fa524 100644
--- a/backend/internal/handler/admin/channel_handler_test.go
+++ b/backend/internal/handler/admin/channel_handler_test.go
@@ -421,6 +421,49 @@ func TestPricingRequestToService_NilPriceFields(t *testing.T) {
 	require.Nil(t, r.PerRequestPrice)
 }
 
+func TestPricingRequestToService_TimePricing(t *testing.T) {
+	req := channelModelPricingRequest{
+		Models:      []string{"gpt-5"},
+		BillingMode: "token",
+		TimePricing: &channelTimePricingRequest{
+			Timezone: "Asia/Shanghai",
+			Periods: []channelTimePricingPeriodRequest{{
+				StartTime: "09:00", EndTime: "12:00", Multiplier: 2,
+			}},
+		},
+	}
+
+	got := pricingRequestToService([]channelModelPricingRequest{req})
+	require.Equal(t, "Asia/Shanghai", got[0].TimePricing.Timezone)
+	require.Equal(t, 2.0, got[0].TimePricing.Periods[0].Multiplier)
+}
+
+func TestPricingRequestToService_TimePricingNil(t *testing.T) {
+	got := pricingRequestToService([]channelModelPricingRequest{{Models: []string{"gpt-5"}}})
+	require.Nil(t, got[0].TimePricing)
+}
+
+func TestPricingToResponse_TimePricing(t *testing.T) {
+	got := pricingToResponse(&service.ChannelModelPricing{
+		BillingMode: service.BillingModeToken,
+		TimePricing: &service.ChannelTimePricing{
+			Timezone: "Asia/Shanghai",
+			Periods: []service.ChannelTimePricingPeriod{{
+				StartTime: "14:00", EndTime: "18:00", Multiplier: 1.25,
+			}},
+		},
+	})
+
+	require.NotNil(t, got.TimePricing)
+	require.Equal(t, "Asia/Shanghai", got.TimePricing.Timezone)
+	require.Equal(t, 1.25, got.TimePricing.Periods[0].Multiplier)
+}
+
+func TestPricingToResponse_TimePricingNil(t *testing.T) {
+	got := pricingToResponse(&service.ChannelModelPricing{})
+	require.Nil(t, got.TimePricing)
+}
+
 // ---------------------------------------------------------------------------
 // 3. SyncPricingModels handler
 // ---------------------------------------------------------------------------
diff --git a/backend/internal/handler/openai_gateway_handler.go b/backend/internal/handler/openai_gateway_handler.go
index 28515d98f0d7..c9b034c141de 100644
--- a/backend/internal/handler/openai_gateway_handler.go
+++ b/backend/internal/handler/openai_gateway_handler.go
@@ -1420,12 +1420,10 @@ const (
 // 由 BeforeTurn 在每个 turn 开始时冻结，AfterTurn 的用量提交读取它；turn 在
 // 连接内串行推进，互斥锁只为跨用量提交 goroutine 的读取安全。
 //
-// 零值语义（重要）：ws_v2 passthrough ingress 只实现了 AfterTurn，没有任何
-// turn 起始回调，BeforeTurn 永远不会被调用。此时本值保持零，RecordUsage 经
-// openAIUsagePricingAt 回退到记录时刻——与引入分组利润控制前的基线一致。
-// 绝不能用建连时刻初始化：那会把透传连接的所有 turn 钉死在建连时的高峰因子，
-// 客户端只要峰前一分钟建连并保活，整条连接就能全程按谷价结算，正是利润控制
-// 想堵的漏洞。透传 ingress 目前不做 turn 级利润复核，只有建连时的准入门。
+// ws_v2 passthrough ingress 没有 BeforeTurn，因此本值会保持零；AfterTurn 必须
+// 以 TurnStarted 已记录的所属 turn 开始时刻为回退，而不是用建连或记录时刻。
+// 这样每个 passthrough turn 都按自己的开始时刻计价，但不改变其仅在建连时执行
+// 准入门、没有 turn 级利润复核的既有行为。
 type openAIWSTurnPricing struct {
 	mu sync.Mutex
 	at time.Time
@@ -1437,10 +1435,13 @@ func (p *openAIWSTurnPricing) freeze(at time.Time) {
 	p.mu.Unlock()
 }
 
-func (p *openAIWSTurnPricing) current() time.Time {
+func (p *openAIWSTurnPricing) currentOr(fallback time.Time) time.Time {
 	p.mu.Lock()
 	defer p.mu.Unlock()
-	return p.at
+	if !p.at.IsZero() {
+		return p.at
+	}
+	return fallback
 }
 
 // recordOpenAIProfitVeto 记录 OpenAI 侧选号循环的一次利润门终检否决：把账号
@@ -1703,6 +1704,7 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
 		closeOpenAIClientWS(wsConn, coderws.StatusPolicyViolation, "missing first response.create message")
 		return
 	}
+	firstTurnStartedAt := time.Now()
 	if msgType != coderws.MessageText && msgType != coderws.MessageBinary {
 		closeOpenAIClientWS(wsConn, coderws.StatusPolicyViolation, "unsupported websocket message type")
 		return
@@ -2031,19 +2033,37 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
 
 		maxReasoningEffort, reasoningEffortMappings, _ := openAIReasoningEffortPolicyForRequest(c, apiKey)
 		var requestPayloadHash string
+		var turnStartsMu sync.Mutex
+		turnStarts := make(map[int]time.Time, 4)
+		recordTurnStart := func(turn int, startedAt time.Time) {
+			if turn <= 0 || startedAt.IsZero() {
+				return
+			}
+			turnStartsMu.Lock()
+			turnStarts[turn] = startedAt
+			turnStartsMu.Unlock()
+		}
+		getTurnStart := func(turn int) time.Time {
+			turnStartsMu.Lock()
+			startedAt := turnStarts[turn]
+			delete(turnStarts, turn)
+			turnStartsMu.Unlock()
+			return startedAt
+		}
 		// Passthrough rejects overlapping response.create frames, so one immutable
 		// turn-tagged slot preserves the exact mapping used for the in-flight request.
 		var turnChannelMapping atomic.Pointer[openAIWSTurnChannelMappingSnapshot]
 		turnChannelMapping.Store(&openAIWSTurnChannelMappingSnapshot{turn: 1, mapping: channelMappingWS})
-		// turn 级定价：BeforeTurn 重新冻结 pricingAt 并按最新门复核当前账号，
-		// AfterTurn 的计费读取所属 turn 的时刻。零值起步的语义见
-		// openAIWSTurnPricing 的注释——绝不能用建连时刻初始化。
+		// turn 级定价：BeforeTurn 重新冻结 pricingAt 并按最新门复核当前账号；
+		// passthrough 没有 BeforeTurn 时，AfterTurn 回退到 TurnStarted 的所属 turn 时刻。
 		var turnPricing openAIWSTurnPricing
 		hooks := &service.OpenAIWSIngressHooks{
 			ClientLifecycleContext:  clientLifecycleCtx,
 			InitialRequestModel:     reqModel,
+			InitialTurnStartedAt:    firstTurnStartedAt,
 			MaxReasoningEffort:      maxReasoningEffort,
 			ReasoningEffortMappings: reasoningEffortMappings,
+			TurnStarted:             recordTurnStart,
 			BeforeRequest: func(turn int, payload []byte, originalModel string) error {
 				c.Set(securityAuditWSTurnContextKey, turn)
 				if turn == 1 {
@@ -2129,6 +2149,7 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
 				return nil
 			},
 			AfterTurn: func(turn int, result *service.OpenAIForwardResult, turnErr error) {
+				turnStart := getTurnStart(turn)
 				// F1: cyber 标记按 turn 生命周期清理——defer 保证任意早返回路径都执行；
 				// CyberBlocked 必须在 submit 前同步预捕获（task 闭包由 worker 池异步执行，
 				// 届时 defer 已清除标记）。
@@ -2196,7 +2217,7 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
 				upstreamEndpoint := resolveOpenAIUpstreamEndpoint(c, account, result)
 				quotaPlatform := service.QuotaPlatform(c.Request.Context(), apiKey)
 				sessionID := service.ExtractClientSessionID(c)
-				turnRecordPricingAt := turnPricing.current()
+				turnRecordPricingAt := turnPricing.currentOr(turnStart)
 				cyberBlocked := service.GetOpsCyberPolicy(c) != nil
 				h.submitOpenAIUsageRecordTask(ctx, result, func(taskCtx context.Context) {
 					if err := h.gatewayService.RecordUsage(taskCtx, &service.OpenAIRecordUsageInput{
diff --git a/backend/internal/handler/openai_ws_turn_pricing_test.go b/backend/internal/handler/openai_ws_turn_pricing_test.go
index b9bb204fd60b..3a59f323ba4b 100644
--- a/backend/internal/handler/openai_ws_turn_pricing_test.go
+++ b/backend/internal/handler/openai_ws_turn_pricing_test.go
@@ -7,16 +7,20 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
-// TestOpenAIWSTurnPricingZeroValue 钉死 WS turn 定价的零值语义：
-// 没有 turn 起始回调的 ingress 模式（ws_v2 passthrough 只实现 AfterTurn）
-// 必须让 pricingAt 保持零，由 RecordUsage 回退到记录时刻。
-//
-// 反例（本 PR 引入的回归）：用建连时刻初始化，会把透传连接的所有 turn 钉死在
-// 建连时的高峰因子——客户端峰前一分钟建连并保活，整条连接就按谷价结算。
-func TestOpenAIWSTurnPricingZeroValue(t *testing.T) {
-	var p openAIWSTurnPricing
-	require.True(t, p.current().IsZero(),
-		"未经 turn 起始回调冻结时必须保持零值，交由 RecordUsage 回退记录时刻")
+func TestOpenAIWSTurnPricingCurrentOr(t *testing.T) {
+	fallback := time.Date(2024, time.January, 2, 2, 0, 0, 0, time.UTC)
+
+	t.Run("frozen time takes precedence", func(t *testing.T) {
+		frozen := fallback.Add(time.Minute)
+		var p openAIWSTurnPricing
+		p.freeze(frozen)
+		require.Equal(t, frozen, p.currentOr(fallback))
+	})
+
+	t.Run("zero value falls back to turn start", func(t *testing.T) {
+		var p openAIWSTurnPricing
+		require.Equal(t, fallback, p.currentOr(fallback))
+	})
 }
 
 // TestOpenAIWSTurnPricingFreezePerTurn 钉死每个 turn 的 BeforeTurn 都会覆盖
@@ -27,8 +31,8 @@ func TestOpenAIWSTurnPricingFreezePerTurn(t *testing.T) {
 	turn2 := time.Now()
 
 	p.freeze(turn1)
-	require.Equal(t, turn1, p.current())
+	require.Equal(t, turn1, p.currentOr(time.Time{}))
 
 	p.freeze(turn2)
-	require.Equal(t, turn2, p.current(), "后续 turn 必须使用自己的定价时刻")
+	require.Equal(t, turn2, p.currentOr(time.Time{}), "后续 turn 必须使用自己的定价时刻")
 }
diff --git a/backend/internal/repository/channel_repo_pricing.go b/backend/internal/repository/channel_repo_pricing.go
index 995621139e9d..b7a9fd6b02ad 100644
--- a/backend/internal/repository/channel_repo_pricing.go
+++ b/backend/internal/repository/channel_repo_pricing.go
@@ -16,7 +16,7 @@ import (
 
 func (r *channelRepository) ListModelPricing(ctx context.Context, channelID int64) ([]service.ChannelModelPricing, error) {
 	rows, err := r.db.QueryContext(ctx,
-		`SELECT id, channel_id, platform, models, billing_mode, input_price, output_price, cache_write_price, cache_read_price, image_input_price, image_output_price, per_request_price, created_at, updated_at
+		`SELECT id, channel_id, platform, models, billing_mode, input_price, output_price, cache_write_price, cache_read_price, image_input_price, image_output_price, per_request_price, time_pricing, created_at, updated_at
 		 FROM channel_model_pricing WHERE channel_id = $1 ORDER BY id`, channelID,
 	)
 	if err != nil {
@@ -51,16 +51,20 @@ func (r *channelRepository) UpdateModelPricing(ctx context.Context, pricing *ser
 	if err != nil {
 		return fmt.Errorf("marshal models: %w", err)
 	}
+	timePricingJSON, err := marshalChannelTimePricing(pricing.TimePricing)
+	if err != nil {
+		return err
+	}
 	billingMode := pricing.BillingMode
 	if billingMode == "" {
 		billingMode = service.BillingModeToken
 	}
 	result, err := r.db.ExecContext(ctx,
 		`UPDATE channel_model_pricing
-		 SET models = $1, billing_mode = $2, input_price = $3, output_price = $4, cache_write_price = $5, cache_read_price = $6, image_input_price = $7, image_output_price = $8, per_request_price = $9, platform = $10, updated_at = NOW()
-		 WHERE id = $11`,
+		 SET models = $1, billing_mode = $2, input_price = $3, output_price = $4, cache_write_price = $5, cache_read_price = $6, image_input_price = $7, image_output_price = $8, per_request_price = $9, time_pricing = $10, platform = $11, updated_at = NOW()
+		 WHERE id = $12`,
 		modelsJSON, billingMode, pricing.InputPrice, pricing.OutputPrice, pricing.CacheWritePrice, pricing.CacheReadPrice,
-		pricing.ImageInputPrice, pricing.ImageOutputPrice, pricing.PerRequestPrice, pricing.Platform, pricing.ID,
+		pricing.ImageInputPrice, pricing.ImageOutputPrice, pricing.PerRequestPrice, timePricingJSON, pricing.Platform, pricing.ID,
 	)
 	if err != nil {
 		return fmt.Errorf("update model pricing: %w", err)
@@ -91,7 +95,7 @@ func (r *channelRepository) ReplaceModelPricing(ctx context.Context, channelID i
 // batchLoadModelPricing 批量加载多个渠道的模型定价（含区间）
 func (r *channelRepository) batchLoadModelPricing(ctx context.Context, channelIDs []int64) (map[int64][]service.ChannelModelPricing, error) {
 	rows, err := r.db.QueryContext(ctx,
-		`SELECT id, channel_id, platform, models, billing_mode, input_price, output_price, cache_write_price, cache_read_price, image_input_price, image_output_price, per_request_price, created_at, updated_at
+		`SELECT id, channel_id, platform, models, billing_mode, input_price, output_price, cache_write_price, cache_read_price, image_input_price, image_output_price, per_request_price, time_pricing, created_at, updated_at
 		 FROM channel_model_pricing WHERE channel_id = ANY($1) ORDER BY channel_id, id`,
 		pq.Array(channelIDs),
 	)
@@ -169,16 +173,22 @@ func scanModelPricingRows(rows *sql.Rows) ([]service.ChannelModelPricing, []int6
 	for rows.Next() {
 		var p service.ChannelModelPricing
 		var modelsJSON []byte
+		var timePricingJSON []byte
 		if err := rows.Scan(
 			&p.ID, &p.ChannelID, &p.Platform, &modelsJSON, &p.BillingMode,
 			&p.InputPrice, &p.OutputPrice, &p.CacheWritePrice, &p.CacheReadPrice,
-			&p.ImageInputPrice, &p.ImageOutputPrice, &p.PerRequestPrice, &p.CreatedAt, &p.UpdatedAt,
+			&p.ImageInputPrice, &p.ImageOutputPrice, &p.PerRequestPrice, &timePricingJSON, &p.CreatedAt, &p.UpdatedAt,
 		); err != nil {
 			return nil, nil, fmt.Errorf("scan model pricing: %w", err)
 		}
 		if err := json.Unmarshal(modelsJSON, &p.Models); err != nil {
 			p.Models = []string{}
 		}
+		timePricing, err := unmarshalChannelTimePricing(timePricingJSON)
+		if err != nil {
+			return nil, nil, err
+		}
+		p.TimePricing = timePricing
 		pricingIDs = append(pricingIDs, p.ID)
 		result = append(result, p)
 	}
@@ -220,6 +230,10 @@ func createModelPricingExec(ctx context.Context, exec dbExec, pricing *service.C
 	if err != nil {
 		return fmt.Errorf("marshal models: %w", err)
 	}
+	timePricingJSON, err := marshalChannelTimePricing(pricing.TimePricing)
+	if err != nil {
+		return err
+	}
 	billingMode := pricing.BillingMode
 	if billingMode == "" {
 		billingMode = service.BillingModeToken
@@ -229,11 +243,11 @@ func createModelPricingExec(ctx context.Context, exec dbExec, pricing *service.C
 		platform = "anthropic"
 	}
 	err = exec.QueryRowContext(ctx,
-		`INSERT INTO channel_model_pricing (channel_id, platform, models, billing_mode, input_price, output_price, cache_write_price, cache_read_price, image_input_price, image_output_price, per_request_price)
-		 VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11) RETURNING id, created_at, updated_at`,
+		`INSERT INTO channel_model_pricing (channel_id, platform, models, billing_mode, input_price, output_price, cache_write_price, cache_read_price, image_input_price, image_output_price, per_request_price, time_pricing)
+		 VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12) RETURNING id, created_at, updated_at`,
 		pricing.ChannelID, platform, modelsJSON, billingMode,
 		pricing.InputPrice, pricing.OutputPrice, pricing.CacheWritePrice, pricing.CacheReadPrice,
-		pricing.ImageInputPrice, pricing.ImageOutputPrice, pricing.PerRequestPrice,
+		pricing.ImageInputPrice, pricing.ImageOutputPrice, pricing.PerRequestPrice, timePricingJSON,
 	).Scan(&pricing.ID, &pricing.CreatedAt, &pricing.UpdatedAt)
 	if err != nil {
 		return fmt.Errorf("insert model pricing: %w", err)
@@ -249,6 +263,28 @@ func createModelPricingExec(ctx context.Context, exec dbExec, pricing *service.C
 	return nil
 }
 
+func marshalChannelTimePricing(config *service.ChannelTimePricing) (any, error) {
+	if config == nil || len(config.Periods) == 0 {
+		return nil, nil
+	}
+	data, err := json.Marshal(config)
+	if err != nil {
+		return nil, fmt.Errorf("marshal time pricing: %w", err)
+	}
+	return string(data), nil
+}
+
+func unmarshalChannelTimePricing(data []byte) (*service.ChannelTimePricing, error) {
+	if len(data) == 0 {
+		return nil, nil
+	}
+	var config service.ChannelTimePricing
+	if err := json.Unmarshal(data, &config); err != nil {
+		return nil, fmt.Errorf("unmarshal time pricing: %w", err)
+	}
+	return &config, nil
+}
+
 func createIntervalExec(ctx context.Context, exec dbExec, iv *service.PricingInterval) error {
 	return exec.QueryRowContext(ctx,
 		`INSERT INTO channel_pricing_intervals
diff --git a/backend/internal/repository/channel_repo_pricing_time_test.go b/backend/internal/repository/channel_repo_pricing_time_test.go
new file mode 100644
index 000000000000..d7af1886e7e8
--- /dev/null
+++ b/backend/internal/repository/channel_repo_pricing_time_test.go
@@ -0,0 +1,181 @@
+//go:build unit
+
+package repository
+
+import (
+	"context"
+	"regexp"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/DATA-DOG/go-sqlmock"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/stretchr/testify/require"
+)
+
+var channelModelPricingTimePricingColumns = []string{
+	"id", "channel_id", "platform", "models", "billing_mode", "input_price", "output_price",
+	"cache_write_price", "cache_read_price", "image_input_price", "image_output_price",
+	"per_request_price", "time_pricing", "created_at", "updated_at",
+}
+
+const channelModelPricingTimePricingJSON = `{"timezone":"Asia/Shanghai","periods":[{"start_time":"09:00","end_time":"12:00","multiplier":2}]}`
+
+func newChannelModelPricingTimePricingRepo(t *testing.T) (*channelRepository, sqlmock.Sqlmock) {
+	t.Helper()
+	db, mock, err := sqlmock.New()
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+	return &channelRepository{db: db}, mock
+}
+
+func modelPricingTimePricingRow(timePricing any) *sqlmock.Rows {
+	return sqlmock.NewRows(channelModelPricingTimePricingColumns).AddRow(
+		int64(11), int64(7), "openai", `["gpt-5"]`, service.BillingModeToken,
+		nil, nil, nil, nil, nil, nil, nil, timePricing,
+		time.Date(2026, 8, 17, 0, 0, 0, 0, time.UTC), time.Date(2026, 8, 17, 1, 0, 0, 0, time.UTC),
+	)
+}
+
+func expectEmptyModelPricingIntervals(mock sqlmock.Sqlmock) {
+	mock.ExpectQuery(`SELECT id, pricing_id, min_tokens, max_tokens, tier_label`).
+		WithArgs(sqlmock.AnyArg()).
+		WillReturnRows(sqlmock.NewRows([]string{"id"}))
+}
+
+func TestChannelModelPricingTimePricingListRoundTrip(t *testing.T) {
+	repo, mock := newChannelModelPricingTimePricingRepo(t)
+	mock.ExpectQuery(`(?s)SELECT .*per_request_price, time_pricing, created_at, updated_at.*FROM channel_model_pricing.*channel_id = \$1`).
+		WithArgs(int64(7)).
+		WillReturnRows(modelPricingTimePricingRow(channelModelPricingTimePricingJSON))
+	expectEmptyModelPricingIntervals(mock)
+
+	pricing, err := repo.ListModelPricing(context.Background(), 7)
+	require.NoError(t, err)
+	require.Len(t, pricing, 1)
+	require.NotNil(t, pricing[0].TimePricing)
+	require.Equal(t, "Asia/Shanghai", pricing[0].TimePricing.Timezone)
+	require.Len(t, pricing[0].TimePricing.Periods, 1)
+	require.Equal(t, 2.0, pricing[0].TimePricing.Periods[0].Multiplier)
+	require.NoError(t, mock.ExpectationsWereMet())
+}
+
+func TestChannelModelPricingTimePricingListNullAndMalformed(t *testing.T) {
+	t.Run("SQL NULL maps to nil", func(t *testing.T) {
+		repo, mock := newChannelModelPricingTimePricingRepo(t)
+		mock.ExpectQuery(`(?s)SELECT .*per_request_price, time_pricing, created_at, updated_at.*FROM channel_model_pricing.*channel_id = \$1`).
+			WithArgs(int64(7)).
+			WillReturnRows(modelPricingTimePricingRow(nil))
+		expectEmptyModelPricingIntervals(mock)
+
+		pricing, err := repo.ListModelPricing(context.Background(), 7)
+		require.NoError(t, err)
+		require.Len(t, pricing, 1)
+		require.Nil(t, pricing[0].TimePricing)
+		require.NoError(t, mock.ExpectationsWereMet())
+	})
+
+	t.Run("malformed JSON returns repository error", func(t *testing.T) {
+		repo, mock := newChannelModelPricingTimePricingRepo(t)
+		mock.ExpectQuery(`(?s)SELECT .*per_request_price, time_pricing, created_at, updated_at.*FROM channel_model_pricing.*channel_id = \$1`).
+			WithArgs(int64(7)).
+			WillReturnRows(modelPricingTimePricingRow(`{"timezone":`))
+
+		_, err := repo.ListModelPricing(context.Background(), 7)
+		require.Error(t, err)
+		require.True(t, strings.Contains(err.Error(), "unmarshal time pricing"), "unexpected error: %v", err)
+		require.NoError(t, mock.ExpectationsWereMet())
+	})
+}
+
+func TestChannelModelPricingTimePricingCreateAndUpdateRoundTrip(t *testing.T) {
+	pricing := &service.ChannelModelPricing{
+		ID:        11,
+		ChannelID: 7,
+		Platform:  "openai",
+		Models:    []string{"gpt-5"},
+		TimePricing: &service.ChannelTimePricing{
+			Timezone: "Asia/Shanghai",
+			Periods: []service.ChannelTimePricingPeriod{{
+				StartTime: "09:00", EndTime: "12:00", Multiplier: 2,
+			}},
+		},
+	}
+
+	t.Run("create writes JSON", func(t *testing.T) {
+		repo, mock := newChannelModelPricingTimePricingRepo(t)
+		mock.ExpectQuery(regexp.QuoteMeta("INSERT INTO channel_model_pricing (channel_id, platform, models, billing_mode, input_price, output_price, cache_write_price, cache_read_price, image_input_price, image_output_price, per_request_price, time_pricing)")).
+			WithArgs(
+				int64(7), "openai", []byte(`["gpt-5"]`), service.BillingModeToken,
+				nil, nil, nil, nil, nil, nil, nil, channelModelPricingTimePricingJSON,
+			).
+			WillReturnRows(sqlmock.NewRows([]string{"id", "created_at", "updated_at"}).AddRow(int64(11), time.Time{}, time.Time{}))
+
+		require.NoError(t, repo.CreateModelPricing(context.Background(), pricing))
+		require.NoError(t, mock.ExpectationsWereMet())
+	})
+
+	t.Run("update writes JSON and entry ID", func(t *testing.T) {
+		repo, mock := newChannelModelPricingTimePricingRepo(t)
+		mock.ExpectExec(`(?s)UPDATE channel_model_pricing.*per_request_price = \$9, time_pricing = \$10, platform = \$11.*WHERE id = \$12`).
+			WithArgs(
+				[]byte(`["gpt-5"]`), service.BillingModeToken,
+				nil, nil, nil, nil, nil, nil, nil, channelModelPricingTimePricingJSON, "openai", int64(11),
+			).
+			WillReturnResult(sqlmock.NewResult(0, 1))
+
+		require.NoError(t, repo.UpdateModelPricing(context.Background(), pricing))
+		require.NoError(t, mock.ExpectationsWereMet())
+	})
+}
+
+func TestChannelModelPricingTimePricingCreateAndUpdateWriteNullWhenDisabled(t *testing.T) {
+	tests := []struct {
+		name        string
+		timePricing *service.ChannelTimePricing
+	}{
+		{name: "nil", timePricing: nil},
+		{name: "empty periods", timePricing: &service.ChannelTimePricing{Timezone: "Asia/Shanghai"}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			newPricing := func() *service.ChannelModelPricing {
+				return &service.ChannelModelPricing{
+					ID:          11,
+					ChannelID:   7,
+					Platform:    "openai",
+					Models:      []string{"gpt-5"},
+					TimePricing: tt.timePricing,
+				}
+			}
+
+			t.Run("create writes SQL NULL", func(t *testing.T) {
+				repo, mock := newChannelModelPricingTimePricingRepo(t)
+				mock.ExpectQuery(regexp.QuoteMeta("INSERT INTO channel_model_pricing (channel_id, platform, models, billing_mode, input_price, output_price, cache_write_price, cache_read_price, image_input_price, image_output_price, per_request_price, time_pricing)")).
+					WithArgs(
+						int64(7), "openai", []byte(`["gpt-5"]`), service.BillingModeToken,
+						nil, nil, nil, nil, nil, nil, nil, nil,
+					).
+					WillReturnRows(sqlmock.NewRows([]string{"id", "created_at", "updated_at"}).AddRow(int64(11), time.Time{}, time.Time{}))
+
+				require.NoError(t, repo.CreateModelPricing(context.Background(), newPricing()))
+				require.NoError(t, mock.ExpectationsWereMet())
+			})
+
+			t.Run("update writes SQL NULL", func(t *testing.T) {
+				repo, mock := newChannelModelPricingTimePricingRepo(t)
+				mock.ExpectExec(`(?s)UPDATE channel_model_pricing.*per_request_price = \$9, time_pricing = \$10, platform = \$11.*WHERE id = \$12`).
+					WithArgs(
+						[]byte(`["gpt-5"]`), service.BillingModeToken,
+						nil, nil, nil, nil, nil, nil, nil, nil, "openai", int64(11),
+					).
+					WillReturnResult(sqlmock.NewResult(0, 1))
+
+				require.NoError(t, repo.UpdateModelPricing(context.Background(), newPricing()))
+				require.NoError(t, mock.ExpectationsWereMet())
+			})
+		})
+	}
+}
diff --git a/backend/internal/service/admin_group.go b/backend/internal/service/admin_group.go
index ac6259509d1a..f8f9acb4e72a 100644
--- a/backend/internal/service/admin_group.go
+++ b/backend/internal/service/admin_group.go
@@ -987,6 +987,12 @@ func normalizeGroupModelPricing(platform string, pricing []ChannelModelPricing)
 		out[i] = pricing[i].Clone()
 		out[i].ID = 0
 		out[i].ChannelID = 0
+		if out[i].TimePricing != nil && len(out[i].TimePricing.Periods) > 0 {
+			return nil, infraerrors.BadRequest(
+				"GROUP_MODEL_TIME_PRICING_UNSUPPORTED",
+				"group model pricing does not support time pricing",
+			)
+		}
 		if strings.TrimSpace(out[i].Platform) == "" {
 			out[i].Platform = platform
 		}
diff --git a/backend/internal/service/admin_service_group_test.go b/backend/internal/service/admin_service_group_test.go
index c0b4faef5232..64bfdaeb0200 100644
--- a/backend/internal/service/admin_service_group_test.go
+++ b/backend/internal/service/admin_service_group_test.go
@@ -4,8 +4,10 @@ package service
 
 import (
 	"context"
+	"net/http"
 	"testing"
 
+	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/pagination"
 	"github.com/stretchr/testify/require"
 )
@@ -156,6 +158,61 @@ func (s *groupRepoStubForAdmin) UpdateSortOrders(_ context.Context, _ []GroupSor
 	return nil
 }
 
+func TestAdminService_CreateGroup_RejectsTimePricing(t *testing.T) {
+	repo := &groupRepoStubForAdmin{createID: 51}
+	svc := &adminServiceImpl{groupRepo: repo}
+
+	_, err := svc.CreateGroup(context.Background(), &CreateGroupInput{
+		Name:           "time-pricing-group",
+		Platform:       PlatformOpenAI,
+		RateMultiplier: 1,
+		ModelPricing: []ChannelModelPricing{{
+			Platform:    PlatformOpenAI,
+			Models:      []string{"gpt-5"},
+			BillingMode: BillingModeToken,
+			TimePricing: validTimePricingForTest(),
+		}},
+	})
+
+	require.Error(t, err)
+	appErr := infraerrors.FromError(err)
+	require.Equal(t, int32(http.StatusBadRequest), appErr.Code)
+	require.Equal(t, "GROUP_MODEL_TIME_PRICING_UNSUPPORTED", appErr.Reason)
+	require.Nil(t, repo.created)
+}
+
+func TestAdminService_UpdateGroup_RejectsTimePricing(t *testing.T) {
+	existing := &Group{ID: 1, Name: "existing", Platform: PlatformOpenAI, Status: StatusActive}
+	repo := &groupRepoStubForAdmin{getByID: existing}
+	svc := &adminServiceImpl{groupRepo: repo}
+	pricing := []ChannelModelPricing{{
+		Platform:    PlatformOpenAI,
+		Models:      []string{"gpt-5"},
+		BillingMode: BillingModeToken,
+		TimePricing: validTimePricingForTest(),
+	}}
+
+	_, err := svc.UpdateGroup(context.Background(), existing.ID, &UpdateGroupInput{ModelPricing: &pricing})
+
+	require.Error(t, err)
+	appErr := infraerrors.FromError(err)
+	require.Equal(t, int32(http.StatusBadRequest), appErr.Code)
+	require.Equal(t, "GROUP_MODEL_TIME_PRICING_UNSUPPORTED", appErr.Reason)
+	require.Nil(t, repo.updated)
+}
+
+func TestNormalizeGroupModelPricing_NormalizesEmptyTimePricing(t *testing.T) {
+	pricing, err := normalizeGroupModelPricing(PlatformOpenAI, []ChannelModelPricing{{
+		Models:      []string{"gpt-5"},
+		BillingMode: BillingModeToken,
+		TimePricing: &ChannelTimePricing{Timezone: "Asia/Shanghai"},
+	}})
+
+	require.NoError(t, err)
+	require.Len(t, pricing, 1)
+	require.Nil(t, pricing[0].TimePricing)
+}
+
 type compositeRouteRepoStubForAdmin struct {
 	routes    []CompositeModelRoute
 	created   *CompositeModelRoute
diff --git a/backend/internal/service/billing_service.go b/backend/internal/service/billing_service.go
index 0356f698a009..6ead8da02131 100644
--- a/backend/internal/service/billing_service.go
+++ b/backend/internal/service/billing_service.go
@@ -167,6 +167,27 @@ type CostBreakdown struct {
 	LongContextBillingApplied bool
 }
 
+func applyCostBreakdownMultiplier(cost *CostBreakdown, multiplier float64) {
+	if cost == nil || multiplier == 1 {
+		return
+	}
+	cost.InputCost *= multiplier
+	cost.ImageInputCost *= multiplier
+	cost.OutputCost *= multiplier
+	cost.ImageOutputCost *= multiplier
+	cost.CacheCreationCost *= multiplier
+	cost.CacheReadCost *= multiplier
+	cost.TotalCost *= multiplier
+	cost.ActualCost *= multiplier
+}
+
+func resolvedChannelTimeMultiplier(resolved *ResolvedPricing, at time.Time) float64 {
+	if resolved == nil || resolved.Source != PricingSourceChannel || resolved.channelPricing == nil {
+		return 1
+	}
+	return resolved.channelPricing.TimePricing.MultiplierAt(at)
+}
+
 // ErrModelPricingUnavailable indicates that none of the configured pricing
 // sources can price the requested model.
 var ErrModelPricingUnavailable = errors.New("pricing not found")
@@ -1030,6 +1051,7 @@ type CostInput struct {
 	UsageUnits                float64 // 音频等连续计量单位（分钟/小时/百万字符）
 	SizeTier                  string  // 按次/图片模式的层级标签（"1K","2K","4K","HD" 等）
 	RateMultiplier            float64
+	PricingAt                 time.Time             // 渠道分时定价使用的计费时刻
 	ServiceTier               string                // "priority","flex","" 等
 	Resolver                  *ModelPricingResolver // 定价解析器
 	Resolved                  *ResolvedPricing      // 可选：预解析的定价结果（避免重复 Resolve 调用）
@@ -1104,7 +1126,9 @@ func (s *BillingService) calculateTokenCost(resolved *ResolvedPricing, input Cos
 		applyLongCtx = applyLongCtx && *input.LongContextBillingEnabled
 	}
 
-	return s.computeTokenBreakdown(pricing, input.Tokens, input.RateMultiplier, input.ServiceTier, applyLongCtx), nil
+	breakdown := s.computeTokenBreakdown(pricing, input.Tokens, input.RateMultiplier, input.ServiceTier, applyLongCtx)
+	applyCostBreakdownMultiplier(breakdown, resolvedChannelTimeMultiplier(resolved, input.PricingAt))
+	return breakdown, nil
 }
 
 // computeTokenBreakdown 是 token 计费的核心逻辑，由 calculateTokenCost 和 calculateCostInternal 共用。
diff --git a/backend/internal/service/billing_service_unified_test.go b/backend/internal/service/billing_service_unified_test.go
index eabbab3dfc7e..e61a8995511a 100644
--- a/backend/internal/service/billing_service_unified_test.go
+++ b/backend/internal/service/billing_service_unified_test.go
@@ -169,6 +169,154 @@ func TestCalculateCostUnified_ImageMode(t *testing.T) {
 	require.Equal(t, string(BillingModeImage), cost.BillingMode)
 }
 
+func channelTimeResolvedForTest(base *ModelPricing, intervals []PricingInterval) *ResolvedPricing {
+	return &ResolvedPricing{
+		Mode:        BillingModeToken,
+		BasePricing: base,
+		Intervals:   intervals,
+		Source:      PricingSourceChannel,
+		channelPricing: &ChannelModelPricing{
+			BillingMode: BillingModeToken,
+			TimePricing: &ChannelTimePricing{
+				Timezone: "Asia/Shanghai",
+				Periods: []ChannelTimePricingPeriod{{
+					StartTime:  "09:00",
+					EndTime:    "12:00",
+					Multiplier: 2,
+				}},
+			},
+		},
+		longContextPricingEnabled: true,
+	}
+}
+
+func TestCalculateCostUnified_ChannelTimePricingScalesBaseAndActualCost(t *testing.T) {
+	billing := NewBillingService(&config.Config{}, nil)
+	resolved := channelTimeResolvedForTest(&ModelPricing{InputPricePerToken: 0.001}, nil)
+
+	cost, err := billing.CalculateCostUnified(CostInput{
+		Ctx:            context.Background(),
+		Model:          "model",
+		Tokens:         UsageTokens{InputTokens: 1000},
+		RateMultiplier: 0.8,
+		Resolver:       &ModelPricingResolver{},
+		Resolved:       resolved,
+		PricingAt:      time.Date(2026, 8, 17, 1, 0, 0, 0, time.UTC),
+	})
+	require.NoError(t, err)
+	require.InDelta(t, 2.0, cost.InputCost, 1e-12)
+	require.InDelta(t, 2.0, cost.TotalCost, 1e-12)
+	require.InDelta(t, 1.6, cost.ActualCost, 1e-12)
+}
+
+func TestCalculateCostUnified_ChannelTimePricingScalesMatchingInterval(t *testing.T) {
+	intervalInputPrice := 0.003
+	resolved := channelTimeResolvedForTest(
+		&ModelPricing{InputPricePerToken: 0.001},
+		[]PricingInterval{{MinTokens: 0, InputPrice: &intervalInputPrice}},
+	)
+	billing := NewBillingService(&config.Config{}, nil)
+
+	cost, err := billing.CalculateCostUnified(CostInput{
+		Ctx:       context.Background(),
+		Model:     "model",
+		Tokens:    UsageTokens{InputTokens: 1000},
+		Resolver:  &ModelPricingResolver{},
+		Resolved:  resolved,
+		PricingAt: time.Date(2026, 8, 17, 1, 0, 0, 0, time.UTC),
+	})
+	require.NoError(t, err)
+	require.InDelta(t, 6.0, cost.InputCost, 1e-12)
+	require.InDelta(t, 6.0, cost.TotalCost, 1e-12)
+}
+
+func TestCalculateCostUnified_ChannelTimePricingScalesBaseOnUnmatchedInterval(t *testing.T) {
+	intervalInputPrice := 0.003
+	resolved := channelTimeResolvedForTest(
+		&ModelPricing{InputPricePerToken: 0.001},
+		[]PricingInterval{{MinTokens: 2000, InputPrice: &intervalInputPrice}},
+	)
+	billing := NewBillingService(&config.Config{}, nil)
+
+	cost, err := billing.CalculateCostUnified(CostInput{
+		Ctx:       context.Background(),
+		Model:     "model",
+		Tokens:    UsageTokens{InputTokens: 1000},
+		Resolver:  &ModelPricingResolver{},
+		Resolved:  resolved,
+		PricingAt: time.Date(2026, 8, 17, 1, 0, 0, 0, time.UTC),
+	})
+	require.NoError(t, err)
+	require.InDelta(t, 2.0, cost.InputCost, 1e-12)
+	require.InDelta(t, 2.0, cost.TotalCost, 1e-12)
+}
+
+func TestCalculateCostUnified_ChannelTimePricingDoesNotApplyToGroupPricing(t *testing.T) {
+	resolved := channelTimeResolvedForTest(&ModelPricing{InputPricePerToken: 0.001}, nil)
+	resolved.Source = PricingSourceGroup
+	billing := NewBillingService(&config.Config{}, nil)
+
+	cost, err := billing.CalculateCostUnified(CostInput{
+		Ctx:       context.Background(),
+		Model:     "model",
+		Tokens:    UsageTokens{InputTokens: 1000},
+		Resolver:  &ModelPricingResolver{},
+		Resolved:  resolved,
+		PricingAt: time.Date(2026, 8, 17, 1, 0, 0, 0, time.UTC),
+	})
+	require.NoError(t, err)
+	require.InDelta(t, 1.0, cost.TotalCost, 1e-12)
+}
+
+func TestCalculateCostUnified_ChannelTimePricingDoesNotApplyOutsideMatchingTime(t *testing.T) {
+	resolved := channelTimeResolvedForTest(&ModelPricing{InputPricePerToken: 0.001}, nil)
+	billing := NewBillingService(&config.Config{}, nil)
+
+	for _, pricingAt := range []time.Time{
+		time.Time{},
+		time.Date(2026, 8, 17, 5, 0, 0, 0, time.UTC),
+	} {
+		cost, err := billing.CalculateCostUnified(CostInput{
+			Ctx:       context.Background(),
+			Model:     "model",
+			Tokens:    UsageTokens{InputTokens: 1000},
+			Resolver:  &ModelPricingResolver{},
+			Resolved:  resolved,
+			PricingAt: pricingAt,
+		})
+		require.NoError(t, err)
+		require.InDelta(t, 1.0, cost.TotalCost, 1e-12)
+	}
+}
+
+func TestApplyCostBreakdownMultiplierScalesAllMonetaryFields(t *testing.T) {
+	cost := &CostBreakdown{
+		InputCost:                 1,
+		ImageInputCost:            2,
+		OutputCost:                3,
+		ImageOutputCost:           4,
+		CacheCreationCost:         5,
+		CacheReadCost:             6,
+		TotalCost:                 21,
+		ActualCost:                42,
+		BillingMode:               string(BillingModeToken),
+		LongContextBillingApplied: true,
+	}
+
+	applyCostBreakdownMultiplier(cost, 1.5)
+
+	require.InDelta(t, 1.5, cost.InputCost, 1e-12)
+	require.InDelta(t, 3.0, cost.ImageInputCost, 1e-12)
+	require.InDelta(t, 4.5, cost.OutputCost, 1e-12)
+	require.InDelta(t, 6.0, cost.ImageOutputCost, 1e-12)
+	require.InDelta(t, 7.5, cost.CacheCreationCost, 1e-12)
+	require.InDelta(t, 9.0, cost.CacheReadCost, 1e-12)
+	require.InDelta(t, 31.5, cost.TotalCost, 1e-12)
+	require.InDelta(t, 63.0, cost.ActualCost, 1e-12)
+	require.Equal(t, string(BillingModeToken), cost.BillingMode)
+	require.True(t, cost.LongContextBillingApplied)
+}
+
 // TestCalculateCostUnified_RateMultiplierZeroProducesZero 锁定新行为：
 // 保存时强制 > 0；若 0 仍泄漏到计费层，按 0 计费（而非历史上的 1.0）。
 func TestCalculateCostUnified_RateMultiplierZeroProducesZero(t *testing.T) {
diff --git a/backend/internal/service/channel.go b/backend/internal/service/channel.go
index 4d5523b0a37d..5345693eb024 100644
--- a/backend/internal/service/channel.go
+++ b/backend/internal/service/channel.go
@@ -87,21 +87,35 @@ type AccountStatsPricingRule struct {
 
 // ChannelModelPricing 渠道模型定价条目
 type ChannelModelPricing struct {
-	ID               int64             `json:"id,omitempty"`
-	ChannelID        int64             `json:"channel_id,omitempty"`
-	Platform         string            `json:"platform"` // 所属平台（anthropic/openai/gemini/...）
-	Models           []string          `json:"models"`
-	BillingMode      BillingMode       `json:"billing_mode"`
-	InputPrice       *float64          `json:"input_price"`
-	OutputPrice      *float64          `json:"output_price"`
-	CacheWritePrice  *float64          `json:"cache_write_price"`
-	CacheReadPrice   *float64          `json:"cache_read_price"`
-	ImageInputPrice  *float64          `json:"image_input_price"`
-	ImageOutputPrice *float64          `json:"image_output_price"`
-	PerRequestPrice  *float64          `json:"per_request_price"`
-	Intervals        []PricingInterval `json:"intervals"`
-	CreatedAt        time.Time         `json:"created_at,omitempty"`
-	UpdatedAt        time.Time         `json:"updated_at,omitempty"`
+	ID               int64               `json:"id,omitempty"`
+	ChannelID        int64               `json:"channel_id,omitempty"`
+	Platform         string              `json:"platform"` // 所属平台（anthropic/openai/gemini/...）
+	Models           []string            `json:"models"`
+	BillingMode      BillingMode         `json:"billing_mode"`
+	InputPrice       *float64            `json:"input_price"`
+	OutputPrice      *float64            `json:"output_price"`
+	CacheWritePrice  *float64            `json:"cache_write_price"`
+	CacheReadPrice   *float64            `json:"cache_read_price"`
+	ImageInputPrice  *float64            `json:"image_input_price"`
+	ImageOutputPrice *float64            `json:"image_output_price"`
+	PerRequestPrice  *float64            `json:"per_request_price"`
+	Intervals        []PricingInterval   `json:"intervals"`
+	TimePricing      *ChannelTimePricing `json:"time_pricing,omitempty"`
+	CreatedAt        time.Time           `json:"created_at,omitempty"`
+	UpdatedAt        time.Time           `json:"updated_at,omitempty"`
+}
+
+// ChannelTimePricing 渠道模型定价的分时倍率配置。
+type ChannelTimePricing struct {
+	Timezone string                     `json:"timezone"`
+	Periods  []ChannelTimePricingPeriod `json:"periods"`
+}
+
+// ChannelTimePricingPeriod 是秒级的左闭右开分时倍率区间，并兼容历史 HH:mm 数据。
+type ChannelTimePricingPeriod struct {
+	StartTime  string  `json:"start_time"`
+	EndTime    string  `json:"end_time"`
+	Multiplier float64 `json:"multiplier"`
 }
 
 // PricingInterval 定价区间（token 区间 / 按次分层 / 图片分辨率分层）
@@ -195,6 +209,12 @@ func (p ChannelModelPricing) Clone() ChannelModelPricing {
 		cp.Intervals = make([]PricingInterval, len(p.Intervals))
 		copy(cp.Intervals, p.Intervals)
 	}
+	if p.TimePricing != nil {
+		cp.TimePricing = &ChannelTimePricing{Timezone: p.TimePricing.Timezone}
+		if p.TimePricing.Periods != nil {
+			cp.TimePricing.Periods = append([]ChannelTimePricingPeriod(nil), p.TimePricing.Periods...)
+		}
+	}
 	return cp
 }
 
diff --git a/backend/internal/service/channel_service.go b/backend/internal/service/channel_service.go
index 3a8c5556add6..a2c4bf6ebfba 100644
--- a/backend/internal/service/channel_service.go
+++ b/backend/internal/service/channel_service.go
@@ -646,7 +646,47 @@ func validatePricingEntries(pricing []ChannelModelPricing) error {
 	if err := validatePricingIntervals(pricing); err != nil {
 		return err
 	}
-	return validatePricingBillingMode(pricing)
+	if err := validatePricingBillingMode(pricing); err != nil {
+		return err
+	}
+	return validatePricingTimePricing(pricing)
+}
+
+func validatePricingTimePricing(pricing []ChannelModelPricing) error {
+	for i := range pricing {
+		config := pricing[i].TimePricing
+		if config == nil {
+			continue
+		}
+		if len(config.Periods) == 0 {
+			pricing[i].TimePricing = nil
+			continue
+		}
+		mode := pricing[i].BillingMode
+		if mode != "" && mode != BillingModeToken {
+			return infraerrors.BadRequest("TIME_PRICING_UNSUPPORTED_MODE", "time pricing only supports token billing mode")
+		}
+		if err := validateChannelTimePricing(config); err != nil {
+			return infraerrors.BadRequest("INVALID_TIME_PRICING", fmt.Sprintf(
+				"invalid time pricing for platform '%s' models %v: %v", pricing[i].Platform, pricing[i].Models, err))
+		}
+	}
+	return nil
+}
+
+func validateAccountStatsPricingRules(rules []AccountStatsPricingRule) error {
+	for i := range rules {
+		for _, pricing := range rules[i].Pricing {
+			if pricing.TimePricing != nil && len(pricing.TimePricing.Periods) > 0 {
+				return fmt.Errorf("account stats pricing rule #%d: %w", i+1,
+					infraerrors.BadRequest("ACCOUNT_STATS_TIME_PRICING_UNSUPPORTED", "account stats pricing does not support time pricing"))
+			}
+		}
+		if err := validatePricingEntries(rules[i].Pricing); err != nil {
+			return fmt.Errorf("account stats pricing rule #%d: %w", i+1, err)
+		}
+	}
+	return nil
 }
 
 // validatePricingBillingMode 校验计费模式配置：按次/图片模式必须配价格或区间，所有价格字段不能为负，区间至少有一个价格字段。
@@ -755,10 +795,8 @@ func (s *ChannelService) Create(ctx context.Context, input *CreateChannelInput)
 	if err := validateChannelConfig(channel.ModelPricing, channel.ModelMapping); err != nil {
 		return nil, err
 	}
-	for i, rule := range channel.AccountStatsPricingRules {
-		if err := validatePricingEntries(rule.Pricing); err != nil {
-			return nil, fmt.Errorf("account stats pricing rule #%d: %w", i+1, err)
-		}
+	if err := validateAccountStatsPricingRules(channel.AccountStatsPricingRules); err != nil {
+		return nil, err
 	}
 
 	if err := s.repo.Create(ctx, channel); err != nil {
@@ -799,10 +837,8 @@ func (s *ChannelService) Update(ctx context.Context, id int64, input *UpdateChan
 	if err := validateChannelConfig(channel.ModelPricing, channel.ModelMapping); err != nil {
 		return nil, err
 	}
-	for i, rule := range channel.AccountStatsPricingRules {
-		if err := validatePricingEntries(rule.Pricing); err != nil {
-			return nil, fmt.Errorf("account stats pricing rule #%d: %w", i+1, err)
-		}
+	if err := validateAccountStatsPricingRules(channel.AccountStatsPricingRules); err != nil {
+		return nil, err
 	}
 
 	oldGroupIDs := s.getOldGroupIDs(ctx, id)
diff --git a/backend/internal/service/channel_service_test.go b/backend/internal/service/channel_service_test.go
index f56f61e9173e..9a81913b04d1 100644
--- a/backend/internal/service/channel_service_test.go
+++ b/backend/internal/service/channel_service_test.go
@@ -5,8 +5,10 @@ package service
 import (
 	"context"
 	"errors"
+	"net/http"
 	"testing"
 
+	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/pagination"
 	"github.com/stretchr/testify/require"
 	"github.com/tidwall/gjson"
@@ -2460,6 +2462,66 @@ func TestValidatePricingBillingMode(t *testing.T) {
 	}
 }
 
+func validTimePricingForTest() *ChannelTimePricing {
+	return &ChannelTimePricing{Timezone: "Asia/Shanghai", Periods: []ChannelTimePricingPeriod{
+		{StartTime: "09:00", EndTime: "12:00", Multiplier: 2},
+	}}
+}
+
+func TestValidatePricingTimePricing(t *testing.T) {
+	token := []ChannelModelPricing{{BillingMode: BillingModeToken, TimePricing: validTimePricingForTest()}}
+	require.NoError(t, validatePricingTimePricing(token))
+
+	implicitToken := []ChannelModelPricing{{TimePricing: validTimePricingForTest()}}
+	require.NoError(t, validatePricingTimePricing(implicitToken))
+
+	image := []ChannelModelPricing{{BillingMode: BillingModeImage, TimePricing: validTimePricingForTest()}}
+	modeErr := infraerrors.FromError(validatePricingTimePricing(image))
+	require.Equal(t, int32(http.StatusBadRequest), modeErr.Code)
+	require.Equal(t, "TIME_PRICING_UNSUPPORTED_MODE", modeErr.Reason)
+
+	invalid := []ChannelModelPricing{{
+		Platform:    PlatformOpenAI,
+		Models:      []string{"gpt-5"},
+		BillingMode: BillingModeToken,
+		TimePricing: &ChannelTimePricing{Timezone: "UTC+8", Periods: validTimePricingForTest().Periods},
+	}}
+	invalidErr := infraerrors.FromError(validatePricingTimePricing(invalid))
+	require.Equal(t, int32(http.StatusBadRequest), invalidErr.Code)
+	require.Equal(t, "INVALID_TIME_PRICING", invalidErr.Reason)
+	require.Contains(t, invalidErr.Message, "platform 'openai'")
+	require.Contains(t, invalidErr.Message, "models [gpt-5]")
+
+	invalidMultiplier := []ChannelModelPricing{{
+		Platform:    PlatformOpenAI,
+		Models:      []string{"gpt-5"},
+		BillingMode: BillingModeToken,
+		TimePricing: &ChannelTimePricing{Timezone: "Asia/Shanghai", Periods: []ChannelTimePricingPeriod{{
+			StartTime: "09:00", EndTime: "12:00", Multiplier: 1e-12,
+		}}},
+	}}
+	invalidMultiplierRawErr := validatePricingTimePricing(invalidMultiplier)
+	require.Error(t, invalidMultiplierRawErr)
+	invalidMultiplierErr := infraerrors.FromError(invalidMultiplierRawErr)
+	require.Equal(t, int32(http.StatusBadRequest), invalidMultiplierErr.Code)
+	require.Equal(t, "INVALID_TIME_PRICING", invalidMultiplierErr.Reason)
+
+	empty := []ChannelModelPricing{{BillingMode: BillingModeToken, TimePricing: &ChannelTimePricing{Timezone: "Asia/Shanghai"}}}
+	require.NoError(t, validatePricingTimePricing(empty))
+	require.Nil(t, empty[0].TimePricing)
+}
+
+func TestValidateAccountStatsPricingRulesRejectsTimePricing(t *testing.T) {
+	rules := []AccountStatsPricingRule{{Pricing: []ChannelModelPricing{{
+		BillingMode: BillingModeToken,
+		TimePricing: validTimePricingForTest(),
+	}}}}
+
+	appErr := infraerrors.FromError(validateAccountStatsPricingRules(rules))
+	require.Equal(t, int32(http.StatusBadRequest), appErr.Code)
+	require.Equal(t, "ACCOUNT_STATS_TIME_PRICING_UNSUPPORTED", appErr.Reason)
+}
+
 // ---------------------------------------------------------------------------
 // 12. Antigravity wildcard mapping isolation
 // ---------------------------------------------------------------------------
diff --git a/backend/internal/service/channel_test.go b/backend/internal/service/channel_test.go
index 2f371f8a1c8d..19e45a02db65 100644
--- a/backend/internal/service/channel_test.go
+++ b/backend/internal/service/channel_test.go
@@ -197,6 +197,14 @@ func TestChannelModelPricingClone(t *testing.T) {
 		Intervals: []PricingInterval{
 			{MinTokens: 0, TierLabel: "tier1"},
 		},
+		TimePricing: &ChannelTimePricing{
+			Timezone: "Asia/Shanghai",
+			Periods: []ChannelTimePricingPeriod{{
+				StartTime:  "09:00",
+				EndTime:    "12:00",
+				Multiplier: 2,
+			}},
+		},
 	}
 
 	cloned := original.Clone()
@@ -207,6 +215,13 @@ func TestChannelModelPricingClone(t *testing.T) {
 
 	cloned.Intervals[0].TierLabel = "hacked"
 	require.Equal(t, "tier1", original.Intervals[0].TierLabel)
+
+	cloned.TimePricing.Timezone = "America/New_York"
+	cloned.TimePricing.Periods[0].StartTime = "10:00"
+	cloned.TimePricing.Periods[0].Multiplier = 3
+	require.Equal(t, "Asia/Shanghai", original.TimePricing.Timezone)
+	require.Equal(t, "09:00", original.TimePricing.Periods[0].StartTime)
+	require.Equal(t, 2.0, original.TimePricing.Periods[0].Multiplier)
 }
 
 // --- BillingMode.IsValid ---
@@ -513,7 +528,6 @@ func TestSupportedModels_WildcardExpandedFromPricing(t *testing.T) {
 	}
 }
 
-
 func TestSupportedModels_MissingPricingKeepsNilPricing(t *testing.T) {
 	ch := &Channel{
 		ModelMapping: map[string]map[string]string{
diff --git a/backend/internal/service/custom_channel_time_pricing.go b/backend/internal/service/custom_channel_time_pricing.go
new file mode 100644
index 000000000000..759eac987eff
--- /dev/null
+++ b/backend/internal/service/custom_channel_time_pricing.go
@@ -0,0 +1,140 @@
+package service
+
+import (
+	"fmt"
+	"math"
+	"sort"
+	"strings"
+	"sync"
+	"time"
+)
+
+var channelTimePricingLocations sync.Map
+
+type parsedChannelTimePeriod struct {
+	start      int
+	end        int
+	multiplier float64
+}
+
+// validateChannelTimePricing 校验分时倍率配置。nil 或空 periods 表示未启用。
+func validateChannelTimePricing(config *ChannelTimePricing) error {
+	if config == nil || len(config.Periods) == 0 {
+		return nil
+	}
+	if _, err := loadChannelTimePricingLocation(config.Timezone); err != nil {
+		return fmt.Errorf("timezone: %w", err)
+	}
+	_, err := parseChannelTimePeriods(config.Periods)
+	return err
+}
+
+// MultiplierAt 返回 at 对应的分时倍率。无配置或脏配置均安全降级为 1。
+func (config *ChannelTimePricing) MultiplierAt(at time.Time) float64 {
+	if config == nil || len(config.Periods) == 0 || at.IsZero() {
+		return 1.0
+	}
+	if err := validateChannelTimePricing(config); err != nil {
+		return 1.0
+	}
+	location, err := loadChannelTimePricingLocation(config.Timezone)
+	if err != nil {
+		return 1.0
+	}
+	periods, err := parseChannelTimePeriods(config.Periods)
+	if err != nil {
+		return 1.0
+	}
+
+	local := at.In(location)
+	second := local.Hour()*60*60 + local.Minute()*60 + local.Second()
+	for _, period := range periods {
+		if second >= period.start && second < period.end {
+			return period.multiplier
+		}
+	}
+	return 1.0
+}
+
+func loadChannelTimePricingLocation(name string) (*time.Location, error) {
+	if strings.TrimSpace(name) == "" {
+		return nil, fmt.Errorf("timezone is required")
+	}
+	if name == "Local" {
+		return nil, fmt.Errorf("local is not a supported timezone")
+	}
+	if cached, ok := channelTimePricingLocations.Load(name); ok {
+		location, valid := cached.(*time.Location)
+		if valid && location != nil {
+			return location, nil
+		}
+		channelTimePricingLocations.Delete(name)
+	}
+	location, err := time.LoadLocation(name)
+	if err != nil {
+		return nil, err
+	}
+	actual, _ := channelTimePricingLocations.LoadOrStore(name, location)
+	actualLocation, ok := actual.(*time.Location)
+	if !ok || actualLocation == nil {
+		return nil, fmt.Errorf("invalid cached timezone %q", name)
+	}
+	return actualLocation, nil
+}
+
+func parseChannelTime(value string, end bool) (int, error) {
+	if end && (value == "00:00" || value == "00:00:00") {
+		return 24 * 60 * 60, nil
+	}
+	layout := "15:04:05"
+	if len(value) == len("15:04") {
+		layout = "15:04"
+	}
+	parsed, err := time.Parse(layout, value)
+	if err != nil || parsed.Format(layout) != value {
+		return 0, fmt.Errorf("time %q must use HH:mm or HH:mm:ss format", value)
+	}
+	return parsed.Hour()*60*60 + parsed.Minute()*60 + parsed.Second(), nil
+}
+
+func parseChannelTimePeriods(periods []ChannelTimePricingPeriod) ([]parsedChannelTimePeriod, error) {
+	parsed := make([]parsedChannelTimePeriod, 0, len(periods))
+	for _, period := range periods {
+		if math.IsNaN(period.Multiplier) || math.IsInf(period.Multiplier, 0) || period.Multiplier <= 0 {
+			return nil, fmt.Errorf("multiplier must be finite and greater than 0")
+		}
+		if period.Multiplier < 0.01 {
+			return nil, fmt.Errorf("multiplier must be at least 0.01")
+		}
+		scaled := period.Multiplier * 100
+		if math.IsNaN(scaled) || math.IsInf(scaled, 0) {
+			return nil, fmt.Errorf("multiplier must remain finite when scaled")
+		}
+		if math.Abs(scaled-math.Round(scaled)) > 1e-9 {
+			return nil, fmt.Errorf("multiplier must have at most two decimal places")
+		}
+
+		start, err := parseChannelTime(period.StartTime, false)
+		if err != nil {
+			return nil, err
+		}
+		end, err := parseChannelTime(period.EndTime, true)
+		if err != nil {
+			return nil, err
+		}
+		if period.StartTime == period.EndTime || start >= end {
+			return nil, fmt.Errorf("start time must be before end time")
+		}
+		parsed = append(parsed, parsedChannelTimePeriod{start: start, end: end, multiplier: period.Multiplier})
+	}
+
+	sort.Slice(parsed, func(i, j int) bool {
+		return parsed[i].start < parsed[j].start
+	})
+	for i := 1; i < len(parsed); i++ {
+		if parsed[i].start < parsed[i-1].end {
+			return nil, fmt.Errorf("time pricing periods overlap")
+		}
+	}
+	return parsed, nil
+}
diff --git a/backend/internal/service/custom_channel_time_pricing_test.go b/backend/internal/service/custom_channel_time_pricing_test.go
new file mode 100644
index 000000000000..f1038af0d372
--- /dev/null
+++ b/backend/internal/service/custom_channel_time_pricing_test.go
@@ -0,0 +1,166 @@
+//go:build unit
+
+package service
+
+import (
+	"math"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+)
+
+func timeConfig(periods ...ChannelTimePricingPeriod) *ChannelTimePricing {
+	return &ChannelTimePricing{Timezone: "Asia/Shanghai", Periods: periods}
+}
+
+func onePeriod() []ChannelTimePricingPeriod {
+	return []ChannelTimePricingPeriod{{StartTime: "09:00", EndTime: "12:00", Multiplier: 2}}
+}
+
+func TestValidateChannelTimePricing(t *testing.T) {
+	tests := []struct {
+		name    string
+		config  *ChannelTimePricing
+		wantErr string
+	}{
+		{name: "nil disabled", config: nil},
+		{name: "empty disabled", config: &ChannelTimePricing{Timezone: "Asia/Shanghai"}},
+		{name: "adjacent", config: timeConfig(
+			ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 2},
+			ChannelTimePricingPeriod{StartTime: "12:00", EndTime: "14:00", Multiplier: 1.5})},
+		{name: "midnight split", config: timeConfig(
+			ChannelTimePricingPeriod{StartTime: "22:00", EndTime: "00:00", Multiplier: 2},
+			ChannelTimePricingPeriod{StartTime: "00:00", EndTime: "02:00", Multiplier: 2})},
+		{name: "second precision", config: timeConfig(
+			ChannelTimePricingPeriod{StartTime: "09:00:00", EndTime: "12:00:00", Multiplier: 2},
+			ChannelTimePricingPeriod{StartTime: "14:00:00", EndTime: "18:00:00", Multiplier: 2})},
+		{name: "second precision overlap", config: timeConfig(
+			ChannelTimePricingPeriod{StartTime: "09:00:00", EndTime: "12:00:00", Multiplier: 2},
+			ChannelTimePricingPeriod{StartTime: "11:59:59", EndTime: "14:00:00", Multiplier: 2}), wantErr: "overlap"},
+		{name: "empty timezone", config: &ChannelTimePricing{Periods: onePeriod()}, wantErr: "timezone"},
+		{name: "whitespace timezone", config: &ChannelTimePricing{Timezone: "  ", Periods: onePeriod()}, wantErr: "timezone"},
+		{name: "timezone", config: &ChannelTimePricing{Timezone: "UTC+8", Periods: onePeriod()}, wantErr: "timezone"},
+		{name: "format", config: timeConfig(ChannelTimePricingPeriod{StartTime: "9:00", EndTime: "12:00", Multiplier: 2}), wantErr: "HH:mm"},
+		{name: "equal midnight", config: timeConfig(ChannelTimePricingPeriod{StartTime: "00:00", EndTime: "00:00", Multiplier: 2}), wantErr: "before"},
+		{name: "cross midnight", config: timeConfig(ChannelTimePricingPeriod{StartTime: "22:00", EndTime: "02:00", Multiplier: 2}), wantErr: "before"},
+		{name: "overlap", config: timeConfig(
+			ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 2},
+			ChannelTimePricingPeriod{StartTime: "11:59", EndTime: "14:00", Multiplier: 2}), wantErr: "overlap"},
+		{name: "zero", config: timeConfig(ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 0}), wantErr: "greater than 0"},
+		{name: "minimum positive", config: timeConfig(ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 0.01})},
+		{name: "tiny positive", config: timeConfig(ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 1e-12}), wantErr: "at least 0.01"},
+		{name: "below minimum", config: timeConfig(ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 0.001}), wantErr: "at least 0.01"},
+		{name: "three decimals", config: timeConfig(ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 1.001}), wantErr: "decimal"},
+		{name: "scaled overflow", config: timeConfig(ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: math.MaxFloat64}), wantErr: "finite"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			err := validateChannelTimePricing(tt.config)
+			if tt.wantErr == "" {
+				require.NoError(t, err)
+				return
+			}
+			require.Error(t, err)
+			require.True(t, strings.Contains(err.Error(), tt.wantErr), "error %q does not contain %q", err, tt.wantErr)
+		})
+	}
+}
+
+func TestChannelTimePricingMultiplierAt(t *testing.T) {
+	config := timeConfig(ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 2})
+	tests := []struct {
+		name string
+		at   time.Time
+		want float64
+	}{
+		{name: "Shanghai 08:59", at: time.Date(2026, 6, 29, 0, 59, 0, 0, time.UTC), want: 1},
+		{name: "Shanghai 09:00", at: time.Date(2026, 6, 29, 1, 0, 0, 0, time.UTC), want: 2},
+		{name: "Shanghai 11:59", at: time.Date(2026, 6, 29, 3, 59, 0, 0, time.UTC), want: 2},
+		{name: "Shanghai 12:00", at: time.Date(2026, 6, 29, 4, 0, 0, 0, time.UTC), want: 1},
+		{name: "Shanghai 14:00", at: time.Date(2026, 6, 29, 6, 0, 0, 0, time.UTC), want: 1},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			require.Equal(t, tt.want, config.MultiplierAt(tt.at))
+		})
+	}
+
+	newYork := &ChannelTimePricing{Timezone: "America/New_York", Periods: onePeriod()}
+	at := time.Date(2026, 6, 29, 14, 0, 0, 0, time.UTC)
+	require.Equal(t, 1.0, config.MultiplierAt(at))
+	require.Equal(t, 2.0, newYork.MultiplierAt(at))
+}
+
+func TestChannelTimePricingMultiplierAtSecondPrecision(t *testing.T) {
+	config := timeConfig(ChannelTimePricingPeriod{StartTime: "09:00:30", EndTime: "09:00:45", Multiplier: 2})
+	shanghai, err := time.LoadLocation("Asia/Shanghai")
+	require.NoError(t, err)
+
+	tests := []struct {
+		name string
+		at   time.Time
+		want float64
+	}{
+		{name: "before", at: time.Date(2026, 6, 29, 9, 0, 29, 0, shanghai), want: 1},
+		{name: "start", at: time.Date(2026, 6, 29, 9, 0, 30, 0, shanghai), want: 2},
+		{name: "last matching second", at: time.Date(2026, 6, 29, 9, 0, 44, 999_999_999, shanghai), want: 2},
+		{name: "end", at: time.Date(2026, 6, 29, 9, 0, 45, 0, shanghai), want: 1},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			require.Equal(t, tt.want, config.MultiplierAt(tt.at))
+		})
+	}
+}
+
+func TestChannelTimePricingMultiplierAtMidnightSplit(t *testing.T) {
+	config := timeConfig(
+		ChannelTimePricingPeriod{StartTime: "22:00", EndTime: "00:00", Multiplier: 2},
+		ChannelTimePricingPeriod{StartTime: "00:00", EndTime: "02:00", Multiplier: 3},
+	)
+	shanghai, err := time.LoadLocation("Asia/Shanghai")
+	require.NoError(t, err)
+	tests := []struct {
+		name string
+		at   time.Time
+		want float64
+	}{
+		{name: "23:59", at: time.Date(2026, 6, 29, 23, 59, 0, 0, shanghai), want: 2},
+		{name: "next day 00:00", at: time.Date(2026, 6, 30, 0, 0, 0, 0, shanghai), want: 3},
+		{name: "02:00", at: time.Date(2026, 6, 30, 2, 0, 0, 0, shanghai), want: 1},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			require.Equal(t, tt.want, config.MultiplierAt(tt.at))
+		})
+	}
+}
+
+func TestChannelTimePricingMultiplierAtDegradesForInvalidConfigurations(t *testing.T) {
+	var nilConfig *ChannelTimePricing
+	zeroTime := time.Time{}
+	validAt := time.Date(2026, 6, 29, 1, 0, 0, 0, time.UTC)
+
+	require.Equal(t, 1.0, nilConfig.MultiplierAt(validAt))
+	require.Equal(t, 1.0, timeConfig().MultiplierAt(validAt))
+	require.Equal(t, 1.0, timeConfig(ChannelTimePricingPeriod{StartTime: "09:00", EndTime: "12:00", Multiplier: 2}).MultiplierAt(zeroTime))
+	require.Equal(t, 1.0, (&ChannelTimePricing{Periods: onePeriod()}).MultiplierAt(validAt))
+	require.Equal(t, 1.0, (&ChannelTimePricing{Timezone: "  ", Periods: onePeriod()}).MultiplierAt(validAt))
+	require.Equal(t, 1.0, (&ChannelTimePricing{Timezone: "UTC+8", Periods: onePeriod()}).MultiplierAt(validAt))
+	require.Equal(t, 1.0, timeConfig(ChannelTimePricingPeriod{StartTime: "22:00", EndTime: "02:00", Multiplier: 2}).MultiplierAt(validAt))
+}
+
+func TestChannelTimePricingRejectsLocalTimezone(t *testing.T) {
+	config := &ChannelTimePricing{Timezone: "Local", Periods: onePeriod()}
+
+	err := validateChannelTimePricing(config)
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "timezone")
+	require.Equal(t, 1.0, config.MultiplierAt(time.Date(2026, 6, 29, 1, 0, 0, 0, time.UTC)))
+}
diff --git a/backend/internal/service/gateway_record_usage_test.go b/backend/internal/service/gateway_record_usage_test.go
index 517d4723cbdd..53616447452e 100644
--- a/backend/internal/service/gateway_record_usage_test.go
+++ b/backend/internal/service/gateway_record_usage_test.go
@@ -345,6 +345,38 @@ func TestGatewayServiceRecordUsage_PeakRateAffectsTokenModeImageOutputTokens(t *
 	require.InDelta(t, expectedActual, userRepo.lastAmount, 1e-12)
 }
 
+func TestGatewayServiceRecordUsage_TimePricingUsesPricingAt(t *testing.T) {
+	groupID := int64(904)
+	requestStart := time.Date(2024, time.January, 2, 2, 0, 0, 0, time.UTC) // 上海 10:00
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	svc := newGatewayRecordUsageServiceForTest(usageRepo, userRepo, &openAIRecordUsageSubRepoStub{})
+	svc.resolver = newOpenAITokenImageChannelPricingResolverWithTimeForTest(t, groupID, "gpt-5.1", &ChannelTimePricing{
+		Timezone: "Asia/Shanghai",
+		Periods:  []ChannelTimePricingPeriod{{StartTime: "09:00", EndTime: "12:00", Multiplier: 2}},
+	})
+
+	err := svc.RecordUsage(context.Background(), &RecordUsageInput{
+		Result: &ForwardResult{
+			RequestID: "gateway_time_pricing_request_start",
+			Model:     "gpt-5.1",
+			Usage:     ClaudeUsage{InputTokens: 1000, OutputTokens: 500},
+		},
+		APIKey: &APIKey{ID: 804, GroupID: i64p(groupID), Group: &Group{
+			ID: groupID, RateMultiplier: 0.8, SubscriptionType: SubscriptionTypeSubscription,
+		}},
+		User:      &User{ID: 604},
+		Account:   &Account{ID: 704},
+		PricingAt: requestStart,
+	})
+
+	require.NoError(t, err)
+	require.NotNil(t, usageRepo.lastLog)
+	baseCost := 1000*3e-6 + 500*15e-6
+	require.InDelta(t, baseCost*2, usageRepo.lastLog.TotalCost, 1e-12)
+	require.InDelta(t, baseCost*2*0.8, usageRepo.lastLog.ActualCost, 1e-12)
+	require.InDelta(t, 0.8, usageRepo.lastLog.RateMultiplier, 1e-12)
+}
 func TestGatewayServiceRecordUsage_UsesExplicitPricingAtForPeakRate(t *testing.T) {
 	for _, platform := range []string{PlatformAnthropic, PlatformGemini, PlatformGrok, PlatformAntigravity} {
 		t.Run(platform, func(t *testing.T) {
diff --git a/backend/internal/service/gateway_usage_billing.go b/backend/internal/service/gateway_usage_billing.go
index d28a1be19afb..c9671328bec3 100644
--- a/backend/internal/service/gateway_usage_billing.go
+++ b/backend/internal/service/gateway_usage_billing.go
@@ -838,7 +838,7 @@ func (s *GatewayService) recordUsageCore(ctx context.Context, input *recordUsage
 	}
 
 	// 计算费用
-	cost := s.calculateRecordUsageCost(ctx, result, apiKey, billingModel, multiplier, imageMultiplier, opts)
+	cost := s.calculateRecordUsageCost(ctx, result, apiKey, billingModel, multiplier, imageMultiplier, pricingAt, opts)
 	// response_model：按上游成功响应自报的模型计费（渠道显式开启才生效）。
 	// 采纳条件见 responseModelBillingDeclaration + hasIdentifiedResponseModelPricing
 	// + responseModelBillingAdoptable。任一条件不满足都静默回落基线，即开启本模式前的
@@ -850,7 +850,7 @@ func (s *GatewayService) recordUsageCore(ctx context.Context, input *recordUsage
 		result.ImageCount > 0 || result.AudioUsage != nil || result.SearchCount > 0,
 	); responseModel != "" && !strings.EqualFold(responseModel, strings.TrimSpace(billingModel)) {
 		if identified, responseChannelPriced := s.hasIdentifiedResponseModelPricing(ctx, responseModel, apiKey); identified {
-			responseCost := s.calculateRecordUsageCost(ctx, result, apiKey, responseModel, multiplier, imageMultiplier, opts)
+			responseCost := s.calculateRecordUsageCost(ctx, result, apiKey, responseModel, multiplier, imageMultiplier, pricingAt, opts)
 			baselineChannelPriced := s.resolveChannelPricing(ctx, billingModel, apiKey) != nil
 			if responseModelBillingAdoptable(cost, responseCost, baselineChannelPriced, responseChannelPriced) {
 				// billingModel 到此为止只是定价查表的入参，后续流程只消费 cost，
@@ -939,12 +939,13 @@ func (s *GatewayService) calculateRecordUsageCost(
 	billingModel string,
 	multiplier float64,
 	imageMultiplier float64,
+	pricingAt time.Time,
 	opts *recordUsageOpts,
 ) *CostBreakdown {
 	// 图片生成：渠道定价为 token 计费时走 token 路径，否则走图片计费
 	if result.ImageCount > 0 {
 		if resolved := s.resolveChannelPricing(ctx, billingModel, apiKey); resolved != nil && resolved.Mode == BillingModeToken {
-			return s.calculateTokenCost(ctx, result, apiKey, billingModel, multiplier, opts)
+			return s.calculateTokenCost(ctx, result, apiKey, billingModel, multiplier, pricingAt, opts)
 		}
 		return s.calculateImageCost(ctx, result, apiKey, billingModel, imageMultiplier)
 	}
@@ -968,7 +969,7 @@ func (s *GatewayService) calculateRecordUsageCost(
 	}
 
 	// Token 计费；SearchCount 为叠加 surcharge（不替代 token）。
-	tokenCost := s.calculateTokenCost(ctx, result, apiKey, billingModel, multiplier, opts)
+	tokenCost := s.calculateTokenCost(ctx, result, apiKey, billingModel, multiplier, pricingAt, opts)
 	if result.SearchCount > 0 {
 		price := groupSearchPricePer1kFromAPIKey(apiKey)
 		if price != nil && *price == 0 {
@@ -1127,6 +1128,7 @@ func (s *GatewayService) calculateTokenCost(
 	apiKey *APIKey,
 	billingModel string,
 	multiplier float64,
+	pricingAt time.Time,
 	opts *recordUsageOpts,
 ) *CostBreakdown {
 	tokens := UsageTokens{
@@ -1154,6 +1156,7 @@ func (s *GatewayService) calculateTokenCost(
 			Tokens:         tokens,
 			RequestCount:   1,
 			RateMultiplier: multiplier,
+			PricingAt:      pricingAt,
 			Resolver:       s.resolver,
 			Resolved:       resolved,
 		})
@@ -1164,7 +1167,7 @@ func (s *GatewayService) calculateTokenCost(
 		gid := apiKey.Group.ID
 		cost, err = s.billingService.CalculateCostUnified(CostInput{
 			Ctx: ctx, Model: billingModel, GroupID: &gid, Group: apiKey.Group,
-			Tokens: tokens, RequestCount: 1, RateMultiplier: multiplier, Resolver: s.resolver,
+			Tokens: tokens, RequestCount: 1, RateMultiplier: multiplier, PricingAt: pricingAt, Resolver: s.resolver,
 		})
 	} else {
 		cost, err = s.billingService.CalculateCost(billingModel, tokens, multiplier)
diff --git a/backend/internal/service/openai_alpha_search_billing_test.go b/backend/internal/service/openai_alpha_search_billing_test.go
index 99cacc88f7d7..b8c6236d9dba 100644
--- a/backend/internal/service/openai_alpha_search_billing_test.go
+++ b/backend/internal/service/openai_alpha_search_billing_test.go
@@ -5,6 +5,7 @@ package service
 import (
 	"context"
 	"testing"
+	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/config"
 	"github.com/stretchr/testify/require"
@@ -50,7 +51,7 @@ func TestCalculateOpenAIRecordUsageCostWebSearchPerCall(t *testing.T) {
 	// 即使 token 倍率（含高峰，3.0）更高也不采用。
 	apiKey := &APIKey{ID: 1, GroupID: &groupID, Group: &Group{ID: groupID, Platform: PlatformOpenAI}}
 	result := &OpenAIForwardResult{Model: "gpt-5.6-sol", UpstreamModel: "gpt-5.6-sol", WebSearchCalls: 1}
-	cost, err := svc.calculateOpenAIRecordUsageCost(context.Background(), result, apiKey, []string{"gpt-5.6-sol"}, 3.0, 1.0, 1.0, 2.0, UsageTokens{}, "", boolPtr(false))
+	cost, err := svc.calculateOpenAIRecordUsageCost(context.Background(), result, apiKey, []string{"gpt-5.6-sol"}, 3.0, 1.0, 1.0, 2.0, UsageTokens{}, "", boolPtr(false), time.Time{})
 	require.NoError(t, err)
 	require.Equal(t, string(BillingModePerRequest), cost.BillingMode)
 	require.InDelta(t, 0.01, cost.TotalCost, 1e-12)
@@ -58,7 +59,7 @@ func TestCalculateOpenAIRecordUsageCostWebSearchPerCall(t *testing.T) {
 
 	// 分组配置单价 0.005
 	apiKey.Group.WebSearchPricePerCall = float64Ptr(0.005)
-	cost, err = svc.calculateOpenAIRecordUsageCost(context.Background(), result, apiKey, []string{"gpt-5.6-sol"}, 1.0, 1.0, 1.0, 1.0, UsageTokens{}, "", boolPtr(false))
+	cost, err = svc.calculateOpenAIRecordUsageCost(context.Background(), result, apiKey, []string{"gpt-5.6-sol"}, 1.0, 1.0, 1.0, 1.0, UsageTokens{}, "", boolPtr(false), time.Time{})
 	require.NoError(t, err)
 	require.InDelta(t, 0.005, cost.TotalCost, 1e-12)
 	require.InDelta(t, 0.005, cost.ActualCost, 1e-12)
@@ -66,7 +67,7 @@ func TestCalculateOpenAIRecordUsageCostWebSearchPerCall(t *testing.T) {
 	// WebSearchCalls = 0 时不得走按次分支（无定价数据会返回 pricing 错误，
 	// 证明回落到了 token 路径而不是被按次分支吞掉）。
 	result.WebSearchCalls = 0
-	_, err = svc.calculateOpenAIRecordUsageCost(context.Background(), result, apiKey, []string{"gpt-5.6-sol"}, 1.0, 1.0, 1.0, 1.0, UsageTokens{InputTokens: 10}, "", boolPtr(false))
+	_, err = svc.calculateOpenAIRecordUsageCost(context.Background(), result, apiKey, []string{"gpt-5.6-sol"}, 1.0, 1.0, 1.0, 1.0, UsageTokens{InputTokens: 10}, "", boolPtr(false), time.Time{})
 	require.Error(t, err)
 }
 
diff --git a/backend/internal/service/openai_gateway_record_usage_test.go b/backend/internal/service/openai_gateway_record_usage_test.go
index e59b3b205a27..21e5cd409676 100644
--- a/backend/internal/service/openai_gateway_record_usage_test.go
+++ b/backend/internal/service/openai_gateway_record_usage_test.go
@@ -504,6 +504,71 @@ func TestOpenAIGatewayServiceRecordUsage_PeakRateAffectsTokenModeImageOutputToke
 	require.InDelta(t, expectedActual, userRepo.lastAmount, 1e-12)
 }
 
+func TestOpenAIGatewayServiceRecordUsage_TimePricingUsesPricingAt(t *testing.T) {
+	groupID := int64(16)
+	requestStart := time.Date(2024, time.January, 2, 2, 0, 0, 0, time.UTC) // 上海 10:00
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	svc := newOpenAIRecordUsageServiceForTest(usageRepo, userRepo, &openAIRecordUsageSubRepoStub{}, nil)
+	svc.resolver = newOpenAITokenImageChannelPricingResolverWithTimeForTest(t, groupID, "gpt-5.1", &ChannelTimePricing{
+		Timezone: "Asia/Shanghai",
+		Periods:  []ChannelTimePricingPeriod{{StartTime: "09:00", EndTime: "12:00", Multiplier: 2}},
+	})
+
+	err := svc.RecordUsage(context.Background(), &OpenAIRecordUsageInput{
+		Result: &OpenAIForwardResult{
+			RequestID: "openai_time_pricing_request_start",
+			Model:     "gpt-5.1",
+			Usage:     OpenAIUsage{InputTokens: 1000, OutputTokens: 500},
+		},
+		APIKey: &APIKey{ID: 1006, GroupID: i64p(groupID), Group: &Group{
+			ID: groupID, RateMultiplier: 0.8, SubscriptionType: SubscriptionTypeSubscription,
+		}},
+		User:      &User{ID: 2006},
+		Account:   &Account{ID: 3006},
+		PricingAt: requestStart,
+	})
+
+	require.NoError(t, err)
+	require.NotNil(t, usageRepo.lastLog)
+	baseCost := 1000*3e-6 + 500*15e-6
+	require.InDelta(t, baseCost*2, usageRepo.lastLog.TotalCost, 1e-12)
+	require.InDelta(t, baseCost*2*0.8, usageRepo.lastLog.ActualCost, 1e-12)
+	require.InDelta(t, 0.8, usageRepo.lastLog.RateMultiplier, 1e-12)
+}
+
+func TestOpenAIGatewayServiceRecordUsage_TimePricingUsesExplicitPricingAt(t *testing.T) {
+	groupID := int64(17)
+	pricingAt := time.Date(2024, time.January, 2, 0, 0, 0, 0, time.UTC) // 上海 08:00
+	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
+	userRepo := &openAIRecordUsageUserRepoStub{}
+	svc := newOpenAIRecordUsageServiceForTest(usageRepo, userRepo, &openAIRecordUsageSubRepoStub{}, nil)
+	svc.resolver = newOpenAITokenImageChannelPricingResolverWithTimeForTest(t, groupID, "gpt-5.1", &ChannelTimePricing{
+		Timezone: "Asia/Shanghai",
+		Periods:  []ChannelTimePricingPeriod{{StartTime: "09:00", EndTime: "12:00", Multiplier: 2}},
+	})
+
+	err := svc.RecordUsage(context.Background(), &OpenAIRecordUsageInput{
+		Result: &OpenAIForwardResult{
+			RequestID: "openai_time_pricing_explicit",
+			Model:     "gpt-5.1",
+			Usage:     OpenAIUsage{InputTokens: 1000, OutputTokens: 500},
+		},
+		APIKey: &APIKey{ID: 1007, GroupID: i64p(groupID), Group: &Group{
+			ID: groupID, RateMultiplier: 0.8, SubscriptionType: SubscriptionTypeSubscription,
+		}},
+		User:      &User{ID: 2007},
+		Account:   &Account{ID: 3007},
+		PricingAt: pricingAt,
+	})
+
+	require.NoError(t, err)
+	require.NotNil(t, usageRepo.lastLog)
+	baseCost := 1000*3e-6 + 500*15e-6
+	require.InDelta(t, baseCost, usageRepo.lastLog.TotalCost, 1e-12)
+	require.InDelta(t, baseCost*0.8, usageRepo.lastLog.ActualCost, 1e-12)
+	require.InDelta(t, 0.8, usageRepo.lastLog.RateMultiplier, 1e-12)
+}
 func TestOpenAIGatewayServiceRecordUsage_IncludesEndpointMetadata(t *testing.T) {
 	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
 	userRepo := &openAIRecordUsageUserRepoStub{}
@@ -2640,6 +2705,20 @@ func newOpenAITokenImageChannelPricingResolverForTest(t *testing.T, groupID int6
 	return NewModelPricingResolver(cs, NewBillingService(&config.Config{}, nil))
 }
 
+func newOpenAITokenImageChannelPricingResolverWithTimeForTest(
+	t *testing.T,
+	groupID int64,
+	model string,
+	timePricing *ChannelTimePricing,
+) *ModelPricingResolver {
+	t.Helper()
+	resolver := newOpenAITokenImageChannelPricingResolverForTest(t, groupID, model)
+	cached, ok := resolver.channelService.cache.Load().(*channelCache)
+	require.True(t, ok)
+	cached.pricingByGroupModel[channelModelKey{groupID: groupID, model: model}].TimePricing = timePricing
+	return resolver
+}
+
 type openAIMediaPriceGroupRepoStub struct {
 	GroupRepository
 	group *Group
@@ -2668,6 +2747,7 @@ func TestGatewayServiceCalculateRecordUsageCost_ChannelImageBillingUsesImageCoun
 		"gemini-image",
 		0.15,
 		1.0,
+		time.Time{},
 		nil,
 	)
 
@@ -2707,6 +2787,7 @@ func TestGatewayServiceCalculateRecordUsageCost_ChannelImageBillingUsesSizeTier(
 		"gemini-image",
 		1.0,
 		1.0,
+		time.Time{},
 		nil,
 	)
 
@@ -2739,6 +2820,7 @@ func TestGatewayServiceCalculateRecordUsageCost_GroupImagePriceOverridesChannelI
 		"gemini-image",
 		1.0,
 		1.0,
+		time.Time{},
 		nil,
 	)
 
@@ -2802,6 +2884,7 @@ func TestGatewayServiceCalculateRecordUsageCost_ChannelImageBillingNormalizesMis
 		"gemini-image",
 		1.0,
 		1.0,
+		time.Time{},
 		nil,
 	)
 
diff --git a/backend/internal/service/openai_gateway_search_surcharge_test.go b/backend/internal/service/openai_gateway_search_surcharge_test.go
index 2633a5b930f1..d2d069ff5c26 100644
--- a/backend/internal/service/openai_gateway_search_surcharge_test.go
+++ b/backend/internal/service/openai_gateway_search_surcharge_test.go
@@ -5,6 +5,7 @@ package service
 import (
 	"context"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/require"
 )
@@ -37,6 +38,7 @@ func TestCalculateOpenAIRecordUsageCost_SearchIsAdditiveToTokens(t *testing.T) {
 		UsageTokens{InputTokens: 1000, OutputTokens: 500},
 		"",
 		boolPtr(false),
+		time.Time{},
 	)
 	require.NoError(t, err)
 	require.NotNil(t, cost)
@@ -67,6 +69,7 @@ func TestCalculateOpenAIRecordUsageCost_SearchOnlyWhenNoTokenPricing(t *testing.
 		UsageTokens{},
 		"",
 		boolPtr(false),
+		time.Time{},
 	)
 	require.NoError(t, err)
 	require.NotNil(t, cost)
@@ -112,6 +115,7 @@ func TestCalculateOpenAIRecordUsageCost_TokenPricingErrorNotSwallowedBySearch(t
 		UsageTokens{InputTokens: 1000, OutputTokens: 500},
 		"",
 		boolPtr(false),
+		time.Time{},
 	)
 	require.Error(t, err)
 	require.Nil(t, cost)
diff --git a/backend/internal/service/openai_gateway_usage.go b/backend/internal/service/openai_gateway_usage.go
index c55bcdbeac89..a8b3c0fc0f5d 100644
--- a/backend/internal/service/openai_gateway_usage.go
+++ b/backend/internal/service/openai_gateway_usage.go
@@ -177,7 +177,8 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 	// 变价）；未装配 PricingAt 的路径回退记录时刻，保持既有行为。不并入上面的
 	// Resolve，以免污染 user:group 倍率缓存。
 	baseMultiplier := multiplier
-	multiplier, imageMultiplier := computePeakAwareMultipliers(apiKey, baseMultiplier, openAIUsagePricingAt(input))
+	pricingAt := openAIUsagePricingAt(input)
+	multiplier, imageMultiplier := computePeakAwareMultipliers(apiKey, baseMultiplier, pricingAt)
 	videoMultiplier := resolveVideoRateMultiplier(apiKey, baseMultiplier)
 
 	var cost *CostBreakdown
@@ -224,6 +225,7 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 		tokens,
 		serviceTier,
 		longContextBillingGate,
+		pricingAt,
 	)
 	if err != nil {
 		if !isUsagePricingUnavailableError(err) {
@@ -256,7 +258,7 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 			responseModels := usageBillingModelCandidates(responseModel)
 			responseCost, responseErr := s.calculateOpenAIRecordUsageCost(
 				ctx, result, apiKey, responseModels, multiplier, imageMultiplier,
-				videoMultiplier, baseMultiplier, tokens, serviceTier, longContextBillingGate,
+				videoMultiplier, baseMultiplier, tokens, serviceTier, longContextBillingGate, pricingAt,
 			)
 			// 基线定价源以 baselineBillingModel 为准：它正是 calculateOpenAIRecordUsageCost
 			// 内部做渠道定价判断时使用的模型，且"首候选有渠道价"必然意味着首候选就是实际
@@ -505,6 +507,7 @@ func (s *OpenAIGatewayService) calculateOpenAIRecordUsageCost(
 	tokens UsageTokens,
 	serviceTier string,
 	longContextBillingGate *bool,
+	pricingAt time.Time,
 ) (*CostBreakdown, error) {
 	billingModel := firstUsageBillingModel(billingModels)
 	if result != nil && result.WebSearchCalls > 0 {
@@ -554,6 +557,7 @@ func (s *OpenAIGatewayService) calculateOpenAIRecordUsageCost(
 				apiKey,
 				candidate,
 				multiplier,
+				pricingAt,
 				tokens,
 				serviceTier,
 				longContextBillingGate,
@@ -643,6 +647,7 @@ func (s *OpenAIGatewayService) calculateOpenAIRecordUsageTokenCost(
 	apiKey *APIKey,
 	billingModel string,
 	multiplier float64,
+	pricingAt time.Time,
 	tokens UsageTokens,
 	serviceTier string,
 	longContextBillingGate *bool,
@@ -651,7 +656,7 @@ func (s *OpenAIGatewayService) calculateOpenAIRecordUsageTokenCost(
 		gid := apiKey.Group.ID
 		return s.billingService.CalculateCostUnified(CostInput{
 			Ctx: ctx, Model: billingModel, GroupID: &gid, Group: apiKey.Group,
-			Tokens: tokens, RequestCount: 1, RateMultiplier: multiplier,
+			Tokens: tokens, RequestCount: 1, RateMultiplier: multiplier, PricingAt: pricingAt,
 			ServiceTier: serviceTier, Resolver: s.resolver,
 			LongContextBillingEnabled: longContextBillingGate,
 		})
diff --git a/backend/internal/service/openai_ws_forwarder.go b/backend/internal/service/openai_ws_forwarder.go
index 51898060acae..1928c1d6e4a8 100644
--- a/backend/internal/service/openai_ws_forwarder.go
+++ b/backend/internal/service/openai_ws_forwarder.go
@@ -215,10 +215,13 @@ type OpenAIWSIngressHooks struct {
 	// before channel or account mapping. Ingress modes preserve it for usage
 	// attribution while MapRequestModel determines the upstream model.
 	InitialRequestModel string
+	// InitialTurnStartedAt freezes when the first response.create was accepted.
+	InitialTurnStartedAt time.Time
 	// MaxReasoningEffort limits explicit reasoning effort values for this WS session.
 	MaxReasoningEffort string
 	// ReasoningEffortMappings rewrites explicit effort values for this WS session.
 	ReasoningEffortMappings []ReasoningEffortMapping
+	TurnStarted             func(turn int, startedAt time.Time)
 	BeforeTurn              func(turn int) error
 	BeforeRequest           func(turn int, payload []byte, originalModel string) error
 	// MapRequestModel resolves the current turn's client model to the model
diff --git a/backend/internal/service/openai_ws_forwarder_ingress.go b/backend/internal/service/openai_ws_forwarder_ingress.go
index ce7e8f038666..bdd59f5e1cb1 100644
--- a/backend/internal/service/openai_ws_forwarder_ingress.go
+++ b/backend/internal/service/openai_ws_forwarder_ingress.go
@@ -92,11 +92,9 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient(
 			if wsDecision.Transport != OpenAIUpstreamTransportResponsesWebsocketV2 {
 				return fmt.Errorf("websocket ingress requires ws_v2 transport, got=%s", wsDecision.Transport)
 			}
-			// 注意：透传 relay 只回调 hooks.AfterTurn，没有 turn 起始回调，
-			// 因此下面这条路径永远不会触发 hooks.BeforeTurn——分组利润控制的
-			// turn 级复核与 turn 级 pricingAt 冻结都不覆盖透传 ingress，
-			// 只有建连时的准入门生效。handler 侧据此把 turn 定价留作零值，
-			// 由 RecordUsage 回退到记录时刻（见 openAIWSTurnPricing 注释）。
+			// 透传 relay 通过 TurnStarted 记录每个 turn 的开始时刻，但不触发
+			// BeforeTurn；因此仍只有建连时的利润准入门，没有 turn 级复核。
+			// handler 计费在 turn 定价未冻结时回退到对应的 turn 开始时刻。
 			return s.proxyResponsesWebSocketV2Passthrough(
 				ctx,
 				c,
diff --git a/backend/internal/service/openai_ws_passthrough_turn_pricing_test.go b/backend/internal/service/openai_ws_passthrough_turn_pricing_test.go
index bb6e5ebde604..16be2bd4c62d 100644
--- a/backend/internal/service/openai_ws_passthrough_turn_pricing_test.go
+++ b/backend/internal/service/openai_ws_passthrough_turn_pricing_test.go
@@ -60,18 +60,13 @@ func startPassthroughHookRecordingServer(
 	return server, serverErr
 }
 
-// TestPassthroughIngressNeverCallsBeforeTurn 钉死 ws_v2 透传 ingress 与 handler
-// 侧 turn 定价的耦合：透传 relay 只回调 AfterTurn，没有任何 turn 起始回调，
-// 因此 hooks.BeforeTurn 永远不会触发。
+// TestPassthroughIngressReportsTurnStartedBeforeAfterTurnWithoutBeforeTurn 钉死
+// ws_v2 透传 ingress 与 handler 侧 turn 定价的耦合：透传 relay 不触发
+// BeforeTurn，但会在每个 AfterTurn 前通过 TurnStarted 报告同一 turn 的开始时刻。
 //
-// handler 依赖这一点：openAIWSTurnPricing 零值起步，透传连接的每个 turn 都拿
-// 不到冻结的 pricingAt，RecordUsage 回退到记录时刻——与引入分组利润控制前的
-// 基线一致。若把 turn 定价初始化成建连时刻，透传连接的所有 turn 就会被钉死在
-// 建连时的高峰因子，客户端峰前建连保活即可全程按谷价结算。
-//
-// 若本断言因为透传补齐了 turn 起始回调而失败：这是好事，请同步复核
-// openAIWSTurnPricing 的零值语义与透传路径的 turn 级利润复核。
-func TestPassthroughIngressNeverCallsBeforeTurn(t *testing.T) {
+// handler 的 recordTurnStart 保存该时刻，AfterTurn 再用 currentOr(turnStart)
+// 作为计费 PricingAt；不触发 BeforeTurn 也意味着透传仍没有 turn 级利润复核。
+func TestPassthroughIngressReportsTurnStartedBeforeAfterTurnWithoutBeforeTurn(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	controlCtx, cancelControl := context.WithCancelCause(context.Background())
 	defer cancelControl(context.Canceled)
@@ -81,17 +76,29 @@ func TestPassthroughIngressNeverCallsBeforeTurn(t *testing.T) {
 
 	var hooksMu sync.Mutex
 	beforeTurnCalls := 0
-	afterTurnCalls := 0
+	expectedTurnStartedAt := time.Date(2026, time.August, 17, 9, 59, 59, 0, time.UTC)
+	type hookEvent struct {
+		name      string
+		turn      int
+		startedAt time.Time
+	}
+	var hookEvents []hookEvent
 	hooks := &OpenAIWSIngressHooks{
+		InitialTurnStartedAt: expectedTurnStartedAt,
+		TurnStarted: func(turn int, startedAt time.Time) {
+			hooksMu.Lock()
+			hookEvents = append(hookEvents, hookEvent{name: "TurnStarted", turn: turn, startedAt: startedAt})
+			hooksMu.Unlock()
+		},
 		BeforeTurn: func(int) error {
 			hooksMu.Lock()
 			beforeTurnCalls++
 			hooksMu.Unlock()
 			return nil
 		},
-		AfterTurn: func(int, *OpenAIForwardResult, error) {
+		AfterTurn: func(turn int, _ *OpenAIForwardResult, _ error) {
 			hooksMu.Lock()
-			afterTurnCalls++
+			hookEvents = append(hookEvents, hookEvent{name: "AfterTurn", turn: turn})
 			hooksMu.Unlock()
 		},
 	}
@@ -120,9 +127,109 @@ func TestPassthroughIngressNeverCallsBeforeTurn(t *testing.T) {
 	}
 
 	hooksMu.Lock()
-	gotBefore, gotAfter := beforeTurnCalls, afterTurnCalls
+	gotBefore := beforeTurnCalls
+	gotEvents := append([]hookEvent(nil), hookEvents...)
 	hooksMu.Unlock()
 
-	require.Zero(t, gotBefore, "透传 ingress 没有 turn 起始回调，BeforeTurn 不应被调用")
-	require.Positive(t, gotAfter, "透传 ingress 仍应回调 AfterTurn 提交用量")
+	require.Zero(t, gotBefore, "透传 ingress 不应调用 BeforeTurn")
+	require.GreaterOrEqual(t, len(gotEvents), 2, "透传 ingress 应报告 TurnStarted 和 AfterTurn")
+	require.Equal(t, "TurnStarted", gotEvents[0].name)
+	require.Equal(t, expectedTurnStartedAt, gotEvents[0].startedAt, "TurnStarted 必须携带入口冻结的首轮开始时刻")
+	require.Equal(t, "AfterTurn", gotEvents[1].name)
+	require.Equal(t, gotEvents[0].turn, gotEvents[1].turn, "TurnStarted 后应提交同一 turn 的 AfterTurn")
+}
+
+func TestPassthroughIngressFreezesSubsequentTurnBeforeRequestPolicy(t *testing.T) {
+	testPassthroughIngressFreezesSubsequentTurnBeforeRequestPolicy(t, coderws.MessageText)
+}
+
+func TestPassthroughIngressFreezesBinarySubsequentTurnBeforeRequestPolicy(t *testing.T) {
+	testPassthroughIngressFreezesSubsequentTurnBeforeRequestPolicy(t, coderws.MessageBinary)
+}
+
+func testPassthroughIngressFreezesSubsequentTurnBeforeRequestPolicy(t *testing.T, secondMessageType coderws.MessageType) {
+	t.Helper()
+	gin.SetMode(gin.TestMode)
+	controlCtx, cancelControl := context.WithCancelCause(context.Background())
+	defer cancelControl(context.Canceled)
+
+	upstream := newStagedPassthroughConn()
+	upstream.Send(`{"type":"response.completed","response":{"id":"resp_first","model":"gpt-5.1","usage":{"input_tokens":1,"output_tokens":1}}}`)
+
+	type turnStart struct {
+		turn      int
+		startedAt time.Time
+	}
+	turnStarts := make(chan turnStart, 2)
+	beforeRequestEntered := make(chan time.Time, 1)
+	releaseBeforeRequest := make(chan struct{})
+	hooks := &OpenAIWSIngressHooks{
+		InitialTurnStartedAt: time.Now(),
+		TurnStarted: func(turn int, startedAt time.Time) {
+			turnStarts <- turnStart{turn: turn, startedAt: startedAt}
+		},
+		BeforeRequest: func(turn int, _ []byte, _ string) error {
+			if turn == 2 {
+				beforeRequestEntered <- time.Now()
+				<-releaseBeforeRequest
+			}
+			return nil
+		},
+	}
+
+	server, serverErr := startPassthroughHookRecordingServer(
+		t,
+		controlCtx,
+		newPassthroughLifecycleService(passthroughLifecycleConfig(), upstream),
+		passthroughLifecycleAccount(),
+		hooks,
+	)
+	defer server.Close()
+	clientConn := dialPassthroughLifecycleClient(t, server)
+	defer func() { _ = clientConn.CloseNow() }()
+
+	require.Equal(t, "response.create", gjson.GetBytes(requirePassthroughUpstreamWrite(t, upstream, 3*time.Second), "type").String())
+	firstCompleted, err := readPassthroughLifecycleFrame(t, clientConn, 3*time.Second)
+	require.NoError(t, err)
+	require.Equal(t, "resp_first", gjson.GetBytes(firstCompleted, "response.id").String())
+	select {
+	case first := <-turnStarts:
+		require.Equal(t, 1, first.turn)
+	case <-time.After(time.Second):
+		t.Fatal("first turn start was not reported")
+	}
+
+	writeCtx, cancelWrite := context.WithTimeout(context.Background(), 3*time.Second)
+	err = clientConn.Write(writeCtx, secondMessageType, []byte(`{"type":"response.create","model":"gpt-5.1","previous_response_id":"resp_first"}`))
+	cancelWrite()
+	require.NoError(t, err)
+
+	var policyEnteredAt time.Time
+	select {
+	case policyEnteredAt = <-beforeRequestEntered:
+	case <-time.After(time.Second):
+		t.Fatal("second turn did not enter BeforeRequest")
+	}
+	close(releaseBeforeRequest)
+	require.Equal(t, "response.create", gjson.GetBytes(requirePassthroughUpstreamWrite(t, upstream, 3*time.Second), "type").String())
+	upstream.Send(`{"type":"response.completed","response":{"id":"resp_second","model":"gpt-5.1","usage":{"input_tokens":1,"output_tokens":1}}}`)
+	secondCompleted, err := readPassthroughLifecycleFrame(t, clientConn, 3*time.Second)
+	require.NoError(t, err)
+	require.Equal(t, "resp_second", gjson.GetBytes(secondCompleted, "response.id").String())
+
+	select {
+	case second := <-turnStarts:
+		require.Equal(t, 2, second.turn)
+		require.False(t, second.startedAt.After(policyEnteredAt), "第二轮开始时刻必须在 BeforeRequest 策略执行前冻结")
+	case <-time.After(time.Second):
+		t.Fatal("second turn start was not reported")
+	}
+
+	_ = clientConn.CloseNow()
+	cancelControl(context.Canceled)
+	select {
+	case <-serverErr:
+	case <-time.After(3 * time.Second):
+		t.Fatal("passthrough ingress did not exit")
+	}
 }
diff --git a/backend/internal/service/openai_ws_v2/passthrough_relay.go b/backend/internal/service/openai_ws_v2/passthrough_relay.go
index 626faf891ae7..abc94e8f1cdb 100644
--- a/backend/internal/service/openai_ws_v2/passthrough_relay.go
+++ b/backend/internal/service/openai_ws_v2/passthrough_relay.go
@@ -50,6 +50,7 @@ type RelayTurnResult struct {
 	Usage                 Usage
 	RequestID             string
 	TerminalEventType     string
+	StartedAt             time.Time
 	Duration              time.Duration
 	FirstTokenMs          *int
 }
@@ -65,6 +66,8 @@ type RelayOptions struct {
 	WriteTimeout                    time.Duration
 	IdleTimeout                     time.Duration
 	UpstreamDrainTimeout            time.Duration
+	FirstTurnStartedAt              time.Time
+	TakeNextTurnStartedAt           func() time.Time
 	FirstMessageType                coderws.MessageType
 	FirstMessageSent                bool
 	StartClientAfterFirstDownstream bool
@@ -93,6 +96,7 @@ type relayState struct {
 	usage             Usage
 	requestModelMu    sync.RWMutex
 	requestModel      string
+	pendingTurnStart  atomic.Pointer[time.Time]
 	lastResponseID    string
 	lastResponseModel string
 	responseConflict  bool
@@ -114,6 +118,7 @@ type observedUpstreamEvent struct {
 	eventType        string
 	responseID       string
 	usage            Usage
+	startedAt        time.Time
 	responseModel    string
 	responseConflict bool
 	duration         time.Duration
@@ -161,6 +166,13 @@ func Relay(
 	}
 	startAt := nowFn()
 	state := &relayState{requestModel: result.RequestModel}
+	if isClientResponseCreateFrame(firstMessageType, firstClientMessage) {
+		firstTurnStartedAt := options.FirstTurnStartedAt
+		if firstTurnStartedAt.IsZero() {
+			firstTurnStartedAt = startAt
+		}
+		state.setPendingTurnStartedAt(firstTurnStartedAt)
+	}
 	onTrace := options.OnTrace
 
 	relayCtx, relayCancel := context.WithCancel(ctx)
@@ -178,8 +190,16 @@ func Relay(
 		return upstreamConn.WriteFrame(writeCtx, msgType, payload)
 	}
 	writeClientFrameUpstream := func(msgType coderws.MessageType, payload []byte) error {
-		if msgType == coderws.MessageText && strings.TrimSpace(gjson.GetBytes(payload, "type").String()) == "response.create" {
+		if isClientResponseCreateFrame(msgType, payload) {
 			state.setRequestModel(strings.TrimSpace(gjson.GetBytes(payload, "model").String()))
+			turnStartedAt := time.Time{}
+			if options.TakeNextTurnStartedAt != nil {
+				turnStartedAt = options.TakeNextTurnStartedAt()
+			}
+			if turnStartedAt.IsZero() {
+				turnStartedAt = nowFn()
+			}
+			state.setPendingTurnStartedAt(turnStartedAt)
 		}
 		return writeUpstream(msgType, payload)
 	}
@@ -412,6 +432,13 @@ func Relay(
 	return result, nil
 }
 
+func isClientResponseCreateFrame(msgType coderws.MessageType, payload []byte) bool {
+	if msgType != coderws.MessageText && msgType != coderws.MessageBinary {
+		return false
+	}
+	return strings.TrimSpace(gjson.GetBytes(payload, "type").String()) == "response.create"
+}
+
 func runClientToUpstream(
 	ctx context.Context,
 	clientConn FrameConn,
@@ -724,6 +751,7 @@ func observeUpstreamMessage(
 			if duration < 0 {
 				duration = 0
 			}
+			observed.startedAt = turnTiming.startAt
 			observed.duration = duration
 			observed.firstToken = openAIWSRelayCloneIntPtr(turnTiming.firstTokenMs)
 		}
@@ -754,6 +782,7 @@ func emitTurnComplete(
 		Usage:                 observed.usage,
 		RequestID:             responseID,
 		TerminalEventType:     observed.eventType,
+		StartedAt:             observed.startedAt,
 		Duration:              observed.duration,
 		FirstTokenMs:          openAIWSRelayCloneIntPtr(observed.firstToken),
 	})
@@ -815,7 +844,11 @@ func openAIWSRelayGetOrInitTurnTiming(state *relayState, responseID string, now
 	}
 	timing, ok := state.turnTimingByID[responseID]
 	if !ok || timing == nil || timing.startAt.IsZero() {
-		timing = &relayTurnTiming{startAt: now}
+		startAt := state.consumePendingTurnStartedAt()
+		if startAt.IsZero() {
+			startAt = now
+		}
+		timing = &relayTurnTiming{startAt: startAt}
 		state.turnTimingByID[responseID] = timing
 		state.activeTurn = timing
 		return timing
@@ -823,6 +856,25 @@ func openAIWSRelayGetOrInitTurnTiming(state *relayState, responseID string, now
 	return timing
 }
 
+func (s *relayState) setPendingTurnStartedAt(startedAt time.Time) {
+	if s == nil || startedAt.IsZero() {
+		return
+	}
+	startedAtCopy := startedAt
+	s.pendingTurnStart.Store(&startedAtCopy)
+}
+
+func (s *relayState) consumePendingTurnStartedAt() time.Time {
+	if s == nil {
+		return time.Time{}
+	}
+	startedAt := s.pendingTurnStart.Swap(nil)
+	if startedAt == nil {
+		return time.Time{}
+	}
+	return *startedAt
+}
+
 func openAIWSRelayDeleteTurnTiming(state *relayState, responseID string) (relayTurnTiming, bool) {
 	if state == nil || state.turnTimingByID == nil {
 		return relayTurnTiming{}, false
diff --git a/backend/internal/service/openai_ws_v2/passthrough_relay_test.go b/backend/internal/service/openai_ws_v2/passthrough_relay_test.go
index c41e7d293b04..12e550669802 100644
--- a/backend/internal/service/openai_ws_v2/passthrough_relay_test.go
+++ b/backend/internal/service/openai_ws_v2/passthrough_relay_test.go
@@ -565,6 +565,143 @@ func TestRelay_OnTurnComplete_ProvidesTurnMetrics(t *testing.T) {
 	require.Greater(t, result.Duration.Milliseconds(), int64(0))
 }
 
+func TestRelay_OnTurnComplete_UsesResponseCreateTimeAcrossPricingBoundary(t *testing.T) {
+	t.Parallel()
+
+	clientConn := newPassthroughTestFrameConn(nil, false)
+	upstreamConn := newPassthroughTestFrameConn([]passthroughTestFrame{
+		{
+			msgType: coderws.MessageText,
+			payload: []byte(`{"type":"response.completed","response":{"id":"resp_boundary","usage":{"input_tokens":1,"output_tokens":1}}}`),
+		},
+	}, true)
+
+	responseCreateAt := time.Date(2026, time.August, 17, 9, 59, 59, 0, time.UTC)
+	upstreamResponseAt := responseCreateAt.Add(time.Second)
+	var nowCalls atomic.Int64
+	nowFn := func() time.Time {
+		if nowCalls.Add(1) == 1 {
+			return responseCreateAt
+		}
+		return upstreamResponseAt
+	}
+
+	var turn RelayTurnResult
+	_, relayExit := Relay(
+		context.Background(),
+		clientConn,
+		upstreamConn,
+		[]byte(`{"type":"response.create","model":"gpt-5.3-codex","input":[]}`),
+		RelayOptions{
+			Now:            nowFn,
+			OnTurnComplete: func(current RelayTurnResult) { turn = current },
+		},
+	)
+
+	require.Nil(t, relayExit)
+	require.Equal(t, responseCreateAt, turn.StartedAt)
+}
+
+func TestRelay_OnTurnComplete_UsesExplicitFirstTurnStartedAt(t *testing.T) {
+	t.Parallel()
+
+	clientConn := newPassthroughTestFrameConn(nil, false)
+	upstreamConn := newPassthroughTestFrameConn([]passthroughTestFrame{
+		{
+			msgType: coderws.MessageText,
+			payload: []byte(`{"type":"response.completed","response":{"id":"resp_initial_boundary","usage":{"input_tokens":1,"output_tokens":1}}}`),
+		},
+	}, true)
+
+	responseCreateAt := time.Date(2026, time.August, 17, 9, 59, 59, 0, time.UTC)
+	relayStartedAt := responseCreateAt.Add(time.Second)
+	var turn RelayTurnResult
+	_, relayExit := Relay(
+		context.Background(),
+		clientConn,
+		upstreamConn,
+		[]byte(`{"type":"response.create","model":"gpt-5.3-codex","input":[]}`),
+		RelayOptions{
+			FirstTurnStartedAt: responseCreateAt,
+			Now:                func() time.Time { return relayStartedAt },
+			OnTurnComplete:     func(current RelayTurnResult) { turn = current },
+		},
+	)
+
+	require.Nil(t, relayExit)
+	require.Equal(t, responseCreateAt, turn.StartedAt)
+}
+
+func TestRelay_OnTurnComplete_UsesSubsequentResponseCreateTimeAcrossPricingBoundary(t *testing.T) {
+	t.Parallel()
+
+	clientConn := newPassthroughTestFrameConn(nil, false)
+	upstreamConn := newPassthroughTestFrameConn(nil, false)
+	firstTurnAt := time.Date(2026, time.August, 17, 9, 0, 0, 0, time.UTC)
+	secondTurnAt := time.Date(2026, time.August, 17, 9, 59, 59, 0, time.UTC)
+	secondResponseAt := secondTurnAt.Add(time.Second)
+	var clock atomic.Int64
+	clock.Store(firstTurnAt.UnixNano())
+
+	turns := make(chan RelayTurnResult, 2)
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		_, _ = Relay(
+			ctx,
+			clientConn,
+			upstreamConn,
+			[]byte(`{"type":"response.create","model":"gpt-5.3-codex","input":[]}`),
+			RelayOptions{
+				Now: func() time.Time { return time.Unix(0, clock.Load()).UTC() },
+				OnTurnComplete: func(current RelayTurnResult) {
+					turns <- current
+				},
+			},
+		)
+	}()
+
+	require.Eventually(t, func() bool { return len(upstreamConn.Writes()) == 1 }, time.Second, time.Millisecond)
+	upstreamConn.readCh <- passthroughTestFrame{
+		msgType: coderws.MessageText,
+		payload: []byte(`{"type":"response.completed","response":{"id":"resp_first","usage":{"input_tokens":1,"output_tokens":1}}}`),
+	}
+	select {
+	case <-turns:
+	case <-time.After(time.Second):
+		t.Fatal("first turn did not complete")
+	}
+
+	clock.Store(secondTurnAt.UnixNano())
+	clientConn.readCh <- passthroughTestFrame{
+		msgType: coderws.MessageText,
+		payload: []byte(`{"type":"response.create","model":"gpt-5.3-codex","input":[]}`),
+	}
+	require.Eventually(t, func() bool { return len(upstreamConn.Writes()) == 2 }, time.Second, time.Millisecond)
+	clock.Store(secondResponseAt.UnixNano())
+	upstreamConn.readCh <- passthroughTestFrame{
+		msgType: coderws.MessageText,
+		payload: []byte(`{"type":"response.completed","response":{"id":"resp_second","usage":{"input_tokens":1,"output_tokens":1}}}`),
+	}
+
+	var secondTurn RelayTurnResult
+	select {
+	case secondTurn = <-turns:
+	case <-time.After(time.Second):
+		t.Fatal("second turn did not complete")
+	}
+	require.Equal(t, secondTurnAt, secondTurn.StartedAt)
+
+	cancel()
+	select {
+	case <-done:
+	case <-time.After(time.Second):
+		t.Fatal("relay did not stop after cancellation")
+	}
+}
+
 func TestRelay_BinaryFramePassthrough(t *testing.T) {
 	t.Parallel()
 
diff --git a/backend/internal/service/openai_ws_v2_passthrough_adapter.go b/backend/internal/service/openai_ws_v2_passthrough_adapter.go
index 775e26980ce0..df094a336743 100644
--- a/backend/internal/service/openai_ws_v2_passthrough_adapter.go
+++ b/backend/internal/service/openai_ws_v2_passthrough_adapter.go
@@ -915,6 +915,7 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 
 	completedTurns := atomic.Int32{}
 	turnLifecycle := newOpenAIWSPassthroughTurnLifecycle(true)
+	var acceptedTurnStartedAt atomic.Pointer[time.Time]
 	clientFrameConn := &openAIWSClientFrameConn{
 		conn:                 clientConn,
 		controlCtx:           ctx,
@@ -936,13 +937,15 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 		// capturedSessionModel 的读写都发生在该 goroutine 内，因此无需
 		// 加锁/原子化。
 		filter: func(msgType coderws.MessageType, payload []byte) (out []byte, blocked *OpenAIFastBlockedError, filterErr error) {
-			if msgType != coderws.MessageText {
+			if msgType != coderws.MessageText && msgType != coderws.MessageBinary {
 				return payload, nil, nil
 			}
 			eventType := strings.TrimSpace(gjson.GetBytes(payload, "type").String())
 			isResponseCreate := eventType == "response.create"
+			responseCreateAt := time.Time{}
 			acceptedTurn := false
 			if isResponseCreate {
+				responseCreateAt = time.Now()
 				if !turnLifecycle.beginResponseCreate(clientFrameConn.markTurnStarted) {
 					err := errors.New("overlapping response.create is not supported")
 					return payload, nil, NewOpenAIWSClientCloseError(coderws.StatusPolicyViolation, err.Error(), err)
@@ -1035,6 +1038,8 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 			//     service_tier 时按 default 处理，billing 应如实反映。
 			if policyErr == nil && blocked == nil && isResponseCreate {
 				usageMeta.updateFromResponseCreate(out, model, requestModelForThisFrame)
+				responseCreateAtCopy := responseCreateAt
+				acceptedTurnStartedAt.Store(&responseCreateAtCopy)
 				acceptedTurn = true
 			}
 			return out, blocked, policyErr
@@ -1072,7 +1077,7 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 			if readErr != nil {
 				return msgType, payload, readErr
 			}
-			if msgType == coderws.MessageText && strings.TrimSpace(gjson.GetBytes(payload, "type").String()) == "response.create" {
+			if (msgType == coderws.MessageText || msgType == coderws.MessageBinary) && strings.TrimSpace(gjson.GetBytes(payload, "type").String()) == "response.create" {
 				return msgType, payload, nil
 			}
 			if writeErr := upstreamFrameConn.WriteFrame(readCtx, msgType, payload); writeErr != nil {
@@ -1081,13 +1086,25 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 		}
 	}
 
+	firstTurnStartedAt := time.Time{}
+	if hooks != nil {
+		firstTurnStartedAt = hooks.InitialTurnStartedAt
+	}
 	relayResult, relayExit := openaiwsv2.RunEntry(openaiwsv2.EntryInput{
 		Ctx:                ctx,
 		ClientConn:         policyClientConn,
 		UpstreamConn:       relayUpstreamFrameConn,
 		FirstClientMessage: firstClientMessage,
 		Options: openaiwsv2.RelayOptions{
-			WriteTimeout: s.openAIWSWriteTimeout(),
+			WriteTimeout:       s.openAIWSWriteTimeout(),
+			FirstTurnStartedAt: firstTurnStartedAt,
+			TakeNextTurnStartedAt: func() time.Time {
+				startedAt := acceptedTurnStartedAt.Swap(nil)
+				if startedAt == nil {
+					return time.Time{}
+				}
+				return *startedAt
+			},
 			// Passthrough idle is enforced only after a completed turn by
 			// clientFrameConn. The relay-wide activity watchdog would also
 			// terminate a healthy active upstream turn.
@@ -1105,6 +1122,9 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 			},
 			OnTurnComplete: func(turn openaiwsv2.RelayTurnResult) {
 				turnNo := int(completedTurns.Add(1))
+				if hooks != nil && hooks.TurnStarted != nil && !turn.StartedAt.IsZero() {
+					hooks.TurnStarted(turnNo, turn.StartedAt)
+				}
 				turnRequestModel, turnUpstreamModel := usageMeta.turnModels(turn.RequestModel)
 				turnResult := &OpenAIForwardResult{
 					RequestID: turn.RequestID,
@@ -1265,6 +1285,9 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 		)
 		// 正常路径按 terminal 事件逐 turn 已回调；仅在零 turn 场景兜底回调一次。
 		if turnCount == 0 && hooks != nil && hooks.AfterTurn != nil {
+			if hooks.TurnStarted != nil {
+				hooks.TurnStarted(1, time.Now().Add(-result.Duration))
+			}
 			hooks.AfterTurn(1, result, nil)
 		}
 		return nil
@@ -1331,6 +1354,9 @@ func (s *OpenAIGatewayService) proxyResponsesWebSocketV2Passthrough(
 		relayExit.WroteDownstream,
 	)
 	if hooks != nil && hooks.AfterTurn != nil {
+		if hooks.TurnStarted != nil {
+			hooks.TurnStarted(turnCount+1, time.Now().Add(-result.Duration))
+		}
 		hooks.AfterTurn(turnCount+1, nil, turnErr)
 	}
 	return turnErr
diff --git a/backend/migrations/225_channel_model_time_pricing.sql b/backend/migrations/225_channel_model_time_pricing.sql
new file mode 100644
index 000000000000..dbfb84cde40b
--- /dev/null
+++ b/backend/migrations/225_channel_model_time_pricing.sql
@@ -0,0 +1,5 @@
+ALTER TABLE channel_model_pricing
+    ADD COLUMN IF NOT EXISTS time_pricing JSONB NULL;
+
+COMMENT ON COLUMN channel_model_pricing.time_pricing IS
+    'Optional IANA timezone and recurring daily multiplier periods for channel token pricing';
diff --git a/frontend/src/api/admin/channels.ts b/frontend/src/api/admin/channels.ts
index fdbeadf57a9a..6556417ced89 100644
--- a/frontend/src/api/admin/channels.ts
+++ b/frontend/src/api/admin/channels.ts
@@ -21,6 +21,17 @@ export interface PricingInterval {
   sort_order: number
 }
 
+export interface ChannelTimePricingPeriod {
+  start_time: string
+  end_time: string
+  multiplier: number
+}
+
+export interface ChannelTimePricing {
+  timezone: string
+  periods: ChannelTimePricingPeriod[]
+}
+
 export interface ChannelModelPricing {
   id?: number
   platform: string
@@ -34,6 +45,7 @@ export interface ChannelModelPricing {
   image_output_price: number | null
   per_request_price: number | null
   intervals: PricingInterval[]
+  time_pricing: ChannelTimePricing | null
 }
 
 export interface AccountStatsPricingRule {
diff --git a/frontend/src/components/admin/channel/PricingEntryCard.vue b/frontend/src/components/admin/channel/PricingEntryCard.vue
index d09e19feb296..f2a0d6502bc7 100644
--- a/frontend/src/components/admin/channel/PricingEntryCard.vue
+++ b/frontend/src/components/admin/channel/PricingEntryCard.vue
@@ -87,7 +87,12 @@
             </label>
             <Select
               :modelValue="entry.billing_mode"
-              @update:modelValue="emit('update', { ...entry, billing_mode: $event as BillingMode, intervals: [] })"
+              @update:modelValue="emit('update', {
+                ...entry,
+                billing_mode: $event as BillingMode,
+                intervals: [],
+                time_pricing: { ...entry.time_pricing, periods: [] },
+              })"
               :options="billingModeOptions"
               class="mt-1"
             />
@@ -156,6 +161,12 @@
               />
             </div>
           </div>
+
+          <TimePricingSection
+            v-if="enableTimePricing"
+            :model-value="entry.time_pricing"
+            @update:model-value="emit('update', { ...entry, time_pricing: $event })"
+          />
         </div>
 
         <!-- Per-request mode -->
@@ -238,6 +249,7 @@ import Select from '@/components/common/Select.vue'
 import Icon from '@/components/icons/Icon.vue'
 import IntervalRow from './IntervalRow.vue'
 import ModelTagInput from './ModelTagInput.vue'
+import TimePricingSection from './TimePricingSection.vue'
 import type { PricingFormEntry, IntervalFormEntry } from './types'
 import { perTokenToMTok, getPlatformTagClass } from './types'
 import type { BillingMode } from '@/api/admin/channels'
@@ -249,8 +261,10 @@ const props = withDefaults(defineProps<{
   entry: PricingFormEntry
   platform?: string
   hideTokenIntervals?: boolean
+  enableTimePricing?: boolean
 }>(), {
   hideTokenIntervals: false,
+  enableTimePricing: false,
 })
 
 const emit = defineEmits<{
diff --git a/frontend/src/components/admin/channel/TimePricingSection.vue b/frontend/src/components/admin/channel/TimePricingSection.vue
new file mode 100644
index 000000000000..b2e31ca46dd0
--- /dev/null
+++ b/frontend/src/components/admin/channel/TimePricingSection.vue
@@ -0,0 +1,161 @@
+<template>
+  <section class="mt-3 border-t border-gray-200 pt-3 dark:border-dark-600">
+    <div class="flex flex-col gap-2 sm:flex-row sm:items-end sm:justify-between">
+      <div class="min-w-0 flex-1 sm:max-w-sm">
+        <label class="block text-xs font-medium text-gray-500 dark:text-gray-400">
+          {{ t('admin.channels.form.timePricing') }}
+        </label>
+        <label class="mt-2 block text-xs text-gray-400">
+          {{ t('admin.channels.form.timezone') }}
+        </label>
+        <Select
+          :model-value="modelValue.timezone"
+          :options="timezoneOptions"
+          :aria-label="t('admin.channels.form.timezone')"
+          searchable
+          creatable
+          class="mt-1 w-full"
+          @update:model-value="updateTimezone"
+        />
+      </div>
+      <button
+        type="button"
+        class="self-start text-xs text-primary-600 hover:text-primary-700 sm:self-end sm:pb-2"
+        data-testid="add-time-period"
+        @click="addPeriod"
+      >
+        + {{ t('admin.channels.form.addTimePeriod') }}
+      </button>
+    </div>
+
+    <div v-if="modelValue.periods.length > 0" class="mt-3 space-y-3">
+      <div
+        v-for="(period, index) in modelValue.periods"
+        :key="index"
+        class="grid grid-cols-1 gap-2 border-t border-gray-200 pt-3 sm:grid-cols-[minmax(0,1fr)_minmax(0,1fr)_minmax(0,1fr)_2rem] sm:items-end dark:border-dark-600"
+      >
+        <div class="min-w-0">
+          <label :for="`${inputIdPrefix}-start-${index}`" class="block text-xs text-gray-400">
+            {{ t('admin.channels.form.startTime') }}
+          </label>
+          <input
+            :id="`${inputIdPrefix}-start-${index}`"
+            :value="period.start_time"
+            type="text"
+            inputmode="numeric"
+            maxlength="8"
+            placeholder="HH:mm:ss"
+            pattern="[0-9]{2}:[0-9]{2}:[0-9]{2}"
+            autocomplete="off"
+            class="input mt-1 w-full text-sm"
+            @input="updatePeriod(index, 'start_time', normalizeClockTime(($event.target as HTMLInputElement).value))"
+          />
+        </div>
+        <div class="min-w-0">
+          <label :for="`${inputIdPrefix}-end-${index}`" class="block text-xs text-gray-400">
+            {{ t('admin.channels.form.endTime') }}
+          </label>
+          <input
+            :id="`${inputIdPrefix}-end-${index}`"
+            :value="period.end_time"
+            type="text"
+            inputmode="numeric"
+            maxlength="8"
+            placeholder="HH:mm:ss"
+            pattern="[0-9]{2}:[0-9]{2}:[0-9]{2}"
+            autocomplete="off"
+            class="input mt-1 w-full text-sm"
+            @input="updatePeriod(index, 'end_time', normalizeClockTime(($event.target as HTMLInputElement).value))"
+          />
+        </div>
+        <div class="min-w-0">
+          <label :for="`${inputIdPrefix}-multiplier-${index}`" class="block text-xs text-gray-400">
+            {{ t('admin.channels.form.multiplier') }}
+          </label>
+          <input
+            :id="`${inputIdPrefix}-multiplier-${index}`"
+            :value="period.multiplier"
+            type="number"
+            min="0.01"
+            step="0.01"
+            class="input mt-1 w-full text-sm"
+            @input="updatePeriod(index, 'multiplier', ($event.target as HTMLInputElement).value)"
+            @blur="formatMultiplier(index, ($event.target as HTMLInputElement).value)"
+          />
+        </div>
+        <button
+          type="button"
+          class="flex h-8 w-8 items-center justify-center rounded text-gray-400 hover:text-red-500"
+          :title="t('admin.channels.form.removeTimePeriod')"
+          :aria-label="t('admin.channels.form.removeTimePeriod')"
+          :data-testid="`remove-time-period-${index}`"
+          @click="removePeriod(index)"
+        >
+          <Icon name="trash" size="sm" />
+        </button>
+      </div>
+    </div>
+  </section>
+</template>
+
+<script setup lang="ts">
+import { getCurrentInstance } from 'vue'
+import { useI18n } from 'vue-i18n'
+import Select from '@/components/common/Select.vue'
+import Icon from '@/components/icons/Icon.vue'
+import {
+  COMMON_TIMEZONES,
+  formatTimezoneOffset,
+  isValidTimePricingMultiplier,
+  type TimePricingFormEntry,
+  type TimePricingPeriodFormEntry,
+} from './types'
+
+const { t } = useI18n()
+
+const props = defineProps<{ modelValue: TimePricingFormEntry }>()
+const emit = defineEmits<{ 'update:modelValue': [value: TimePricingFormEntry] }>()
+const inputIdPrefix = `time-pricing-${getCurrentInstance()?.uid}`
+
+const timezoneOptions = COMMON_TIMEZONES.map(value => {
+  const offset = formatTimezoneOffset(value)
+  return { value, label: offset ? `${value} (${offset})` : value }
+})
+
+function updateTimezone(value: string | number | boolean | null) {
+  emit('update:modelValue', { ...props.modelValue, timezone: String(value ?? '') })
+}
+
+function normalizeClockTime(value: string): string {
+  const normalized = value.replace(/：/g, ':')
+  return normalized === '24:00:00' ? '00:00:00' : normalized
+}
+
+function addPeriod() {
+  emit('update:modelValue', {
+    ...props.modelValue,
+    periods: [
+      ...props.modelValue.periods,
+      { start_time: '', end_time: '', multiplier: '1.00' },
+    ],
+  })
+}
+
+function updatePeriod(index: number, field: keyof TimePricingPeriodFormEntry, value: string) {
+  const periods = props.modelValue.periods.map((period, current) =>
+    current === index ? { ...period, [field]: value } : period)
+  emit('update:modelValue', { ...props.modelValue, periods })
+}
+
+function formatMultiplier(index: number, value: string) {
+  if (!isValidTimePricingMultiplier(value)) return
+  updatePeriod(index, 'multiplier', Number(value).toFixed(2))
+}
+
+function removePeriod(index: number) {
+  emit('update:modelValue', {
+    ...props.modelValue,
+    periods: props.modelValue.periods.filter((_period, current) => current !== index),
+  })
+}
+</script>
diff --git a/frontend/src/components/admin/channel/__tests__/PricingEntryCard.timePricing.spec.ts b/frontend/src/components/admin/channel/__tests__/PricingEntryCard.timePricing.spec.ts
new file mode 100644
index 000000000000..45982256ebe4
--- /dev/null
+++ b/frontend/src/components/admin/channel/__tests__/PricingEntryCard.timePricing.spec.ts
@@ -0,0 +1,71 @@
+import { shallowMount } from '@vue/test-utils'
+import { describe, expect, it, vi } from 'vitest'
+import PricingEntryCard from '../PricingEntryCard.vue'
+import type { PricingFormEntry } from '../types'
+
+vi.mock('vue-i18n', async importOriginal => ({
+  ...await importOriginal<typeof import('vue-i18n')>(),
+  useI18n: () => ({ t: (key: string) => key }),
+}))
+
+function createEntry(billingMode: PricingFormEntry['billing_mode'] = 'token'): PricingFormEntry {
+  return {
+    models: [],
+    billing_mode: billingMode,
+    input_price: null,
+    output_price: null,
+    cache_write_price: null,
+    cache_read_price: null,
+    image_input_price: null,
+    image_output_price: null,
+    per_request_price: null,
+    intervals: [],
+    time_pricing: {
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00', end_time: '12:00', multiplier: '2.00' }],
+    },
+  }
+}
+
+describe('PricingEntryCard time pricing visibility', () => {
+  it('is hidden by default', () => {
+    const wrapper = shallowMount(PricingEntryCard, {
+      props: { entry: createEntry() },
+    })
+
+    expect(wrapper.findComponent({ name: 'TimePricingSection' }).exists()).toBe(false)
+  })
+
+  it('is shown for token pricing when explicitly enabled', () => {
+    const wrapper = shallowMount(PricingEntryCard, {
+      props: { entry: createEntry(), enableTimePricing: true },
+    })
+
+    expect(wrapper.findComponent({ name: 'TimePricingSection' }).exists()).toBe(true)
+  })
+
+  it('is hidden for non-token pricing even when explicitly enabled', () => {
+    const wrapper = shallowMount(PricingEntryCard, {
+      props: { entry: createEntry('per_request'), enableTimePricing: true },
+    })
+
+    expect(wrapper.findComponent({ name: 'TimePricingSection' }).exists()).toBe(false)
+  })
+
+  it('clears time periods when changing billing mode', () => {
+    const entry = createEntry()
+    const wrapper = shallowMount(PricingEntryCard, {
+      props: { entry, enableTimePricing: true },
+    })
+
+    wrapper.findComponent({ name: 'Select' }).vm.$emit('update:modelValue', 'image')
+
+    expect(wrapper.emitted('update')?.[0]?.[0]).toEqual({
+      ...entry,
+      billing_mode: 'image',
+      intervals: [],
+      time_pricing: { timezone: 'Asia/Shanghai', periods: [] },
+    })
+    expect(entry.time_pricing.periods).toHaveLength(1)
+  })
+})
diff --git a/frontend/src/components/admin/channel/__tests__/TimePricingSection.spec.ts b/frontend/src/components/admin/channel/__tests__/TimePricingSection.spec.ts
new file mode 100644
index 000000000000..5ce5348a2742
--- /dev/null
+++ b/frontend/src/components/admin/channel/__tests__/TimePricingSection.spec.ts
@@ -0,0 +1,183 @@
+import { mount } from '@vue/test-utils'
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+import TimePricingSection from '../TimePricingSection.vue'
+import { createDefaultTimePricingForm } from '../types'
+
+vi.mock('vue-i18n', async importOriginal => ({
+  ...await importOriginal<typeof import('vue-i18n')>(),
+  useI18n: () => ({ t: (key: string) => key }),
+}))
+
+const SelectStub = {
+  name: 'Select',
+  props: ['modelValue', 'options'],
+  emits: ['update:modelValue'],
+  template: '<button type="button" data-testid="timezone-select" />',
+}
+
+describe('TimePricingSection', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+  })
+
+  it('adds a neutral period immutably', async () => {
+    const value = createDefaultTimePricingForm()
+    const wrapper = mount(TimePricingSection, {
+      props: { modelValue: value },
+      global: { stubs: { Select: SelectStub, Icon: true } },
+    })
+
+    await wrapper.get('[data-testid="add-time-period"]').trigger('click')
+
+    expect(wrapper.emitted('update:modelValue')?.[0]?.[0]).toEqual({
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '', end_time: '', multiplier: '1.00' }],
+    })
+    expect(value.periods).toHaveLength(0)
+  })
+
+  it('removes without mutating props', async () => {
+    const value = {
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }],
+    }
+    const wrapper = mount(TimePricingSection, {
+      props: { modelValue: value },
+      global: { stubs: { Select: SelectStub, Icon: true } },
+    })
+
+    const removeButton = wrapper.get('[data-testid="remove-time-period-0"]')
+    expect(removeButton.attributes('title')).toBe('admin.channels.form.removeTimePeriod')
+    expect(removeButton.attributes('aria-label')).toBe('admin.channels.form.removeTimePeriod')
+    await removeButton.trigger('click')
+
+    expect(wrapper.emitted('update:modelValue')?.[0]?.[0]).toEqual({
+      timezone: 'Asia/Shanghai',
+      periods: [],
+    })
+    expect(value.periods).toHaveLength(1)
+  })
+
+  it('uses locale-independent 24-hour second inputs', () => {
+    const value = {
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }],
+    }
+    const wrapper = mount(TimePricingSection, {
+      props: { modelValue: value },
+      global: { stubs: { Select: SelectStub, Icon: true } },
+    })
+
+    const timeInputs = wrapper.findAll('input[inputmode="numeric"]')
+    expect(timeInputs).toHaveLength(2)
+    expect(timeInputs.every(input => input.attributes('type') === 'text')).toBe(true)
+    expect(timeInputs.every(input => input.attributes('maxlength') === '8')).toBe(true)
+    expect(timeInputs.every(input => input.attributes('placeholder') === 'HH:mm:ss')).toBe(true)
+  })
+
+  it('updates the timezone and normalizes full-width colons immutably', async () => {
+    const value = {
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }],
+    }
+    const wrapper = mount(TimePricingSection, {
+      props: { modelValue: value },
+      global: { stubs: { Select: SelectStub, Icon: true } },
+    })
+
+    wrapper.getComponent(SelectStub).vm.$emit('update:modelValue', 'Europe/London')
+    await wrapper.get('input').setValue('10：30：15')
+
+    expect(wrapper.emitted('update:modelValue')?.[0]?.[0]).toEqual({
+      ...value,
+      timezone: 'Europe/London',
+    })
+    expect(wrapper.emitted('update:modelValue')?.[1]?.[0]).toEqual({
+      ...value,
+      periods: [{ ...value.periods[0], start_time: '10:30:15' }],
+    })
+    expect(value).toEqual({
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }],
+    })
+  })
+
+  it('normalizes 24:00:00 to 00:00:00', async () => {
+    const value = {
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }],
+    }
+    const wrapper = mount(TimePricingSection, {
+      props: { modelValue: value },
+      global: { stubs: { Select: SelectStub, Icon: true } },
+    })
+
+    await wrapper.get('input').setValue('24:00:00')
+
+    expect(wrapper.emitted('update:modelValue')?.[0]?.[0]).toEqual({
+      ...value,
+      periods: [{ ...value.periods[0], start_time: '00:00:00' }],
+    })
+  })
+
+  it('formats only a valid positive multiplier on blur', async () => {
+    const value = {
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2' }],
+    }
+    const wrapper = mount(TimePricingSection, {
+      props: { modelValue: value },
+      global: { stubs: { Select: SelectStub, Icon: true } },
+    })
+
+    const multiplier = wrapper.get('input[type="number"]')
+    expect(multiplier.attributes('min')).toBe('0.01')
+    expect(multiplier.attributes('step')).toBe('0.01')
+    await multiplier.trigger('blur')
+
+    expect(wrapper.emitted('update:modelValue')?.[0]?.[0]).toEqual({
+      ...value,
+      periods: [{ ...value.periods[0], multiplier: '2.00' }],
+    })
+
+  })
+
+  it.each(['1.234', '1e2', '.5'])('keeps invalid multiplier %s unchanged on blur', async invalid => {
+    const value = {
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: invalid }],
+    }
+    const wrapper = mount(TimePricingSection, {
+      props: { modelValue: value },
+      global: { stubs: { Select: SelectStub, Icon: true } },
+    })
+
+    await wrapper.get('input[type="number"]').trigger('blur')
+
+    expect(wrapper.emitted('update:modelValue')).toBeUndefined()
+    expect(value.periods[0].multiplier).toBe(invalid)
+  })
+
+  it('uses unique input ids across component instances', () => {
+    const value = {
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }],
+    }
+    const wrapper = mount({
+      components: { TimePricingSection },
+      data: () => ({ first: value, second: value }),
+      template: `
+        <div>
+          <TimePricingSection v-model="first" />
+          <TimePricingSection v-model="second" />
+        </div>
+      `,
+    }, {
+      global: { stubs: { Select: SelectStub, Icon: true } },
+    })
+
+    const ids = wrapper.findAll('input').map(input => input.attributes('id'))
+    expect(ids).toHaveLength(6)
+    expect(new Set(ids).size).toBe(ids.length)
+  })
+})
diff --git a/frontend/src/components/admin/channel/__tests__/types.spec.ts b/frontend/src/components/admin/channel/__tests__/types.spec.ts
index 04758012f15c..b1ebef252ac4 100644
--- a/frontend/src/components/admin/channel/__tests__/types.spec.ts
+++ b/frontend/src/components/admin/channel/__tests__/types.spec.ts
@@ -1,5 +1,14 @@
 import { describe, expect, it } from 'vitest'
-import { validateIntervals, type IntervalFormEntry } from '../types'
+import {
+  apiTimePricingToForm,
+  createDefaultTimePricingForm,
+  formTimePricingToAPI,
+  validateIntervals,
+  validateTimePricing,
+  type IntervalFormEntry,
+  type TimePricingFormEntry,
+  type TimePricingPeriodFormEntry,
+} from '../types'
 
 function makeInterval(over: Partial<IntervalFormEntry>): IntervalFormEntry {
   return {
@@ -81,3 +90,67 @@ describe('validateIntervals', () => {
     })
   })
 })
+
+describe('time pricing', () => {
+  it('uses a disabled Shanghai default', () => {
+    const form = createDefaultTimePricingForm()
+    expect(form).toEqual({ timezone: 'Asia/Shanghai', periods: [] })
+    expect(formTimePricingToAPI(form)).toBeNull()
+  })
+
+  it('round-trips and formats multiplier', () => {
+    const form = apiTimePricingToForm({
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00', end_time: '12:00', multiplier: 2 }],
+    })
+    expect(form.periods[0]).toEqual({
+      start_time: '09:00:00',
+      end_time: '12:00:00',
+      multiplier: '2.00',
+    })
+    expect(formTimePricingToAPI(form)).toEqual({
+      timezone: 'Asia/Shanghai',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: 2 }],
+    })
+  })
+
+  it.each([
+    ['separated', [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }, { start_time: '14:00:00', end_time: '18:00:00', multiplier: '2.00' }], null],
+    ['adjacent', [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }, { start_time: '12:00:00', end_time: '14:00:00', multiplier: '1.50' }], null],
+    ['midnight split', [{ start_time: '22:00:00', end_time: '00:00:00', multiplier: '2.00' }, { start_time: '00:00:00', end_time: '02:00:00', multiplier: '2.00' }], null],
+    ['overlap by one second', [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }, { start_time: '11:59:59', end_time: '14:00:00', multiplier: '2.00' }], 'overlap'],
+    ['cross midnight', [{ start_time: '22:00:00', end_time: '02:00:00', multiplier: '2.00' }], 'range'],
+    ['equal midnight', [{ start_time: '00:00:00', end_time: '00:00:00', multiplier: '2.00' }], 'range'],
+    ['missing seconds', [{ start_time: '09:00', end_time: '12:00', multiplier: '2.00' }], 'format'],
+    ['zero', [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '0.00' }], 'multiplier'],
+    ['three decimals', [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '1.001' }], 'multiplier'],
+  ])('%s', (_name, periods, errorKey) => {
+    const result = validateTimePricing({
+      timezone: 'Asia/Shanghai',
+      periods: periods as TimePricingPeriodFormEntry[],
+    }, t)
+    if (errorKey === null) expect(result).toBeNull()
+    else expect(result).toContain(String(errorKey))
+  })
+
+  it('rejects non-IANA timezone', () => {
+    expect(validateTimePricing({
+      timezone: 'UTC+8',
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }],
+    }, t)).toContain('timezone')
+  })
+
+  it.each([
+    ['missing', undefined],
+    ['blank', '   '],
+  ])('rejects a %s timezone without throwing during conversion', (_name, timezone) => {
+    const form = {
+      timezone,
+      periods: [{ start_time: '09:00:00', end_time: '12:00:00', multiplier: '2.00' }],
+    } as unknown as TimePricingFormEntry
+
+    expect(validateTimePricing(form, t)).toContain('timezone')
+    expect(() => formTimePricingToAPI(form)).not.toThrow()
+    expect(formTimePricingToAPI(form)?.timezone).toBe('')
+  })
+})
diff --git a/frontend/src/components/admin/channel/types.ts b/frontend/src/components/admin/channel/types.ts
index 5c43b7d34954..8809eff072c0 100644
--- a/frontend/src/components/admin/channel/types.ts
+++ b/frontend/src/components/admin/channel/types.ts
@@ -1,4 +1,4 @@
-import type { BillingMode, PricingInterval } from '@/api/admin/channels'
+import type { BillingMode, ChannelTimePricing, PricingInterval } from '@/api/admin/channels'
 
 type TranslateFn = (key: string, params?: Record<string, unknown>) => string
 
@@ -25,6 +25,134 @@ export interface PricingFormEntry {
   image_output_price: number | string | null
   per_request_price: number | string | null
   intervals: IntervalFormEntry[]
+  time_pricing: TimePricingFormEntry
+}
+
+export interface TimePricingPeriodFormEntry {
+  start_time: string
+  end_time: string
+  multiplier: number | string
+}
+
+export interface TimePricingFormEntry {
+  timezone: string
+  periods: TimePricingPeriodFormEntry[]
+}
+
+export const DEFAULT_TIME_PRICING_TIMEZONE = 'Asia/Shanghai'
+
+const CLOCK_TIME = /^(?:[01]\d|2[0-3]):[0-5]\d:[0-5]\d$/
+const LEGACY_CLOCK_TIME = /^(?:[01]\d|2[0-3]):[0-5]\d$/
+const TWO_DECIMAL_MULTIPLIER = /^\d+(?:\.\d{1,2})?$/
+
+export function isValidTimePricingMultiplier(value: number | string): boolean {
+  const multiplier = String(value)
+  const numericValue = Number(multiplier)
+  return TWO_DECIMAL_MULTIPLIER.test(multiplier) &&
+    Number.isFinite(numericValue) && numericValue > 0
+}
+
+export const COMMON_TIMEZONES = [
+  'UTC', 'Asia/Shanghai', 'Asia/Tokyo', 'Asia/Seoul', 'Asia/Singapore', 'Asia/Kolkata',
+  'Australia/Sydney', 'Europe/London', 'Europe/Paris', 'Europe/Berlin',
+  'America/New_York', 'America/Chicago', 'America/Denver', 'America/Los_Angeles',
+  'America/Toronto', 'America/Sao_Paulo', 'Pacific/Auckland', 'Pacific/Honolulu',
+]
+
+export function createDefaultTimePricingForm(): TimePricingFormEntry {
+  return { timezone: DEFAULT_TIME_PRICING_TIMEZONE, periods: [] }
+}
+
+export function apiTimePricingToForm(value: ChannelTimePricing | null | undefined): TimePricingFormEntry {
+  if (!value) return createDefaultTimePricingForm()
+  return {
+    timezone: value.timezone || DEFAULT_TIME_PRICING_TIMEZONE,
+    periods: (value.periods || []).map(period => ({
+      start_time: LEGACY_CLOCK_TIME.test(period.start_time) ? `${period.start_time}:00` : period.start_time,
+      end_time: LEGACY_CLOCK_TIME.test(period.end_time) ? `${period.end_time}:00` : period.end_time,
+      multiplier: Number(period.multiplier).toFixed(2),
+    })),
+  }
+}
+
+export function formTimePricingToAPI(value: TimePricingFormEntry | null | undefined): ChannelTimePricing | null {
+  if (!value?.periods?.length) return null
+  const timezone = typeof value.timezone === 'string' ? value.timezone.trim() : ''
+  return {
+    timezone,
+    periods: value.periods.map(period => ({
+      start_time: period.start_time,
+      end_time: period.end_time,
+      multiplier: Number(period.multiplier),
+    })),
+  }
+}
+
+function timeToSeconds(time: string, isEnd: boolean): number {
+  if (isEnd && time === '00:00:00') return 24 * 60 * 60
+  const [hours, minutes, seconds] = time.split(':').map(Number)
+  return hours * 60 * 60 + minutes * 60 + seconds
+}
+
+function timePricingValidationMessage(t: TranslateFn, key: string): string {
+  return t(`admin.channels.timePricingValidation.${key}`)
+}
+
+export function validateTimePricing(value: TimePricingFormEntry, t: TranslateFn): string | null {
+  if (!value?.periods?.length) return null
+
+  if (typeof value.timezone !== 'string' || value.timezone.trim() === '') {
+    return timePricingValidationMessage(t, 'timezone')
+  }
+  const timezone = value.timezone.trim()
+
+  try {
+    new Intl.DateTimeFormat('en-US', { timeZone: timezone })
+  } catch {
+    return timePricingValidationMessage(t, 'timezone')
+  }
+
+  const periods = [] as { start: number, end: number }[]
+  for (const period of value.periods) {
+    if (!CLOCK_TIME.test(period.start_time) || !CLOCK_TIME.test(period.end_time)) {
+      return timePricingValidationMessage(t, 'format')
+    }
+    if (period.start_time === period.end_time) {
+      return timePricingValidationMessage(t, 'range')
+    }
+
+    const start = timeToSeconds(period.start_time, false)
+    const end = timeToSeconds(period.end_time, true)
+    if (start >= end) return timePricingValidationMessage(t, 'range')
+
+    if (!isValidTimePricingMultiplier(period.multiplier)) {
+      return timePricingValidationMessage(t, 'multiplier')
+    }
+    periods.push({ start, end })
+  }
+
+  const sorted = [...periods].sort((a, b) => a.start - b.start)
+  for (let i = 1; i < sorted.length; i++) {
+    if (sorted[i].start < sorted[i - 1].end) {
+      return timePricingValidationMessage(t, 'overlap')
+    }
+  }
+  return null
+}
+
+export function formatTimezoneOffset(timezone: string, at = new Date()): string {
+  try {
+    const part = new Intl.DateTimeFormat('en-US', {
+      timeZone: timezone,
+      timeZoneName: 'shortOffset',
+    }).formatToParts(at).find(item => item.type === 'timeZoneName')?.value
+    if (!part || part === 'GMT') return 'UTC+00:00'
+    const match = /^GMT([+-])(\d{1,2})(?::(\d{2}))?$/.exec(part)
+    if (!match) return ''
+    return `UTC${match[1]}${match[2].padStart(2, '0')}:${match[3] || '00'}`
+  } catch {
+    return ''
+  }
 }
 
 // 价格转换：后端存 per-token，前端显示 per-MTok ($/1M tokens)
diff --git a/frontend/src/i18n/locales/en/admin/channels.ts b/frontend/src/i18n/locales/en/admin/channels.ts
index d68ea5060deb..bed17bc0301a 100644
--- a/frontend/src/i18n/locales/en/admin/channels.ts
+++ b/frontend/src/i18n/locales/en/admin/channels.ts
@@ -80,6 +80,13 @@ export default {
           perRequestPrice: 'per-request price'
         }
       },
+      timePricingValidation: {
+        timezone: 'Select a valid IANA time zone',
+        format: 'Start and end times must use HH:mm:ss format',
+        range: 'Start time must be earlier than end time; split ranges across midnight',
+        overlap: 'Time periods must not overlap',
+        multiplier: 'Multiplier must be greater than 0 with at most two decimal places'
+      },
       deleteConfirm: 'Are you sure you want to delete channel "{name}"? This cannot be undone.',
       columns: {
         name: 'Name',
@@ -122,6 +129,13 @@ export default {
         imageOutputPrice: 'Image Output Price',
         pricePlaceholder: 'Default',
         intervals: 'Context Intervals (optional)',
+        timePricing: 'Time-based pricing (optional)',
+        timezone: 'Time zone',
+        addTimePeriod: 'Add period',
+        startTime: 'Start time',
+        endTime: 'End time',
+        multiplier: 'Multiplier',
+        removeTimePeriod: 'Remove period',
         minTokens: 'Min',
         maxTokens: 'Max',
         inclusive: '(inclusive)',
diff --git a/frontend/src/i18n/locales/zh/admin/channels.ts b/frontend/src/i18n/locales/zh/admin/channels.ts
index 808e78fa8823..12cb22d565d3 100644
--- a/frontend/src/i18n/locales/zh/admin/channels.ts
+++ b/frontend/src/i18n/locales/zh/admin/channels.ts
@@ -80,6 +80,13 @@ export default {
           perRequestPrice: '单次价格'
         }
       },
+      timePricingValidation: {
+        timezone: '请选择有效的 IANA 时区',
+        format: '开始时间和结束时间必须使用 HH:mm:ss 格式',
+        range: '开始时间必须早于结束时间；跨午夜请拆分为两个时间段',
+        overlap: '时间段不能重叠',
+        multiplier: '倍率必须大于 0，且最多保留两位小数'
+      },
       deleteConfirm: '确定要删除渠道「{name}」吗？此操作不可撤销。',
       columns: {
         name: '名称',
@@ -122,6 +129,13 @@ export default {
         imageOutputPrice: '图片输出价格',
         pricePlaceholder: '默认',
         intervals: '上下文区间定价（可选）',
+        timePricing: '时间段定价（可选）',
+        timezone: '时区',
+        addTimePeriod: '添加时间段',
+        startTime: '开始时间',
+        endTime: '结束时间',
+        multiplier: '倍率',
+        removeTimePeriod: '删除时间段',
         minTokens: '最小',
         maxTokens: '最大',
         inclusive: '（含）',
diff --git a/frontend/src/views/admin/ChannelsView.vue b/frontend/src/views/admin/ChannelsView.vue
index 9dceb5362a75..d431cb129337 100644
--- a/frontend/src/views/admin/ChannelsView.vue
+++ b/frontend/src/views/admin/ChannelsView.vue
@@ -447,6 +447,7 @@
                   :key="idx"
                   :entry="entry"
                   :platform="section.platform"
+                  enable-time-pricing
                   @update="updatePricingEntry(sIdx, idx, $event)"
                   @remove="removePricingEntry(sIdx, idx)"
                 />
@@ -632,7 +633,7 @@ import { extractApiErrorMessage } from '@/utils/apiError'
 import { adminAPI } from '@/api/admin'
 import type { Channel, ChannelModelPricing, CreateChannelRequest, UpdateChannelRequest, AccountStatsPricingRule } from '@/api/admin/channels'
 import type { PricingFormEntry } from '@/components/admin/channel/types'
-import { mTokToPerToken, perTokenToMTok, apiIntervalsToForm, formIntervalsToAPI, findModelConflict, validateIntervals } from '@/components/admin/channel/types'
+import { apiIntervalsToForm, apiTimePricingToForm, createDefaultTimePricingForm, findModelConflict, formIntervalsToAPI, formTimePricingToAPI, mTokToPerToken, perTokenToMTok, validateIntervals, validateTimePricing } from '@/components/admin/channel/types'
 import type { AdminGroup, GroupPlatform } from '@/types'
 import type { Column } from '@/components/common/types'
 import { platformTextClass, platformBadgeLightClass } from '@/utils/platformColors'
@@ -859,7 +860,8 @@ function addPricingEntry(sectionIdx: number) {
     image_input_price: null,
     image_output_price: null,
     per_request_price: null,
-    intervals: []
+    intervals: [],
+    time_pricing: createDefaultTimePricingForm()
   })
 }
 
@@ -892,7 +894,8 @@ async function syncLatestModels(sectionIdx: number) {
       image_input_price: null,
       image_output_price: null,
       per_request_price: null,
-      intervals: []
+      intervals: [],
+      time_pricing: createDefaultTimePricingForm()
     })
     appStore.showSuccess(t('admin.channels.form.syncModelsSuccess', { count: newModels.length }))
   } catch (error) {
@@ -957,7 +960,8 @@ function addRulePricingEntry(sectionIdx: number, ruleIndex: number) {
     image_input_price: null,
     image_output_price: null,
     per_request_price: null,
-    intervals: []
+    intervals: [],
+    time_pricing: createDefaultTimePricingForm()
   })
 }
 
@@ -1073,7 +1077,8 @@ function accountStatsRulesToAPI(): AccountStatsPricingRule[] {
             image_input_price: mTokToPerToken(p.image_input_price),
             image_output_price: mTokToPerToken(p.image_output_price),
             per_request_price: p.per_request_price != null && p.per_request_price !== '' ? Number(p.per_request_price) : null,
-            intervals: formIntervalsToAPI(p.intervals || [])
+            intervals: formIntervalsToAPI(p.intervals || []),
+            time_pricing: null
           }))
       })
     }
@@ -1114,7 +1119,8 @@ function formToAPI(): { group_ids: number[], model_pricing: ChannelModelPricing[
         image_input_price: mTokToPerToken(entry.image_input_price),
         image_output_price: mTokToPerToken(entry.image_output_price),
         per_request_price: entry.per_request_price != null && entry.per_request_price !== '' ? Number(entry.per_request_price) : null,
-        intervals: formIntervalsToAPI(entry.intervals || [])
+        intervals: formIntervalsToAPI(entry.intervals || []),
+        time_pricing: formTimePricingToAPI(entry.time_pricing)
       })
     }
   }
@@ -1211,7 +1217,8 @@ function apiToForm(channel: Channel): PlatformSection[] {
         image_input_price: perTokenToMTok(p.image_input_price),
         image_output_price: perTokenToMTok(p.image_output_price),
         per_request_price: p.per_request_price,
-        intervals: apiIntervalsToForm(p.intervals || [])
+        intervals: apiIntervalsToForm(p.intervals || []),
+        time_pricing: apiTimePricingToForm(p.time_pricing)
       } as PricingFormEntry))
 
     // Read web_search_emulation from features_config
@@ -1400,7 +1407,8 @@ function distributeRulesToPlatforms(apiRules: AccountStatsPricingRule[]) {
         image_input_price: perTokenToMTok(p.image_input_price),
         image_output_price: perTokenToMTok(p.image_output_price),
         per_request_price: p.per_request_price,
-        intervals: apiIntervalsToForm(p.intervals || [])
+        intervals: apiIntervalsToForm(p.intervals || []),
+        time_pricing: createDefaultTimePricingForm()
       } as PricingFormEntry))
     }
     section.account_stats_pricing_rules.push(formRule)
@@ -1523,6 +1531,20 @@ async function handleSubmit() {
     }
   }
 
+  // 校验时间段定价，并切换到对应平台便于修正
+  for (const section of form.platforms.filter(s => s.enabled)) {
+    for (const entry of section.model_pricing) {
+      const timePricingError = validateTimePricing(entry.time_pricing, t)
+      if (timePricingError) {
+        const platformLabel = t('admin.groups.platforms.' + section.platform, section.platform)
+        const modelLabel = entry.models.join(', ') || t('admin.channels.form.unnamed')
+        appStore.showError(`${platformLabel} - ${modelLabel}: ${timePricingError}`)
+        activeTab.value = section.platform
+        return
+      }
+    }
+  }
+
   const { group_ids, model_pricing, model_mapping, features_config } = formToAPI()
 
   submitting.value = true
diff --git a/frontend/src/views/admin/GroupsView.vue b/frontend/src/views/admin/GroupsView.vue
index 904b8926d4f3..9a427e7c64a6 100644
--- a/frontend/src/views/admin/GroupsView.vue
+++ b/frontend/src/views/admin/GroupsView.vue
@@ -4429,6 +4429,7 @@ import PricingEntryCard from "@/components/admin/channel/PricingEntryCard.vue";
 import type { PricingFormEntry } from "@/components/admin/channel/types";
 import {
   apiIntervalsToForm,
+  createDefaultTimePricingForm,
   formIntervalsToAPI,
   mTokToPerToken,
   perTokenToMTok,
@@ -4499,6 +4500,7 @@ const emptyGroupPricing = (): PricingFormEntry => ({
   image_output_price: null,
   per_request_price: null,
   intervals: [],
+  time_pricing: createDefaultTimePricingForm(),
 });
 
 const addGroupPricing = (entries: PricingFormEntry[]) =>
@@ -4518,6 +4520,7 @@ const groupPricingFromAPI = (
     image_output_price: perTokenToMTok(entry.image_output_price),
     per_request_price: entry.per_request_price,
     intervals: apiIntervalsToForm(entry.intervals || []),
+    time_pricing: createDefaultTimePricingForm(),
   }));
 
 const groupPricingToAPI = (
@@ -4541,6 +4544,7 @@ const groupPricingToAPI = (
         entry.billing_mode === "token"
           ? []
           : formIntervalsToAPI(entry.intervals || []),
+      time_pricing: null,
     }));
 
 const { t } = useI18n();

From 6793d5ac85f4ca469cc49a9cec29285c78e445b1 Mon Sep 17 00:00:00 2001
From: Jarl <okbexx@gmail.com>
Date: Mon, 17 Aug 2026 19:38:03 +0800
Subject: [PATCH 42/68] fix(openai): make Codex convergence identity consistent

---
 backend/internal/repository/account_repo.go   |  43 ++-
 ...ccount_repo_codex_fingerprint_seed_test.go | 123 +++++++
 ...erprint_seed_migration_integration_test.go | 150 ++++++++
 .../internal/repository/scheduler_cache.go    |   2 +
 .../repository/scheduler_cache_unit_test.go   |   4 +
 ...t_codex_fingerprint_seed_lifecycle_test.go | 245 +++++++++++++
 backend/internal/service/account_service.go   |   9 +-
 ...ccount_test_service_openai_compact_test.go |  11 +-
 backend/internal/service/admin_account.go     |  48 ++-
 .../service/openai_codex_fingerprint.go       | 336 ++++++++++++++----
 .../service/openai_codex_fingerprint_test.go  | 325 +++++++++++++++--
 .../service/openai_gateway_forward.go         |   5 +-
 .../service/openai_gateway_passthrough.go     |   1 +
 .../service/openai_oauth_passthrough_test.go  | 206 +++++++++++
 .../service/openai_ws_forwarder_payload.go    |   9 +-
 .../openai_ws_forwarder_success_test.go       | 137 +++++++
 .../service/openai_ws_forwarder_v2.go         |  15 +-
 backend/internal/service/openai_ws_pool.go    | 100 ++++--
 .../internal/service/openai_ws_pool_test.go   | 200 +++++++++++
 .../225_backfill_codex_fingerprint_seed.sql   |  21 ++
 20 files changed, 1832 insertions(+), 158 deletions(-)
 create mode 100644 backend/internal/repository/account_repo_codex_fingerprint_seed_test.go
 create mode 100644 backend/internal/repository/codex_fingerprint_seed_migration_integration_test.go
 create mode 100644 backend/internal/service/account_codex_fingerprint_seed_lifecycle_test.go
 create mode 100644 backend/migrations/225_backfill_codex_fingerprint_seed.sql

diff --git a/backend/internal/repository/account_repo.go b/backend/internal/repository/account_repo.go
index 010442eb6bcf..664a8454608c 100644
--- a/backend/internal/repository/account_repo.go
+++ b/backend/internal/repository/account_repo.go
@@ -71,6 +71,39 @@ var schedulerNeutralExtraKeys = map[string]struct{}{
 
 const postgresParameterBatchSize = 50000
 
+const codexFingerprintSeedCanonicalPattern = "^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$"
+const codexFingerprintNilSeed = "00000000-0000-0000-0000-000000000000"
+
+func codexFingerprintSeedValidSQL(extraExpr string) string {
+	value := "(" + extraExpr + " ->> 'codex_fingerprint_seed')"
+	return "(" + value + " ~ '" + codexFingerprintSeedCanonicalPattern + "' AND " + value + " <> '" + codexFingerprintNilSeed + "')"
+}
+
+func ensureCodexFingerprintSeedSQL(extraExpr string) string {
+	return "CASE WHEN platform = 'openai' AND type = 'oauth' THEN " +
+		"jsonb_set(" + extraExpr + ", '{codex_fingerprint_seed}', " +
+		"CASE WHEN " + codexFingerprintSeedValidSQL("extra") +
+		" THEN to_jsonb(extra ->> 'codex_fingerprint_seed') ELSE to_jsonb(gen_random_uuid()::text) END, true) " +
+		"ELSE " + extraExpr + " END"
+}
+
+func stripCodexFingerprintSeedFromExtraUpdate(extra map[string]any) map[string]any {
+	if extra == nil {
+		return nil
+	}
+	if _, exists := extra["codex_fingerprint_seed"]; !exists {
+		return extra
+	}
+	stripped := make(map[string]any, len(extra)-1)
+	for key, value := range extra {
+		if key == "codex_fingerprint_seed" {
+			continue
+		}
+		stripped[key] = value
+	}
+	return stripped
+}
+
 // NewAccountRepository 创建账户仓储实例。
 // 这是对外暴露的构造函数，返回接口类型以便于依赖注入。
 func NewAccountRepository(client *dbent.Client, sqlDB *sql.DB, schedulerCache service.SchedulerCache) service.AccountRepository {
@@ -2520,6 +2553,7 @@ func (r *accountRepository) AutoPauseExpiredAccounts(ctx context.Context, now ti
 }
 
 func (r *accountRepository) UpdateExtra(ctx context.Context, id int64, updates map[string]any) error {
+	updates = stripCodexFingerprintSeedFromExtraUpdate(updates)
 	if len(updates) == 0 {
 		return nil
 	}
@@ -2552,6 +2586,9 @@ func (r *accountRepository) UpdateExtra(ctx context.Context, id int64, updates m
 	if clearProbeSnapshot {
 		extraExpression = "(" + extraExpression + ") - 'upstream_billing_probe'"
 	}
+	if service.ShouldEnsureCodexFingerprintSeedForExtraUpdates(updates) {
+		extraExpression = ensureCodexFingerprintSeedSQL(extraExpression)
+	}
 	result, err := client.ExecContext(
 		ctx,
 		"UPDATE accounts SET extra = "+extraExpression+", updated_at = NOW() WHERE id = $2 AND deleted_at IS NULL",
@@ -2793,6 +2830,7 @@ func (r *accountRepository) BulkUpdate(ctx context.Context, ids []int64, updates
 	if len(ids) == 0 {
 		return 0, nil
 	}
+	updates.Extra = stripCodexFingerprintSeedFromExtraUpdate(updates.Extra)
 
 	setClauses := make([]string, 0, 8)
 	args := make([]any, 0, 8)
@@ -2880,7 +2918,7 @@ func (r *accountRepository) BulkUpdate(ctx context.Context, ids []int64, updates
 				" AND "+ollamaCloudBaseURLMatchesSQL(credentialPlaceholder+"::jsonb ->> 'base_url'")+")")
 	}
 
-	if len(updates.Extra) > 0 || len(ollamaGroupIdentityChanges) > 0 || ollamaProxyIdentityChanged != "" {
+	if len(updates.Extra) > 0 || len(ollamaGroupIdentityChanges) > 0 || ollamaProxyIdentityChanged != "" || updates.EnsureCodexFingerprintSeed {
 		extraExpression := "COALESCE(extra, '{}'::jsonb)"
 		if len(updates.Extra) > 0 {
 			payload, err := json.Marshal(updates.Extra)
@@ -2919,6 +2957,9 @@ func (r *accountRepository) BulkUpdate(ctx context.Context, ids []int64, updates
 		} else if snapshotIdentityChanged != "" {
 			extraExpression = "CASE WHEN " + snapshotIdentityChanged + " THEN (" + extraExpression + ") - 'ollama_cloud_usage_snapshot' ELSE " + extraExpression + " END"
 		}
+		if updates.EnsureCodexFingerprintSeed {
+			extraExpression = ensureCodexFingerprintSeedSQL(extraExpression)
+		}
 		setClauses = append(setClauses, "extra = "+extraExpression)
 	}
 
diff --git a/backend/internal/repository/account_repo_codex_fingerprint_seed_test.go b/backend/internal/repository/account_repo_codex_fingerprint_seed_test.go
new file mode 100644
index 000000000000..1ad1723748d4
--- /dev/null
+++ b/backend/internal/repository/account_repo_codex_fingerprint_seed_test.go
@@ -0,0 +1,123 @@
+package repository
+
+import (
+	"context"
+	"errors"
+	"regexp"
+	"testing"
+
+	"github.com/DATA-DOG/go-sqlmock"
+	dbent "github.com/Wei-Shaw/sub2api/ent"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/stretchr/testify/require"
+
+	"entgo.io/ent/dialect"
+	entsql "entgo.io/ent/dialect/sql"
+)
+
+func TestBulkUpdateEnsuresCodexFingerprintSeedWithPerRowSQL(t *testing.T) {
+	exec := &recordingSQLExecutor{result: rowsAffectedResult(0)}
+	repo := newAccountRepositoryWithSQL(nil, exec, nil)
+
+	_, err := repo.BulkUpdate(context.Background(), []int64{27, 28}, service.AccountBulkUpdate{
+		Extra: map[string]any{
+			"codex_fingerprint_mode": "session",
+			"codex_fingerprint_seed": "22222222-2222-4222-8222-222222222222",
+		},
+		EnsureCodexFingerprintSeed: true,
+	})
+
+	require.NoError(t, err)
+	require.NotEmpty(t, exec.execQueries)
+	query := normalizeSQLWhitespace(exec.execQueries[0])
+	require.Contains(t, query, "jsonb_set")
+	require.Contains(t, query, "gen_random_uuid()::text")
+	require.Contains(t, query, "platform = 'openai' AND type = 'oauth'")
+	require.Contains(t, query, "to_jsonb(extra ->> 'codex_fingerprint_seed')")
+	require.Contains(t, query, codexFingerprintSeedCanonicalPattern)
+	require.NotContains(t, query, "22222222-2222-4222-8222-222222222222")
+	require.NotEmpty(t, exec.execArgs)
+	payload, ok := exec.execArgs[0][0].([]byte)
+	require.True(t, ok)
+	require.Equal(t, `{"codex_fingerprint_mode":"session"}`, string(payload))
+}
+
+func TestUpdateExtraEnsuresCodexFingerprintSeedAtomicallyWhenEnabling(t *testing.T) {
+	db, mock, err := sqlmock.New()
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+	client := dbent.NewClient(dbent.Driver(entsql.OpenDB(dialect.Postgres, db)))
+	t.Cleanup(func() { _ = client.Close() })
+
+	mock.ExpectBegin()
+	mock.ExpectExec(`(?s)UPDATE accounts SET extra = .*jsonb_set.*gen_random_uuid\(\)::text.*WHERE id = \$2 AND deleted_at IS NULL`).
+		WithArgs(`{"codex_fingerprint_mode":"device"}`, int64(27)).
+		WillReturnResult(sqlmock.NewResult(0, 1))
+	mock.ExpectExec(regexp.QuoteMeta("INSERT INTO scheduler_outbox")).
+		WithArgs(service.SchedulerOutboxEventAccountChanged, int64(27), nil, nil, sqlmock.AnyArg()).
+		WillReturnResult(sqlmock.NewResult(1, 1))
+	mock.ExpectCommit()
+	repo := newAccountRepositoryWithSQL(client, db, nil)
+
+	err = repo.UpdateExtra(context.Background(), 27, map[string]any{
+		"codex_fingerprint_mode": "device",
+		"codex_fingerprint_seed": "22222222-2222-4222-8222-222222222222",
+	})
+
+	require.NoError(t, err)
+	require.NoError(t, mock.ExpectationsWereMet())
+}
+
+func TestBulkUpdateCodexFingerprintSeedRollsBackWhenUpdateFails(t *testing.T) {
+	db, mock, err := sqlmock.New()
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+	client := dbent.NewClient(dbent.Driver(entsql.OpenDB(dialect.Postgres, db)))
+	t.Cleanup(func() { _ = client.Close() })
+
+	mock.ExpectBegin()
+	mock.ExpectExec(`(?s)UPDATE accounts SET extra = .*gen_random_uuid\(\)::text.*WHERE id = ANY\(\$2\)`).
+		WithArgs(sqlmock.AnyArg(), `{27,28}`).
+		WillReturnError(errors.New("update failed"))
+	mock.ExpectRollback()
+
+	repo := newAccountRepositoryWithSQL(client, db, nil)
+	rows, err := repo.BulkUpdate(context.Background(), []int64{27, 28}, service.AccountBulkUpdate{
+		Extra: map[string]any{
+			"codex_fingerprint_mode": "session",
+		},
+		EnsureCodexFingerprintSeed: true,
+	})
+
+	require.EqualError(t, err, "update failed")
+	require.Zero(t, rows)
+	require.NoError(t, mock.ExpectationsWereMet())
+}
+
+func TestBulkUpdateCodexFingerprintSeedRollsBackWhenOutboxFails(t *testing.T) {
+	db, mock, err := sqlmock.New()
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+	client := dbent.NewClient(dbent.Driver(entsql.OpenDB(dialect.Postgres, db)))
+	t.Cleanup(func() { _ = client.Close() })
+
+	mock.ExpectBegin()
+	mock.ExpectExec(`(?s)UPDATE accounts SET extra = .*gen_random_uuid\(\)::text.*WHERE id = ANY\(\$2\)`).
+		WithArgs(sqlmock.AnyArg(), `{27,28}`).
+		WillReturnResult(sqlmock.NewResult(0, 2))
+	mock.ExpectExec(regexp.QuoteMeta("INSERT INTO scheduler_outbox")).
+		WillReturnError(errors.New("outbox failed"))
+	mock.ExpectRollback()
+
+	repo := newAccountRepositoryWithSQL(client, db, nil)
+	rows, err := repo.BulkUpdate(context.Background(), []int64{27, 28}, service.AccountBulkUpdate{
+		Extra: map[string]any{
+			"codex_fingerprint_mode": "full",
+		},
+		EnsureCodexFingerprintSeed: true,
+	})
+
+	require.EqualError(t, err, "outbox failed")
+	require.Zero(t, rows)
+	require.NoError(t, mock.ExpectationsWereMet())
+}
diff --git a/backend/internal/repository/codex_fingerprint_seed_migration_integration_test.go b/backend/internal/repository/codex_fingerprint_seed_migration_integration_test.go
new file mode 100644
index 000000000000..70ae61eb8eee
--- /dev/null
+++ b/backend/internal/repository/codex_fingerprint_seed_migration_integration_test.go
@@ -0,0 +1,150 @@
+//go:build integration
+
+package repository
+
+import (
+	"context"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	dbmigrations "github.com/Wei-Shaw/sub2api/migrations"
+	"github.com/google/uuid"
+	"github.com/lib/pq"
+	"github.com/stretchr/testify/require"
+)
+
+func requireCanonicalUUIDString(t *testing.T, value string) {
+	t.Helper()
+	parsed, err := uuid.Parse(value)
+	require.NoError(t, err)
+	require.NotEqual(t, uuid.Nil, parsed)
+	require.Equal(t, parsed.String(), value)
+}
+
+func TestMigration225BackfillsOnlyEnabledOpenAIOAuthMissingOrMalformedSeeds(t *testing.T) {
+	tx := testTx(t)
+	ctx := context.Background()
+	migrationSQL, err := dbmigrations.FS.ReadFile("225_backfill_codex_fingerprint_seed.sql")
+	require.NoError(t, err)
+
+	var missingID, blankID, malformedID, validID, offID, apiKeyID int64
+	require.NoError(t, tx.QueryRowContext(ctx, `
+INSERT INTO accounts (name, platform, type, extra)
+VALUES ('migration-225-missing', 'openai', 'oauth', '{"codex_fingerprint_mode":"session"}'::jsonb)
+RETURNING id
+`).Scan(&missingID))
+	require.NoError(t, tx.QueryRowContext(ctx, `
+INSERT INTO accounts (name, platform, type, extra)
+VALUES ('migration-225-blank', 'openai', 'oauth', '{"codex_fingerprint_mode":"device","codex_fingerprint_seed":""}'::jsonb)
+RETURNING id
+`).Scan(&blankID))
+	require.NoError(t, tx.QueryRowContext(ctx, `
+INSERT INTO accounts (name, platform, type, extra)
+VALUES ('migration-225-malformed', 'openai', 'oauth', '{"codex_fingerprint_mode":"full","codex_fingerprint_seed":"BAD"}'::jsonb)
+RETURNING id
+`).Scan(&malformedID))
+	require.NoError(t, tx.QueryRowContext(ctx, `
+INSERT INTO accounts (name, platform, type, extra)
+VALUES ('migration-225-valid', 'openai', 'oauth', '{"codex_fingerprint_mode":"session","codex_fingerprint_seed":"11111111-1111-4111-8111-111111111111"}'::jsonb)
+RETURNING id
+`).Scan(&validID))
+	require.NoError(t, tx.QueryRowContext(ctx, `
+INSERT INTO accounts (name, platform, type, extra)
+VALUES ('migration-225-off', 'openai', 'oauth', '{"codex_fingerprint_mode":"off"}'::jsonb)
+RETURNING id
+`).Scan(&offID))
+	require.NoError(t, tx.QueryRowContext(ctx, `
+INSERT INTO accounts (name, platform, type, extra)
+VALUES ('migration-225-apikey', 'openai', 'apikey', '{"codex_fingerprint_mode":"session"}'::jsonb)
+RETURNING id
+`).Scan(&apiKeyID))
+
+	_, err = tx.ExecContext(ctx, string(migrationSQL))
+	require.NoError(t, err)
+
+	seedsAfterFirst := map[int64]string{}
+	for _, id := range []int64{missingID, blankID, malformedID, validID} {
+		var seed string
+		require.NoError(t, tx.QueryRowContext(ctx, `SELECT extra->>'codex_fingerprint_seed' FROM accounts WHERE id = $1`, id).Scan(&seed))
+		requireCanonicalUUIDString(t, seed)
+		seedsAfterFirst[id] = seed
+	}
+	require.Equal(t, "11111111-1111-4111-8111-111111111111", seedsAfterFirst[validID])
+
+	for _, id := range []int64{offID, apiKeyID} {
+		var hasSeed bool
+		require.NoError(t, tx.QueryRowContext(ctx, `SELECT extra ? 'codex_fingerprint_seed' FROM accounts WHERE id = $1`, id).Scan(&hasSeed))
+		require.False(t, hasSeed)
+	}
+
+	_, err = tx.ExecContext(ctx, string(migrationSQL))
+	require.NoError(t, err)
+
+	for id, want := range seedsAfterFirst {
+		var got string
+		require.NoError(t, tx.QueryRowContext(ctx, `SELECT extra->>'codex_fingerprint_seed' FROM accounts WHERE id = $1`, id).Scan(&got))
+		require.Equal(t, want, got)
+	}
+}
+
+func TestBulkUpdateGeneratesDistinctStableCodexFingerprintSeedsPerEligibleRow(t *testing.T) {
+	ctx := context.Background()
+	testName := "bulk-codex-seed-" + uuid.NewString()
+	type fixture struct {
+		name        string
+		accountType string
+		extra       string
+	}
+	fixtures := []fixture{
+		{name: testName + "-missing-a", accountType: service.AccountTypeOAuth, extra: `{}`},
+		{name: testName + "-missing-b", accountType: service.AccountTypeOAuth, extra: `{"codex_fingerprint_seed":"BAD"}`},
+		{name: testName + "-valid", accountType: service.AccountTypeOAuth, extra: `{"codex_fingerprint_seed":"11111111-1111-4111-8111-111111111111"}`},
+		{name: testName + "-apikey", accountType: service.AccountTypeAPIKey, extra: `{}`},
+	}
+
+	ids := make([]int64, 0, len(fixtures))
+	for _, f := range fixtures {
+		var id int64
+		require.NoError(t, integrationDB.QueryRowContext(ctx, `
+INSERT INTO accounts (name, platform, type, extra)
+VALUES ($1, 'openai', $2, $3::jsonb)
+RETURNING id
+`, f.name, f.accountType, f.extra).Scan(&id))
+		ids = append(ids, id)
+	}
+	t.Cleanup(func() {
+		_, _ = integrationDB.ExecContext(context.Background(), `DELETE FROM scheduler_outbox WHERE account_id = ANY($1)`, pq.Array(ids))
+		_, _ = integrationDB.ExecContext(context.Background(), `DELETE FROM accounts WHERE id = ANY($1)`, pq.Array(ids))
+	})
+
+	repo := newAccountRepositoryWithSQL(testEntClient(t), integrationDB, nil)
+	updates := service.AccountBulkUpdate{
+		Extra: map[string]any{
+			"codex_fingerprint_mode": "session",
+		},
+		EnsureCodexFingerprintSeed: true,
+	}
+	rows, err := repo.BulkUpdate(ctx, ids, updates)
+	require.NoError(t, err)
+	require.Equal(t, int64(len(ids)), rows)
+
+	readSeed := func(id int64) string {
+		t.Helper()
+		var seed string
+		require.NoError(t, integrationDB.QueryRowContext(ctx, `SELECT COALESCE(extra->>'codex_fingerprint_seed', '') FROM accounts WHERE id = $1`, id).Scan(&seed))
+		return seed
+	}
+	firstSeeds := []string{readSeed(ids[0]), readSeed(ids[1]), readSeed(ids[2]), readSeed(ids[3])}
+	requireCanonicalUUIDString(t, firstSeeds[0])
+	requireCanonicalUUIDString(t, firstSeeds[1])
+	require.NotEqual(t, firstSeeds[0], firstSeeds[1], "gen_random_uuid must be evaluated per eligible row")
+	require.Equal(t, "11111111-1111-4111-8111-111111111111", firstSeeds[2])
+	require.Empty(t, firstSeeds[3], "API-key accounts must not receive a Codex fingerprint seed")
+
+	rows, err = repo.BulkUpdate(ctx, ids, updates)
+	require.NoError(t, err)
+	require.Equal(t, int64(len(ids)), rows)
+	for i, want := range firstSeeds {
+		require.Equal(t, want, readSeed(ids[i]), "retry must not rotate an existing valid seed")
+	}
+}
diff --git a/backend/internal/repository/scheduler_cache.go b/backend/internal/repository/scheduler_cache.go
index b03e4e663718..b53c51d43f0e 100644
--- a/backend/internal/repository/scheduler_cache.go
+++ b/backend/internal/repository/scheduler_cache.go
@@ -1000,6 +1000,8 @@ func filterSchedulerExtra(extra map[string]any) map[string]any {
 		"openai_ws_force_http",
 		"openai_responses_mode",
 		"openai_responses_supported",
+		"codex_fingerprint_mode",
+		"codex_fingerprint_seed",
 		"codex_5h_used_percent",
 		"codex_7d_used_percent",
 		"codex_5h_reset_at",
diff --git a/backend/internal/repository/scheduler_cache_unit_test.go b/backend/internal/repository/scheduler_cache_unit_test.go
index d0373f16c7b2..354cf6d8aa91 100644
--- a/backend/internal/repository/scheduler_cache_unit_test.go
+++ b/backend/internal/repository/scheduler_cache_unit_test.go
@@ -309,6 +309,8 @@ func TestBuildSchedulerMetadataAccount_KeepsOpenAIWSFlags(t *testing.T) {
 			"openai_ws_force_http":                         true,
 			"openai_responses_mode":                        "force_chat_completions",
 			"openai_responses_supported":                   false,
+			"codex_fingerprint_mode":                       "session",
+			"codex_fingerprint_seed":                       "11111111-1111-4111-8111-111111111111",
 			"mixed_scheduling":                             true,
 			"unused_large_field":                           "drop-me",
 		},
@@ -321,6 +323,8 @@ func TestBuildSchedulerMetadataAccount_KeepsOpenAIWSFlags(t *testing.T) {
 	require.Equal(t, true, got.Extra["openai_ws_force_http"])
 	require.Equal(t, "force_chat_completions", got.Extra["openai_responses_mode"])
 	require.Equal(t, false, got.Extra["openai_responses_supported"])
+	require.Equal(t, "session", got.Extra["codex_fingerprint_mode"])
+	require.Equal(t, "11111111-1111-4111-8111-111111111111", got.Extra["codex_fingerprint_seed"])
 	require.Equal(t, true, got.Extra["mixed_scheduling"])
 	require.Nil(t, got.Extra["unused_large_field"])
 }
diff --git a/backend/internal/service/account_codex_fingerprint_seed_lifecycle_test.go b/backend/internal/service/account_codex_fingerprint_seed_lifecycle_test.go
new file mode 100644
index 000000000000..029b4fb009f9
--- /dev/null
+++ b/backend/internal/service/account_codex_fingerprint_seed_lifecycle_test.go
@@ -0,0 +1,245 @@
+package service
+
+import (
+	"context"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+const userSuppliedCodexFingerprintSeed = "22222222-2222-4222-8222-222222222222"
+
+func requireValidCodexFingerprintSeed(t *testing.T, extra map[string]any) string {
+	t.Helper()
+	seed, ok := codexFingerprintSeed(extra)
+	require.True(t, ok, "expected valid canonical Codex fingerprint seed")
+	return seed
+}
+
+func TestAdminCreateAccountStripsUserSeedAndCreatesFreshSeedWhenEnabled(t *testing.T) {
+	repo := &upstreamBillingProbeAccountRepo{}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	created, err := svc.CreateAccount(context.Background(), &CreateAccountInput{
+		Name:                 "codex-oauth",
+		Platform:             PlatformOpenAI,
+		Type:                 AccountTypeOAuth,
+		SkipDefaultGroupBind: true,
+		Extra: map[string]any{
+			codexFingerprintModeExtraKey: "session",
+			codexFingerprintSeedExtraKey: userSuppliedCodexFingerprintSeed,
+		},
+	})
+
+	require.NoError(t, err)
+	seed := requireValidCodexFingerprintSeed(t, created.Extra)
+	require.NotEqual(t, userSuppliedCodexFingerprintSeed, seed)
+	require.Equal(t, "session", created.Extra[codexFingerprintModeExtraKey])
+}
+
+func TestAdminUpdateAccountPreservesExistingSeedAndStripsUserSeed(t *testing.T) {
+	accountID := int64(201)
+	repo := &upstreamBillingProbeAccountRepo{accounts: map[int64]*Account{
+		accountID: {
+			ID:       accountID,
+			Name:     "before",
+			Platform: PlatformOpenAI,
+			Type:     AccountTypeOAuth,
+			Status:   StatusActive,
+			Extra: map[string]any{
+				codexFingerprintModeExtraKey: "session",
+				codexFingerprintSeedExtraKey: testCodexFingerprintSeed,
+			},
+		},
+	}}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	updated, err := svc.UpdateAccount(context.Background(), accountID, &UpdateAccountInput{
+		Extra: map[string]any{
+			codexFingerprintModeExtraKey: "full",
+			codexFingerprintSeedExtraKey: userSuppliedCodexFingerprintSeed,
+			"custom":                     "value",
+		},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, testCodexFingerprintSeed, requireValidCodexFingerprintSeed(t, updated.Extra))
+	require.Equal(t, "full", updated.Extra[codexFingerprintModeExtraKey])
+	require.Equal(t, "value", updated.Extra["custom"])
+}
+
+func TestAdminUpdateAccountInitializesSeedWhenFullEditEnables(t *testing.T) {
+	accountID := int64(202)
+	repo := &upstreamBillingProbeAccountRepo{accounts: map[int64]*Account{
+		accountID: {
+			ID:       accountID,
+			Name:     "before",
+			Platform: PlatformOpenAI,
+			Type:     AccountTypeOAuth,
+			Status:   StatusActive,
+			Extra: map[string]any{
+				codexFingerprintModeExtraKey: "off",
+				codexFingerprintSeedExtraKey: "not-a-seed",
+			},
+		},
+	}}
+
+	updated, err := (&adminServiceImpl{accountRepo: repo}).UpdateAccount(context.Background(), accountID, &UpdateAccountInput{
+		Extra: map[string]any{codexFingerprintModeExtraKey: "device"},
+	})
+
+	require.NoError(t, err)
+	require.NotEqual(t, "not-a-seed", requireValidCodexFingerprintSeed(t, updated.Extra))
+	require.Equal(t, "device", updated.Extra[codexFingerprintModeExtraKey])
+}
+
+func TestAdminUpdateAccountDisableReenablePreservesValidSeed(t *testing.T) {
+	accountID := int64(203)
+	repo := &upstreamBillingProbeAccountRepo{accounts: map[int64]*Account{
+		accountID: {
+			ID:       accountID,
+			Platform: PlatformOpenAI,
+			Type:     AccountTypeOAuth,
+			Status:   StatusActive,
+			Extra: map[string]any{
+				codexFingerprintModeExtraKey: "session",
+				codexFingerprintSeedExtraKey: testCodexFingerprintSeed,
+			},
+		},
+	}}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	disabled, err := svc.UpdateAccount(context.Background(), accountID, &UpdateAccountInput{
+		Extra: map[string]any{codexFingerprintModeExtraKey: "off"},
+	})
+	require.NoError(t, err)
+	require.Equal(t, testCodexFingerprintSeed, requireValidCodexFingerprintSeed(t, disabled.Extra))
+
+	reenabled, err := svc.UpdateAccount(context.Background(), accountID, &UpdateAccountInput{
+		Extra: map[string]any{codexFingerprintModeExtraKey: "session"},
+	})
+	require.NoError(t, err)
+	require.Equal(t, testCodexFingerprintSeed, requireValidCodexFingerprintSeed(t, reenabled.Extra))
+}
+
+func TestAdminUpdateAccountExtraStripsSeedAndLeavesAtomicEnsureToRepository(t *testing.T) {
+	accountID := int64(204)
+	repo := &upstreamBillingProbeAccountRepo{accounts: map[int64]*Account{
+		accountID: {
+			ID:       accountID,
+			Platform: PlatformOpenAI,
+			Type:     AccountTypeOAuth,
+			Extra:    map[string]any{},
+		},
+	}}
+
+	err := (&adminServiceImpl{accountRepo: repo}).UpdateAccountExtra(context.Background(), accountID, map[string]any{
+		codexFingerprintModeExtraKey: "device",
+		codexFingerprintSeedExtraKey: userSuppliedCodexFingerprintSeed,
+	})
+
+	require.NoError(t, err)
+	require.Len(t, repo.updates[accountID], 1)
+	require.Equal(t, "device", repo.updates[accountID][0][codexFingerprintModeExtraKey])
+	require.NotContains(t, repo.updates[accountID][0], codexFingerprintSeedExtraKey)
+}
+
+func TestBulkUpdateAccountsDoesNotPrewriteCodexSeed(t *testing.T) {
+	repo := &upstreamBillingProbeAccountRepo{}
+
+	result, err := (&adminServiceImpl{accountRepo: repo}).BulkUpdateAccounts(context.Background(), &BulkUpdateAccountsInput{
+		AccountIDs: []int64{301, 302},
+		Extra: map[string]any{
+			codexFingerprintModeExtraKey: "session",
+			codexFingerprintSeedExtraKey: userSuppliedCodexFingerprintSeed,
+		},
+	})
+
+	require.NoError(t, err)
+	require.Equal(t, 2, result.Success)
+	require.Empty(t, repo.updates, "bulk enable must not loop through UpdateExtra before BulkUpdate")
+	require.Len(t, repo.bulkUpdates, 1)
+	require.True(t, repo.bulkUpdates[0].EnsureCodexFingerprintSeed)
+	require.Equal(t, "session", repo.bulkUpdates[0].Extra[codexFingerprintModeExtraKey])
+	require.NotContains(t, repo.bulkUpdates[0].Extra, codexFingerprintSeedExtraKey)
+}
+
+type codexSeedDuplicateRepo struct {
+	*upstreamBillingProbeAccountRepo
+}
+
+func (r *codexSeedDuplicateRepo) CreateWithAccountGroups(ctx context.Context, account *Account, _ []AccountGroup) error {
+	return r.Create(ctx, account)
+}
+
+func TestDuplicateAccountDoesNotCopyCodexFingerprintSeed(t *testing.T) {
+	ctx := context.Background()
+	repo := &codexSeedDuplicateRepo{upstreamBillingProbeAccountRepo: &upstreamBillingProbeAccountRepo{accounts: make(map[int64]*Account)}}
+	svc := &adminServiceImpl{accountRepo: repo, accountDuplicateRepo: repo}
+	source := &Account{
+		Name:     "source",
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeAPIKey,
+		Extra: map[string]any{
+			codexFingerprintModeExtraKey: "session",
+			codexFingerprintSeedExtraKey: testCodexFingerprintSeed,
+		},
+	}
+	require.NoError(t, repo.Create(ctx, source))
+
+	duplicate, err := svc.DuplicateAccount(ctx, source.ID, "admin:1", "")
+
+	require.NoError(t, err)
+	require.NotEqual(t, source.ID, duplicate.ID)
+	require.NotContains(t, duplicate.Extra, codexFingerprintSeedExtraKey)
+	require.Equal(t, "session", duplicate.Extra[codexFingerprintModeExtraKey])
+}
+
+func TestDuplicateCreatePathMintsFreshSeedWhenEligible(t *testing.T) {
+	extra, err := duplicateAccountExtra(map[string]any{
+		codexFingerprintModeExtraKey: "session",
+		codexFingerprintSeedExtraKey: testCodexFingerprintSeed,
+	})
+	require.NoError(t, err)
+
+	account, err := buildAccountForCreate(&CreateAccountInput{
+		Name:     "eligible-copy",
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeOAuth,
+		Extra:    extra,
+	}, extra)
+
+	require.NoError(t, err)
+	require.NotEqual(t, testCodexFingerprintSeed, requireValidCodexFingerprintSeed(t, account.Extra))
+	require.Equal(t, "session", account.Extra[codexFingerprintModeExtraKey])
+}
+
+func TestAccountServiceCreateAndUpdateCodexSeedLifecycle(t *testing.T) {
+	ctx := context.Background()
+	repo := &upstreamBillingProbeAccountRepo{accounts: make(map[int64]*Account)}
+	svc := NewAccountService(repo, nil)
+
+	created, err := svc.Create(ctx, CreateAccountRequest{
+		Name:     "legacy-create",
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeOAuth,
+		Extra: map[string]any{
+			codexFingerprintModeExtraKey: "session",
+			codexFingerprintSeedExtraKey: userSuppliedCodexFingerprintSeed,
+		},
+	})
+	require.NoError(t, err)
+	createdSeed := requireValidCodexFingerprintSeed(t, created.Extra)
+	require.NotEqual(t, userSuppliedCodexFingerprintSeed, createdSeed)
+
+	updateSeed := userSuppliedCodexFingerprintSeed
+	updated, err := svc.Update(ctx, created.ID, UpdateAccountRequest{
+		Extra: &map[string]any{
+			codexFingerprintModeExtraKey: "full",
+			codexFingerprintSeedExtraKey: updateSeed,
+		},
+	})
+	require.NoError(t, err)
+	require.Equal(t, createdSeed, requireValidCodexFingerprintSeed(t, updated.Extra))
+	require.Equal(t, "full", updated.Extra[codexFingerprintModeExtraKey])
+}
diff --git a/backend/internal/service/account_service.go b/backend/internal/service/account_service.go
index be199cb2561c..96689e5000fd 100644
--- a/backend/internal/service/account_service.go
+++ b/backend/internal/service/account_service.go
@@ -166,6 +166,9 @@ type AccountBulkUpdate struct {
 	Credentials    map[string]any
 	Extra          map[string]any
 	ProbeEnabled   *bool
+	// EnsureCodexFingerprintSeed asks the repository to atomically preserve an
+	// existing valid Codex fingerprint seed or create one for eligible rows.
+	EnsureCodexFingerprintSeed bool
 }
 
 // CreateAccountRequest 创建账号请求
@@ -233,7 +236,7 @@ func (s *AccountService) Create(ctx context.Context, req CreateAccountRequest) (
 		Platform:    req.Platform,
 		Type:        req.Type,
 		Credentials: SanitizeStoredCredentials(req.Platform, req.Credentials),
-		Extra:       req.Extra,
+		Extra:       prepareCodexFingerprintExtraForCreate(req.Platform, req.Type, req.Extra),
 		ProxyID:     req.ProxyID,
 		Concurrency: req.Concurrency,
 		Priority:    req.Priority,
@@ -336,7 +339,9 @@ func (s *AccountService) Update(ctx context.Context, id int64, req UpdateAccount
 		delete(extra, OllamaCloudUsageSessionExtraKey)
 		delete(extra, OllamaCloudUsageAutoRefreshExtraKey)
 		delete(extra, OllamaCloudUsageSnapshotExtraKey)
-		account.Extra = extra
+		account.Extra = prepareCodexFingerprintExtraForUpdate(account, extra)
+	} else {
+		account.Extra = prepareCodexFingerprintExtraForUpdate(account, account.Extra)
 	}
 
 	if req.ProxyID != nil {
diff --git a/backend/internal/service/account_test_service_openai_compact_test.go b/backend/internal/service/account_test_service_openai_compact_test.go
index 5b4772508439..aa2b5f70bd86 100644
--- a/backend/internal/service/account_test_service_openai_compact_test.go
+++ b/backend/internal/service/account_test_service_openai_compact_test.go
@@ -284,7 +284,10 @@ func TestAccountTestService_TestAccountConnection_OpenAICompactProbeIdentityMatc
 			"chatgpt_account_id": "chatgpt-acc",
 		},
 		// 收敛是显式 opt-in（#5610），这里显式开启以验证探测身份与真实流量同构。
-		Extra: map[string]any{"codex_fingerprint_mode": "session"},
+		Extra: map[string]any{
+			"codex_fingerprint_mode":     "session",
+			codexFingerprintSeedExtraKey: testCodexFingerprintSeed,
+		},
 	}
 	repo := &snapshotUpdateAccountRepo{
 		stubOpenAIAccountRepo: stubOpenAIAccountRepo{accounts: []Account{account}},
@@ -304,10 +307,12 @@ func TestAccountTestService_TestAccountConnection_OpenAICompactProbeIdentityMatc
 	require.NoError(t, svc.TestAccountConnection(c, account.ID, "gpt-5.4", "", AccountTestModeCompact))
 
 	// 显式 session 收敛模式：出站身份 = 账号级收敛值
-	converged := resolveConvergedSessionID(&account)
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	converged := resolveConvergedSessionID(seed)
 	require.Equal(t, converged, upstream.lastReq.Header.Get("session-id"))
 	require.Equal(t, converged, upstream.lastReq.Header.Get("session_id"))
-	require.Equal(t, resolveConvergedInstallationID(&account), upstream.lastReq.Header.Get("x-codex-installation-id"),
+	require.Equal(t, resolveConvergedInstallationID(&account, seed), upstream.lastReq.Header.Get("x-codex-installation-id"),
 		"真实 Codex 每个请求必带 installation-id，探测不得缺失")
 	require.NotContains(t, upstream.lastReq.Header.Get("session-id"), "probe_compact",
 		"探测标识不得是可被上游一眼识别的字面量")
diff --git a/backend/internal/service/admin_account.go b/backend/internal/service/admin_account.go
index 9958846658ae..ebfe861f1739 100644
--- a/backend/internal/service/admin_account.go
+++ b/backend/internal/service/admin_account.go
@@ -128,22 +128,24 @@ var duplicateAccountDiscardedExtraKeys = map[string]struct{}{
 	"drive_storage_limit":                    {},
 	"drive_storage_usage":                    {},
 	"drive_tier_updated_at":                  {},
-	"codex_primary_used_percent":             {},
-	"codex_primary_reset_after_seconds":      {},
-	"codex_primary_window_minutes":           {},
-	"codex_secondary_used_percent":           {},
-	"codex_secondary_reset_after_seconds":    {},
-	"codex_secondary_window_minutes":         {},
-	"codex_primary_over_secondary_percent":   {},
-	"codex_usage_updated_at":                 {},
-	"codex_5h_used_percent":                  {},
-	"codex_5h_reset_after_seconds":           {},
-	"codex_5h_window_minutes":                {},
-	"codex_5h_reset_at":                      {},
-	"codex_7d_used_percent":                  {},
-	"codex_7d_reset_after_seconds":           {},
-	"codex_7d_window_minutes":                {},
-	"codex_7d_reset_at":                      {},
+	// Codex fingerprint convergence uses a per-account random seed, never copied from another account.
+	codexFingerprintSeedExtraKey:           {},
+	"codex_primary_used_percent":           {},
+	"codex_primary_reset_after_seconds":    {},
+	"codex_primary_window_minutes":         {},
+	"codex_secondary_used_percent":         {},
+	"codex_secondary_reset_after_seconds":  {},
+	"codex_secondary_window_minutes":       {},
+	"codex_primary_over_secondary_percent": {},
+	"codex_usage_updated_at":               {},
+	"codex_5h_used_percent":                {},
+	"codex_5h_reset_after_seconds":         {},
+	"codex_5h_window_minutes":              {},
+	"codex_5h_reset_at":                    {},
+	"codex_7d_used_percent":                {},
+	"codex_7d_reset_after_seconds":         {},
+	"codex_7d_window_minutes":              {},
+	"codex_7d_reset_at":                    {},
 }
 
 func duplicateAccountExtra(value map[string]any) (map[string]any, error) {
@@ -404,6 +406,7 @@ func buildAccountForCreate(input *CreateAccountInput, accountExtra map[string]an
 	delete(accountExtra, OllamaCloudUsageSessionExtraKey)
 	delete(accountExtra, OllamaCloudUsageAutoRefreshExtraKey)
 	delete(accountExtra, OllamaCloudUsageSnapshotExtraKey)
+	accountExtra = prepareCodexFingerprintExtraForCreate(input.Platform, input.Type, accountExtra)
 	account := &Account{
 		Name:        input.Name,
 		Notes:       normalizeAccountNotes(input.Notes),
@@ -651,6 +654,7 @@ func (s *adminServiceImpl) UpdateAccount(ctx context.Context, id int64, input *U
 				normalizedExtra[key] = v
 			}
 		}
+		normalizedExtra = prepareCodexFingerprintExtraForUpdate(account, normalizedExtra)
 		account.Extra = normalizedExtra
 		if account.Platform == PlatformAntigravity && wasOveragesEnabled && !account.IsOveragesEnabled() {
 			delete(account.Extra, "antigravity_credits_overages") // 清理旧版 overages 运行态
@@ -670,6 +674,9 @@ func (s *adminServiceImpl) UpdateAccount(ctx context.Context, id int64, input *U
 		ComputeQuotaResetAt(account.Extra)
 		NormalizeFixedQuotaWindows(account.Extra)
 	}
+	if input.Extra == nil {
+		account.Extra = prepareCodexFingerprintExtraForUpdate(account, account.Extra)
+	}
 	if requestedRateSyncEnabledUpdate != nil && *requestedRateSyncEnabledUpdate {
 		if requestedProbeEnabledUpdate != nil && !*requestedProbeEnabledUpdate {
 			return nil, infraerrors.BadRequest(
@@ -852,6 +859,7 @@ func (s *adminServiceImpl) UpdateAccount(ctx context.Context, id int64, input *U
 // UpdateAccountExtra 仅对 Extra JSONB 做 key 级合并，避免覆盖其它运行态键
 // （如 model_rate_limits / passive_usage_* 等）。
 func (s *adminServiceImpl) UpdateAccountExtra(ctx context.Context, id int64, updates map[string]any) error {
+	updates = sanitizedCodexFingerprintExtraUpdates(updates)
 	delete(updates, UpstreamBillingProbeEnabledExtraKey)
 	delete(updates, UpstreamBillingRateSyncEnabledExtraKey)
 	delete(updates, UpstreamBillingProbeExtraKey)
@@ -877,6 +885,7 @@ func (s *adminServiceImpl) UpdateAccountExtra(ctx context.Context, id int64, upd
 // It merges credentials/extra keys instead of overwriting the whole object.
 func (s *adminServiceImpl) BulkUpdateAccounts(ctx context.Context, input *BulkUpdateAccountsInput) (*BulkUpdateAccountsResult, error) {
 	// Managed probe/session state may only enter through dedicated typed endpoints.
+	input.Extra = sanitizedCodexFingerprintExtraUpdates(input.Extra)
 	delete(input.Extra, UpstreamBillingProbeEnabledExtraKey)
 	delete(input.Extra, UpstreamBillingRateSyncEnabledExtraKey)
 	delete(input.Extra, UpstreamBillingProbeExtraKey)
@@ -1027,9 +1036,10 @@ func (s *adminServiceImpl) BulkUpdateAccounts(ctx context.Context, input *BulkUp
 
 	// Prepare bulk updates for columns and JSONB fields.
 	repoUpdates := AccountBulkUpdate{
-		Credentials:  input.Credentials,
-		Extra:        input.Extra,
-		ProbeEnabled: input.ProbeEnabled,
+		Credentials:                input.Credentials,
+		Extra:                      input.Extra,
+		ProbeEnabled:               input.ProbeEnabled,
+		EnsureCodexFingerprintSeed: ShouldEnsureCodexFingerprintSeedForExtraUpdates(input.Extra),
 	}
 	if input.ProbeEnabled != nil {
 		if repoUpdates.Extra == nil {
diff --git a/backend/internal/service/openai_codex_fingerprint.go b/backend/internal/service/openai_codex_fingerprint.go
index c74669c00c40..0f841fb2fe29 100644
--- a/backend/internal/service/openai_codex_fingerprint.go
+++ b/backend/internal/service/openai_codex_fingerprint.go
@@ -5,6 +5,7 @@ import (
 	"encoding/binary"
 	"encoding/json"
 	"fmt"
+	"maps"
 	"net/http"
 	"strings"
 	"time"
@@ -30,20 +31,30 @@ func stageCodexFingerprintIDs(c *gin.Context, ids *codexFingerprintIDs) {
 	}
 }
 
-// applyStagedCodexFingerprintHeaders 读取 context 暂存的收敛 ID 并改写出站头。
-// 非透传与透传两个请求构造器共用本函数，防止应用语义漂移。仅 OAuth 账号
-// 生效（stale 键在账号类型混合 failover 下由该门挡住）。
-func applyStagedCodexFingerprintHeaders(c *gin.Context, account *Account, h http.Header) {
+func stagedCodexFingerprintIDs(c *gin.Context, account *Account) *codexFingerprintIDs {
 	if c == nil || account == nil || account.Type != AccountTypeOAuth {
-		return
+		return nil
 	}
 	value, ok := c.Get(codexFingerprintIDsContextKey)
 	if !ok {
-		return
+		return nil
 	}
-	if ids, ok := value.(*codexFingerprintIDs); ok {
-		applyCodexFingerprintHeaders(h, ids)
+	ids, ok := value.(*codexFingerprintIDs)
+	if !ok || ids == nil || ids.accountID != account.ID {
+		return nil
 	}
+	return ids
+}
+
+// applyStagedCodexFingerprintHeaders 读取 context 暂存的收敛 ID 并改写出站头。
+// 非透传与透传两个请求构造器共用本函数，防止应用语义漂移。仅解析该
+// snapshot 的 OAuth 账号可读取，避免 stale context 跨账号 failover 泄漏。
+func applyStagedCodexFingerprintHeaders(c *gin.Context, account *Account, h http.Header) {
+	applyCodexFingerprintHeaders(h, stagedCodexFingerprintIDs(c, account))
+}
+
+func applyStagedCodexFingerprintClientMetadata(c *gin.Context, account *Account, reqBody map[string]any) bool {
+	return applyCodexFingerprintClientMetadata(reqBody, stagedCodexFingerprintIDs(c, account))
 }
 
 // codexFingerprintMode 控制 OAuth 账号出站请求的设备指纹收敛强度。
@@ -68,7 +79,117 @@ const (
 	codexFingerprintFull codexFingerprintMode = "full"
 )
 
-const codexFingerprintModeExtraKey = "codex_fingerprint_mode"
+const (
+	codexFingerprintModeExtraKey = "codex_fingerprint_mode"
+	codexFingerprintSeedExtraKey = "codex_fingerprint_seed"
+)
+
+func canonicalCodexFingerprintSeed(value any) (string, bool) {
+	raw, ok := value.(string)
+	if !ok {
+		return "", false
+	}
+	trimmed := strings.TrimSpace(raw)
+	parsed, err := uuid.Parse(trimmed)
+	if err != nil || parsed == uuid.Nil || trimmed != parsed.String() {
+		return "", false
+	}
+	return trimmed, true
+}
+
+func newCodexFingerprintSeed() string {
+	return uuid.NewString()
+}
+
+func stripCodexFingerprintSeed(extra map[string]any) map[string]any {
+	if extra == nil {
+		return nil
+	}
+	stripped := maps.Clone(extra)
+	delete(stripped, codexFingerprintSeedExtraKey)
+	return stripped
+}
+
+func codexFingerprintModeFromExtra(extra map[string]any) codexFingerprintMode {
+	if extra == nil {
+		return codexFingerprintOff
+	}
+	raw, _ := extra[codexFingerprintModeExtraKey].(string)
+	switch codexFingerprintMode(strings.TrimSpace(raw)) {
+	case codexFingerprintOff, codexFingerprintDevice, codexFingerprintSession, codexFingerprintFull:
+		return codexFingerprintMode(strings.TrimSpace(raw))
+	default:
+		return codexFingerprintOff
+	}
+}
+
+func codexFingerprintModeRequiresSeed(mode codexFingerprintMode) bool {
+	switch mode {
+	case codexFingerprintDevice, codexFingerprintSession, codexFingerprintFull:
+		return true
+	default:
+		return false
+	}
+}
+
+func codexFingerprintSeed(extra map[string]any) (string, bool) {
+	if extra == nil {
+		return "", false
+	}
+	return canonicalCodexFingerprintSeed(extra[codexFingerprintSeedExtraKey])
+}
+
+func prepareCodexFingerprintExtraForCreate(platform, accountType string, extra map[string]any) map[string]any {
+	prepared := stripCodexFingerprintSeed(extra)
+	if platform != PlatformOpenAI || accountType != AccountTypeOAuth || !codexFingerprintModeRequiresSeed(codexFingerprintModeFromExtra(prepared)) {
+		return prepared
+	}
+	if prepared == nil {
+		prepared = make(map[string]any, 1)
+	}
+	prepared[codexFingerprintSeedExtraKey] = newCodexFingerprintSeed()
+	return prepared
+}
+
+func prepareCodexFingerprintExtraForUpdate(account *Account, extra map[string]any) map[string]any {
+	prepared := stripCodexFingerprintSeed(extra)
+	if account == nil || account.Platform != PlatformOpenAI || account.Type != AccountTypeOAuth {
+		return prepared
+	}
+	if seed, ok := codexFingerprintSeed(account.Extra); ok {
+		if prepared == nil {
+			prepared = make(map[string]any, 1)
+		}
+		prepared[codexFingerprintSeedExtraKey] = seed
+		return prepared
+	}
+	if codexFingerprintModeRequiresSeed(codexFingerprintModeFromExtra(prepared)) {
+		if prepared == nil {
+			prepared = make(map[string]any, 1)
+		}
+		prepared[codexFingerprintSeedExtraKey] = newCodexFingerprintSeed()
+	}
+	return prepared
+}
+
+func sanitizedCodexFingerprintExtraUpdates(updates map[string]any) map[string]any {
+	if updates == nil {
+		return nil
+	}
+	sanitized := maps.Clone(updates)
+	delete(sanitized, codexFingerprintSeedExtraKey)
+	return sanitized
+}
+
+// ShouldEnsureCodexFingerprintSeedForExtraUpdates reports whether a JSONB key-level
+// extra update is enabling Codex fingerprint convergence and therefore must atomically
+// preserve or create the system-managed per-account seed in the repository update.
+func ShouldEnsureCodexFingerprintSeedForExtraUpdates(updates map[string]any) bool {
+	if updates == nil {
+		return false
+	}
+	return codexFingerprintModeRequiresSeed(codexFingerprintModeFromExtra(updates))
+}
 
 // GetCodexFingerprintMode 从账号 extra JSON 读取指纹收敛模式。
 //
@@ -85,13 +206,7 @@ func (a *Account) GetCodexFingerprintMode() codexFingerprintMode {
 	if a == nil || !a.IsOpenAIOAuth() {
 		return codexFingerprintOff
 	}
-	raw := strings.TrimSpace(a.GetExtraString(codexFingerprintModeExtraKey))
-	switch codexFingerprintMode(raw) {
-	case codexFingerprintOff, codexFingerprintDevice, codexFingerprintSession, codexFingerprintFull:
-		return codexFingerprintMode(raw)
-	default:
-		return codexFingerprintOff
-	}
+	return codexFingerprintModeFromExtra(a.Extra)
 }
 
 // deriveStableUUIDv4 从种子确定性派生一个 UUIDv4 格式的字符串。
@@ -110,45 +225,53 @@ func deriveStableUUIDv4(seed string) string {
 }
 
 // resolveConvergedInstallationID 返回账号级恒定的 installation_id。
-// 优先使用管理员配置的真实 device_id，无则从 accountID 确定性派生。
-func resolveConvergedInstallationID(account *Account) string {
+// 优先使用管理员配置的真实 device_id，无则从系统管理的账号随机种子确定性派生。
+func resolveConvergedInstallationID(account *Account, seed string) string {
 	if account == nil {
 		return ""
 	}
 	if deviceID := account.GetOpenAIDeviceID(); deviceID != "" {
 		return deviceID
 	}
-	return deriveStableUUIDv4(fmt.Sprintf("sub2api:codex-install-id:v1:%d", account.ID))
+	if seed == "" {
+		return ""
+	}
+	return deriveStableUUIDv4("sub2api:codex-install-id:v2:" + seed)
 }
 
 // resolveConvergedSessionID 返回账号级恒定的 session_id。
-func resolveConvergedSessionID(account *Account) string {
-	if account == nil {
+func resolveConvergedSessionID(seed string) string {
+	if seed == "" {
 		return ""
 	}
-	return deriveStableUUIDv4(fmt.Sprintf("sub2api:codex-session-id:v1:%d", account.ID))
+	return deriveStableUUIDv4("sub2api:codex-session-id:v2:" + seed)
 }
 
 // resolveConvergedThreadID 按客户端原始 session-id 确定性派生 thread_id。
 // 每个真实 Codex 会话（不同客户端启动实例）获得一个独立线程，
 // 模拟正常用户 spawn 子代理或开多窗口的模式。
-func resolveConvergedThreadID(account *Account, clientSessionID string) string {
-	if account == nil || clientSessionID == "" {
+func resolveConvergedThreadID(seed, clientSessionID string) string {
+	if seed == "" || clientSessionID == "" {
 		return ""
 	}
-	return deriveStableUUIDv4(fmt.Sprintf("sub2api:codex-thread-id:v1:%d:%s", account.ID, clientSessionID))
+	return deriveStableUUIDv4("sub2api:codex-thread-id:v2:" + seed + ":" + clientSessionID)
 }
 
 // codexFingerprintIDs 收敛后的完整 ID 集合。
 // 由 resolveCodexFingerprintIDs 一次性生成，同一个实例在头改写和体改写之间共享，
-// 确保所有载体中的 turn_id 等随机字段一致。
+// 确保所有载体中的 turn_id 等随机字段一致。体改写时还会补记原始
+// client_metadata.session_id，用于识别 root prompt_cache_key 的默认值。
 type codexFingerprintIDs struct {
-	mode           codexFingerprintMode
-	installationID string
-	sessionID      string
-	threadID       string
-	turnID         string
-	windowID       string
+	accountID                     int64
+	mode                          codexFingerprintMode
+	installationID                string
+	sessionID                     string
+	threadID                      string
+	turnID                        string
+	windowID                      string
+	turnStartedAtUnixMs           int64
+	originalBodySessionID         string
+	originalBodySessionIDCaptured bool
 }
 
 // resolveCodexFingerprintIDs 按收敛模式计算出站 ID 集合。
@@ -157,13 +280,21 @@ type codexFingerprintIDs struct {
 // 返回 nil 表示 off 模式，不需要改写。
 // 注意：包含随机生成的 turn_id，调用方必须只调用一次并共享结果给头改写和体改写。
 func resolveCodexFingerprintIDs(account *Account, clientSessionID string, mode codexFingerprintMode) *codexFingerprintIDs {
-	if mode == codexFingerprintOff {
+	if account == nil || mode == codexFingerprintOff {
+		return nil
+	}
+	seed, ok := codexFingerprintSeed(account.Extra)
+	if !ok {
 		return nil
 	}
 
-	ids := &codexFingerprintIDs{mode: mode}
+	ids := &codexFingerprintIDs{
+		accountID:           account.ID,
+		mode:                mode,
+		turnStartedAtUnixMs: time.Now().UnixMilli(),
+	}
 
-	ids.installationID = resolveConvergedInstallationID(account)
+	ids.installationID = resolveConvergedInstallationID(account, seed)
 	if ids.installationID == "" {
 		return nil
 	}
@@ -173,8 +304,8 @@ func resolveCodexFingerprintIDs(account *Account, clientSessionID string, mode c
 		return ids
 
 	case codexFingerprintSession:
-		ids.sessionID = resolveConvergedSessionID(account)
-		ids.threadID = resolveConvergedThreadID(account, clientSessionID)
+		ids.sessionID = resolveConvergedSessionID(seed)
+		ids.threadID = resolveConvergedThreadID(seed, clientSessionID)
 		if ids.threadID == "" {
 			ids.threadID = ids.sessionID
 		}
@@ -183,7 +314,7 @@ func resolveCodexFingerprintIDs(account *Account, clientSessionID string, mode c
 		return ids
 
 	case codexFingerprintFull:
-		ids.sessionID = resolveConvergedSessionID(account)
+		ids.sessionID = resolveConvergedSessionID(seed)
 		ids.threadID = ids.sessionID
 		ids.turnID = uuid.Must(uuid.NewV7()).String()
 		ids.windowID = ids.threadID + ":0"
@@ -252,20 +383,21 @@ func applyCodexFingerprintHeaders(h http.Header, ids *codexFingerprintIDs) {
 		"thread_id":               ids.threadID,
 		"turn_id":                 ids.turnID,
 		"window_id":               ids.windowID,
-		"turn_started_at_unix_ms": time.Now().UnixMilli(),
+		"turn_started_at_unix_ms": ids.turnStartedAtUnixMs,
 	})
 }
 
 // rewriteCodexTurnMetadataFields 解析 x-codex-turn-metadata 头中的 JSON，
-// 替换指定字段后回写。保留未指定字段原样（如 sandbox、thread_source 等）。
+// 替换指定字段后回写。合法对象保留未指定字段（如 sandbox、thread_source）；
+// 非法/非对象值重建为最小合法 metadata，避免 flat 与 embedded identity 分裂。
 func rewriteCodexTurnMetadataFields(h http.Header, fields map[string]any) {
 	raw := strings.TrimSpace(h.Get("x-codex-turn-metadata"))
 	if raw == "" {
 		return
 	}
 	var metadata map[string]any
-	if err := json.Unmarshal([]byte(raw), &metadata); err != nil {
-		return
+	if err := json.Unmarshal([]byte(raw), &metadata); err != nil || metadata == nil {
+		metadata = make(map[string]any, len(fields))
 	}
 	for k, v := range fields {
 		metadata[k] = v
@@ -284,16 +416,21 @@ func applyCodexFingerprintClientMetadata(reqBody map[string]any, ids *codexFinge
 		return false
 	}
 
+	captureCodexFingerprintOriginalBodySessionID(ids, reqBody["client_metadata"])
 	existing, _ := reqBody["client_metadata"].(map[string]any)
 	if existing == nil {
 		existing = make(map[string]any)
 	}
 
-	if !applyCodexFingerprintToClientMetadataMap(existing, ids) {
-		return false
+	modified := false
+	if applyCodexFingerprintToClientMetadataMap(existing, ids) {
+		reqBody["client_metadata"] = existing
+		modified = true
 	}
-	reqBody["client_metadata"] = existing
-	return true
+	if applyCodexFingerprintPromptCacheKey(reqBody, ids) {
+		modified = true
+	}
+	return modified
 }
 
 // applyCodexFingerprintToClientMetadataMap 是 client_metadata 改写的共享核心，
@@ -330,59 +467,130 @@ func applyCodexFingerprintToClientMetadataMap(existing map[string]any, ids *code
 		"thread_id":               ids.threadID,
 		"turn_id":                 ids.turnID,
 		"window_id":               ids.windowID,
-		"turn_started_at_unix_ms": time.Now().UnixMilli(),
+		"turn_started_at_unix_ms": ids.turnStartedAtUnixMs,
 	})
 	return true
 }
 
+func captureCodexFingerprintOriginalBodySessionID(ids *codexFingerprintIDs, clientMetadata any) {
+	if ids == nil || ids.originalBodySessionIDCaptured {
+		return
+	}
+	ids.originalBodySessionIDCaptured = true
+	if clientMetadata == nil {
+		return
+	}
+	switch metadata := clientMetadata.(type) {
+	case map[string]any:
+		if sessionID, ok := metadata["session_id"].(string); ok {
+			ids.originalBodySessionID = strings.TrimSpace(sessionID)
+		}
+	case map[string]string:
+		ids.originalBodySessionID = strings.TrimSpace(metadata["session_id"])
+	}
+}
+
+func captureCodexFingerprintOriginalBodySessionIDRaw(ids *codexFingerprintIDs, value gjson.Result) {
+	if ids == nil || ids.originalBodySessionIDCaptured {
+		return
+	}
+	ids.originalBodySessionIDCaptured = true
+	if value.Exists() && value.Type == gjson.String {
+		ids.originalBodySessionID = strings.TrimSpace(value.String())
+	}
+}
+
+func shouldRewriteCodexFingerprintPromptCacheKey(ids *codexFingerprintIDs, promptCacheKey string) bool {
+	if ids == nil || !ids.originalBodySessionIDCaptured || ids.originalBodySessionID == "" || ids.sessionID == "" {
+		return false
+	}
+	if ids.mode != codexFingerprintSession && ids.mode != codexFingerprintFull {
+		return false
+	}
+	return promptCacheKey == ids.originalBodySessionID
+}
+
+func applyCodexFingerprintPromptCacheKey(reqBody map[string]any, ids *codexFingerprintIDs) bool {
+	if reqBody == nil {
+		return false
+	}
+	promptCacheKey, ok := reqBody["prompt_cache_key"].(string)
+	if !ok || strings.TrimSpace(promptCacheKey) == "" || !shouldRewriteCodexFingerprintPromptCacheKey(ids, promptCacheKey) {
+		return false
+	}
+	if promptCacheKey == ids.sessionID {
+		return false
+	}
+	reqBody["prompt_cache_key"] = ids.sessionID
+	return true
+}
+
 // applyCodexFingerprintClientMetadataRaw 在原始 JSON 字节上改写 client_metadata，
 // 供透传路径使用——透传是热路径，禁止对可能高达数十 MB 的 body 做全量
 // Unmarshal（见 forwardOpenAIPassthrough 的轻量提取注释）。实现为：gjson 提取
 // client_metadata 小对象单独解码，经共享核心改写后 sjson 一次性拼回，body
-// 其余字节原样保留。语义与 applyCodexFingerprintClientMetadata 逐点一致
-// （含"非对象值整体替换为收敛集合"的行为）。
+// 其余字节原样保留；root prompt_cache_key 仅在可证明是 body session 默认值时
+// 做标量改写。语义与 applyCodexFingerprintClientMetadata 逐点一致（含
+// "非对象值整体替换为收敛集合"的行为）。
 func applyCodexFingerprintClientMetadataRaw(body []byte, ids *codexFingerprintIDs) ([]byte, bool, error) {
 	if len(body) == 0 || ids == nil {
 		return body, false, nil
 	}
 	// 非 JSON 对象的 body（数组/标量/畸形）没有 client_metadata 语义，
 	// sjson 在这类根上写字段会改写整体结构，直接放行保持原样。
-	if !gjson.ParseBytes(body).IsObject() {
+	root := gjson.ParseBytes(body)
+	if !root.IsObject() {
+		captureCodexFingerprintOriginalBodySessionIDRaw(ids, gjson.Result{})
 		return body, false, nil
 	}
 
 	existing := map[string]any{}
 	if cm := gjson.GetBytes(body, "client_metadata"); cm.IsObject() {
+		captureCodexFingerprintOriginalBodySessionIDRaw(ids, gjson.GetBytes(body, "client_metadata.session_id"))
 		if err := json.Unmarshal([]byte(cm.Raw), &existing); err != nil {
 			return body, false, fmt.Errorf("decode client_metadata for fingerprint: %w", err)
 		}
+	} else {
+		captureCodexFingerprintOriginalBodySessionIDRaw(ids, gjson.Result{})
 	}
 
-	if !applyCodexFingerprintToClientMetadataMap(existing, ids) {
-		return body, false, nil
-	}
-
-	raw, err := json.Marshal(existing)
-	if err != nil {
-		return body, false, fmt.Errorf("encode converged client_metadata: %w", err)
+	next := body
+	modified := false
+	if applyCodexFingerprintToClientMetadataMap(existing, ids) {
+		raw, err := json.Marshal(existing)
+		if err != nil {
+			return body, false, fmt.Errorf("encode converged client_metadata: %w", err)
+		}
+		var setErr error
+		next, setErr = sjson.SetRawBytes(body, "client_metadata", raw)
+		if setErr != nil {
+			return body, false, fmt.Errorf("splice converged client_metadata: %w", setErr)
+		}
+		modified = true
 	}
-	next, err := sjson.SetRawBytes(body, "client_metadata", raw)
-	if err != nil {
-		return body, false, fmt.Errorf("splice converged client_metadata: %w", err)
+	promptCacheKey := gjson.GetBytes(body, "prompt_cache_key")
+	if promptCacheKey.Exists() && promptCacheKey.Type == gjson.String && strings.TrimSpace(promptCacheKey.String()) != "" && shouldRewriteCodexFingerprintPromptCacheKey(ids, promptCacheKey.String()) {
+		rewritten, err := sjson.SetBytes(next, "prompt_cache_key", ids.sessionID)
+		if err != nil {
+			return body, false, fmt.Errorf("splice converged prompt_cache_key: %w", err)
+		}
+		next = rewritten
+		modified = true
 	}
-	return next, true, nil
+	return next, modified, nil
 }
 
 // rewriteClientMetadataEmbeddedTurnMetadata 改写 client_metadata 中内嵌的
-// x-codex-turn-metadata JSON 字符串里的指定字段。
+// x-codex-turn-metadata JSON 字符串里的指定字段。非法/非对象值会重建，
+// 避免 flat client_metadata 与 embedded metadata 暴露两套身份。
 func rewriteClientMetadataEmbeddedTurnMetadata(clientMetadata map[string]any, fields map[string]any) {
 	raw, ok := clientMetadata["x-codex-turn-metadata"].(string)
 	if !ok || raw == "" {
 		return
 	}
 	var metadata map[string]any
-	if err := json.Unmarshal([]byte(raw), &metadata); err != nil {
-		return
+	if err := json.Unmarshal([]byte(raw), &metadata); err != nil || metadata == nil {
+		metadata = make(map[string]any, len(fields))
 	}
 	for k, v := range fields {
 		metadata[k] = v
diff --git a/backend/internal/service/openai_codex_fingerprint_test.go b/backend/internal/service/openai_codex_fingerprint_test.go
index 8e74bfa8bf12..841ee3437d49 100644
--- a/backend/internal/service/openai_codex_fingerprint_test.go
+++ b/backend/internal/service/openai_codex_fingerprint_test.go
@@ -13,7 +13,17 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+const testCodexFingerprintSeed = "11111111-1111-4111-8111-111111111111"
+
 func newTestOAuthAccount(id int64, extra map[string]any) *Account {
+	if codexFingerprintModeRequiresSeed(codexFingerprintModeFromExtra(extra)) {
+		if extra == nil {
+			extra = make(map[string]any)
+		}
+		if _, exists := extra[codexFingerprintSeedExtraKey]; !exists {
+			extra[codexFingerprintSeedExtraKey] = testCodexFingerprintSeed
+		}
+	}
 	return &Account{
 		ID:       id,
 		Platform: PlatformOpenAI,
@@ -75,42 +85,40 @@ func TestGetCodexFingerprintMode(t *testing.T) {
 
 func TestResolveConvergedInstallationID_UsesDeviceID(t *testing.T) {
 	account := newTestOAuthAccount(1, map[string]any{"openai_device_id": "real-device-id"})
-	assert.Equal(t, "real-device-id", resolveConvergedInstallationID(account))
+	assert.Equal(t, "real-device-id", resolveConvergedInstallationID(account, testCodexFingerprintSeed))
 }
 
-func TestResolveConvergedInstallationID_DerivesFromAccountID(t *testing.T) {
+func TestResolveConvergedInstallationID_DerivesFromSeed(t *testing.T) {
 	account := newTestOAuthAccount(42, nil)
-	result := resolveConvergedInstallationID(account)
+	result := resolveConvergedInstallationID(account, testCodexFingerprintSeed)
 	_, err := uuid.Parse(result)
 	require.NoError(t, err, "派生值应为合法 UUID")
-	assert.Equal(t, result, resolveConvergedInstallationID(account), "确定性")
+	assert.Equal(t, result, resolveConvergedInstallationID(account, testCodexFingerprintSeed), "确定性")
 }
 
-func TestResolveConvergedInstallationID_DifferentAccounts(t *testing.T) {
-	a := resolveConvergedInstallationID(newTestOAuthAccount(1, nil))
-	b := resolveConvergedInstallationID(newTestOAuthAccount(2, nil))
+func TestResolveConvergedInstallationID_DifferentSeeds(t *testing.T) {
+	account := newTestOAuthAccount(1, nil)
+	a := resolveConvergedInstallationID(account, testCodexFingerprintSeed)
+	b := resolveConvergedInstallationID(account, "22222222-2222-4222-8222-222222222222")
 	assert.NotEqual(t, a, b)
 }
 
 // --- resolveConvergedThreadID ---
 
 func TestResolveConvergedThreadID_PerClientSession(t *testing.T) {
-	account := newTestOAuthAccount(1, nil)
-	a := resolveConvergedThreadID(account, "session-aaa")
-	b := resolveConvergedThreadID(account, "session-bbb")
+	a := resolveConvergedThreadID(testCodexFingerprintSeed, "session-aaa")
+	b := resolveConvergedThreadID(testCodexFingerprintSeed, "session-bbb")
 	assert.NotEqual(t, a, b, "不同客户端 session 应得到不同 thread_id")
 }
 
 func TestResolveConvergedThreadID_Deterministic(t *testing.T) {
-	account := newTestOAuthAccount(1, nil)
-	a := resolveConvergedThreadID(account, "session-aaa")
-	b := resolveConvergedThreadID(account, "session-aaa")
+	a := resolveConvergedThreadID(testCodexFingerprintSeed, "session-aaa")
+	b := resolveConvergedThreadID(testCodexFingerprintSeed, "session-aaa")
 	assert.Equal(t, a, b, "同一客户端 session 应得到相同 thread_id")
 }
 
 func TestResolveConvergedThreadID_EmptySession(t *testing.T) {
-	account := newTestOAuthAccount(1, nil)
-	assert.Equal(t, "", resolveConvergedThreadID(account, ""))
+	assert.Equal(t, "", resolveConvergedThreadID(testCodexFingerprintSeed, ""))
 }
 
 // --- off 模式：resolveCodexFingerprintIDsFromRequest 返回 nil ---
@@ -141,6 +149,25 @@ func TestResolveCodexFingerprintIDsFromRequest_ExplicitOptInHonored(t *testing.T
 	}
 }
 
+func TestResolveCodexFingerprintIDsFromRequest_EnabledModesRequireValidSeed(t *testing.T) {
+	for _, tt := range []struct {
+		name  string
+		extra map[string]any
+	}{
+		{name: "missing", extra: map[string]any{codexFingerprintModeExtraKey: "device"}},
+		{name: "missing with device override", extra: map[string]any{codexFingerprintModeExtraKey: "device", "openai_device_id": "real-device"}},
+		{name: "blank", extra: map[string]any{codexFingerprintModeExtraKey: "session", codexFingerprintSeedExtraKey: ""}},
+		{name: "uppercase", extra: map[string]any{codexFingerprintModeExtraKey: "full", codexFingerprintSeedExtraKey: "11111111-1111-4111-8111-AAAAAAAAAAAA"}},
+		{name: "nil uuid", extra: map[string]any{codexFingerprintModeExtraKey: "device", codexFingerprintSeedExtraKey: "00000000-0000-0000-0000-000000000000"}},
+		{name: "non string", extra: map[string]any{codexFingerprintModeExtraKey: "session", codexFingerprintSeedExtraKey: 123}},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			account := &Account{ID: 1, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Extra: tt.extra}
+			require.Nil(t, resolveCodexFingerprintIDsFromRequest(account, nil))
+		})
+	}
+}
+
 // --- applyCodexFingerprintHeaders: off 模式 ---
 
 func TestApplyCodexFingerprintHeaders_OffMode(t *testing.T) {
@@ -199,9 +226,11 @@ func TestApplyCodexFingerprintHeaders_SessionMode(t *testing.T) {
 	ids := resolveCodexFingerprintIDsFromRequest(account, clientHeaders)
 	applyCodexFingerprintHeaders(h, ids)
 
-	convergedInstall := resolveConvergedInstallationID(account)
-	convergedSession := resolveConvergedSessionID(account)
-	convergedThread := resolveConvergedThreadID(account, "client-session-aaa")
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	convergedInstall := resolveConvergedInstallationID(account, seed)
+	convergedSession := resolveConvergedSessionID(seed)
+	convergedThread := resolveConvergedThreadID(seed, "client-session-aaa")
 
 	assert.Equal(t, convergedInstall, h.Get("x-codex-installation-id"))
 	assert.Equal(t, convergedSession, h.Get("session-id"))
@@ -257,7 +286,9 @@ func TestApplyCodexFingerprintHeaders_FullMode(t *testing.T) {
 	account := newTestOAuthAccount(1, map[string]any{
 		codexFingerprintModeExtraKey: "full",
 	})
-	convergedSession := resolveConvergedSessionID(account)
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	convergedSession := resolveConvergedSessionID(seed)
 
 	clientA := http.Header{}
 	clientA.Set("session-id", "client-A")
@@ -329,6 +360,41 @@ func TestFingerprintIDs_HeaderAndBody_TurnID_Consistent(t *testing.T) {
 	assert.Equal(t, headerTurnID, bodyTurnID, "头和体的 turn_id 必须一致")
 	assert.Equal(t, headerTurnID, bodyEmbeddedTurnID, "头和体内嵌 turn-metadata 的 turn_id 必须一致")
 	assert.Equal(t, ids.turnID, headerTurnID, "所有 turn_id 都应来自同一份 ids")
+	assert.Equal(t, headerMeta["turn_started_at_unix_ms"], bodyMeta["turn_started_at_unix_ms"], "头和体的 timestamp 必须一致")
+	assert.Equal(t, float64(ids.turnStartedAtUnixMs), headerMeta["turn_started_at_unix_ms"])
+}
+
+func TestFingerprintIDs_MalformedEmbeddedMetadataRebuiltConsistently(t *testing.T) {
+	account := newTestOAuthAccount(2, map[string]any{codexFingerprintModeExtraKey: "session"})
+	clientHeaders := make(http.Header)
+	clientHeaders.Set("session-id", "client-session-malformed")
+	ids := resolveCodexFingerprintIDsFromRequest(account, clientHeaders)
+	require.NotNil(t, ids)
+
+	h := make(http.Header)
+	h.Set("x-codex-turn-metadata", "{malformed")
+	applyCodexFingerprintHeaders(h, ids)
+
+	reqBody := map[string]any{
+		"client_metadata": map[string]any{
+			"session_id":            "client-session-malformed",
+			"x-codex-turn-metadata": "[malformed",
+		},
+	}
+	require.True(t, applyCodexFingerprintClientMetadata(reqBody, ids))
+
+	var headerMeta map[string]any
+	require.NoError(t, json.Unmarshal([]byte(h.Get("x-codex-turn-metadata")), &headerMeta))
+	clientMetadata, ok := reqBody["client_metadata"].(map[string]any)
+	require.True(t, ok)
+	bodyRaw, ok := clientMetadata["x-codex-turn-metadata"].(string)
+	require.True(t, ok)
+	var bodyMeta map[string]any
+	require.NoError(t, json.Unmarshal([]byte(bodyRaw), &bodyMeta))
+
+	for _, key := range []string{"installation_id", "session_id", "thread_id", "turn_id", "window_id", "turn_started_at_unix_ms"} {
+		assert.Equal(t, headerMeta[key], bodyMeta[key], "rebuilt metadata field %s must match", key)
+	}
 }
 
 // --- applyCodexFingerprintClientMetadata ---
@@ -400,9 +466,11 @@ func TestApplyCodexFingerprintClientMetadata_SessionMode(t *testing.T) {
 
 	cm, ok := reqBody["client_metadata"].(map[string]any)
 	require.True(t, ok)
-	convergedInstall := resolveConvergedInstallationID(account)
-	convergedSession := resolveConvergedSessionID(account)
-	convergedThread := resolveConvergedThreadID(account, "client-session-aaa")
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	convergedInstall := resolveConvergedInstallationID(account, seed)
+	convergedSession := resolveConvergedSessionID(seed)
+	convergedThread := resolveConvergedThreadID(seed, "client-session-aaa")
 
 	assert.Equal(t, convergedInstall, cm["x-codex-installation-id"])
 	assert.Equal(t, convergedSession, cm["session_id"])
@@ -441,7 +509,9 @@ func TestApplyCodexFingerprintClientMetadata_FullMode(t *testing.T) {
 
 	cm, ok := reqBody["client_metadata"].(map[string]any)
 	require.True(t, ok)
-	convergedSession := resolveConvergedSessionID(account)
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	convergedSession := resolveConvergedSessionID(seed)
 
 	assert.Equal(t, convergedSession, cm["session_id"])
 	assert.Equal(t, convergedSession, cm["thread_id"], "full 模式 thread_id 应等于 session_id")
@@ -496,6 +566,182 @@ func rawVsMapClientMetadata(t *testing.T, body []byte, ids *codexFingerprintIDs)
 	return mapCM, rawCM
 }
 
+func cloneCodexFingerprintIDsForTest(ids *codexFingerprintIDs) *codexFingerprintIDs {
+	if ids == nil {
+		return nil
+	}
+	cloned := *ids
+	cloned.originalBodySessionID = ""
+	cloned.originalBodySessionIDCaptured = false
+	return &cloned
+}
+
+func applyMapAndRawFingerprintBodiesForTest(t *testing.T, body []byte, ids *codexFingerprintIDs) (map[string]any, map[string]any) {
+	t.Helper()
+
+	mapIDs := cloneCodexFingerprintIDsForTest(ids)
+	rawIDs := cloneCodexFingerprintIDsForTest(ids)
+
+	var decoded map[string]any
+	require.NoError(t, json.Unmarshal(body, &decoded))
+	applyCodexFingerprintClientMetadata(decoded, mapIDs)
+
+	rawBody, _, err := applyCodexFingerprintClientMetadataRaw(body, rawIDs)
+	require.NoError(t, err)
+	var rawDecoded map[string]any
+	require.NoError(t, json.Unmarshal(rawBody, &rawDecoded))
+	return decoded, rawDecoded
+}
+
+func TestApplyCodexFingerprintPromptCacheKey_MapRawEquivalence(t *testing.T) {
+	for _, mode := range []codexFingerprintMode{codexFingerprintSession, codexFingerprintFull} {
+		t.Run(string(mode)+"/default", func(t *testing.T) {
+			account := newTestOAuthAccount(4300, map[string]any{codexFingerprintModeExtraKey: string(mode)})
+			ids := resolveCodexFingerprintIDs(account, "header-session", mode)
+			require.NotNil(t, ids)
+
+			body := []byte(`{"model":"gpt-5.6-sol","prompt_cache_key":"body-session","client_metadata":{"session_id":" body-session ","trace":"keep"},"input":[]}`)
+			mapBody, rawBody := applyMapAndRawFingerprintBodiesForTest(t, body, ids)
+
+			require.Equal(t, mapBody["prompt_cache_key"], rawBody["prompt_cache_key"])
+			require.Equal(t, ids.sessionID, mapBody["prompt_cache_key"])
+			mapCM, _ := mapBody["client_metadata"].(map[string]any)
+			rawCM, _ := rawBody["client_metadata"].(map[string]any)
+			require.Equal(t, ids.sessionID, mapCM["session_id"])
+			require.Equal(t, mapCM["session_id"], rawCM["session_id"])
+			require.Equal(t, "keep", rawCM["trace"])
+		})
+	}
+
+	t.Run("explicit override", func(t *testing.T) {
+		account := newTestOAuthAccount(4301, map[string]any{codexFingerprintModeExtraKey: "session"})
+		ids := resolveCodexFingerprintIDs(account, "header-session", codexFingerprintSession)
+		require.NotNil(t, ids)
+
+		body := []byte(`{"model":"gpt-5.6-sol","prompt_cache_key":"explicit-cache","client_metadata":{"session_id":"body-session"},"input":[]}`)
+		mapBody, rawBody := applyMapAndRawFingerprintBodiesForTest(t, body, ids)
+
+		require.Equal(t, "explicit-cache", mapBody["prompt_cache_key"])
+		require.Equal(t, "explicit-cache", rawBody["prompt_cache_key"])
+		mapCM, _ := mapBody["client_metadata"].(map[string]any)
+		rawCM, _ := rawBody["client_metadata"].(map[string]any)
+		require.Equal(t, ids.sessionID, mapCM["session_id"])
+		require.Equal(t, ids.sessionID, rawCM["session_id"])
+	})
+}
+
+func TestApplyCodexFingerprintPromptCacheKey_Negatives(t *testing.T) {
+	sessionAccount := newTestOAuthAccount(4310, map[string]any{codexFingerprintModeExtraKey: "session"})
+	sessionIDs := resolveCodexFingerprintIDs(sessionAccount, "header-session", codexFingerprintSession)
+	require.NotNil(t, sessionIDs)
+	deviceAccount := newTestOAuthAccount(4311, map[string]any{codexFingerprintModeExtraKey: "device"})
+	deviceIDs := resolveCodexFingerprintIDs(deviceAccount, "header-session", codexFingerprintDevice)
+	require.NotNil(t, deviceIDs)
+
+	tests := []struct {
+		name          string
+		body          []byte
+		ids           *codexFingerprintIDs
+		wantExists    bool
+		wantCacheKey  any
+		wantRawString string
+	}{
+		{
+			name:       "missing key is not injected",
+			body:       []byte(`{"client_metadata":{"session_id":"body-session"}}`),
+			ids:        sessionIDs,
+			wantExists: false,
+		},
+		{
+			name:         "empty key preserved",
+			body:         []byte(`{"prompt_cache_key":"","client_metadata":{"session_id":"body-session"}}`),
+			ids:          sessionIDs,
+			wantExists:   true,
+			wantCacheKey: "",
+		},
+		{
+			name:         "whitespace-different key is an explicit override",
+			body:         []byte(`{"prompt_cache_key":" body-session ","client_metadata":{"session_id":"body-session"}}`),
+			ids:          sessionIDs,
+			wantExists:   true,
+			wantCacheKey: " body-session ",
+		},
+		{
+			name:         "non-string key preserved",
+			body:         []byte(`{"prompt_cache_key":123,"client_metadata":{"session_id":"body-session"}}`),
+			ids:          sessionIDs,
+			wantExists:   true,
+			wantCacheKey: float64(123),
+		},
+		{
+			name:         "missing source metadata preserves key",
+			body:         []byte(`{"prompt_cache_key":"body-session"}`),
+			ids:          sessionIDs,
+			wantExists:   true,
+			wantCacheKey: "body-session",
+		},
+		{
+			name:         "non-string source session preserves key",
+			body:         []byte(`{"prompt_cache_key":"123","client_metadata":{"session_id":123}}`),
+			ids:          sessionIDs,
+			wantExists:   true,
+			wantCacheKey: "123",
+		},
+		{
+			name:         "non-object source metadata preserves key",
+			body:         []byte(`{"prompt_cache_key":"body-session","client_metadata":"bad"}`),
+			ids:          sessionIDs,
+			wantExists:   true,
+			wantCacheKey: "body-session",
+		},
+		{
+			name:         "device mode preserves key",
+			body:         []byte(`{"prompt_cache_key":"body-session","client_metadata":{"session_id":"body-session"}}`),
+			ids:          deviceIDs,
+			wantExists:   true,
+			wantCacheKey: "body-session",
+		},
+		{
+			name:          "off mode preserves body",
+			body:          []byte(`{"prompt_cache_key":"body-session","client_metadata":{"session_id":"body-session"}}`),
+			ids:           nil,
+			wantExists:    true,
+			wantCacheKey:  "body-session",
+			wantRawString: `{"prompt_cache_key":"body-session","client_metadata":{"session_id":"body-session"}}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var mapBody map[string]any
+			require.NoError(t, json.Unmarshal(tt.body, &mapBody))
+			changedMap := applyCodexFingerprintClientMetadata(mapBody, cloneCodexFingerprintIDsForTest(tt.ids))
+
+			rawBody, changedRaw, err := applyCodexFingerprintClientMetadataRaw(tt.body, cloneCodexFingerprintIDsForTest(tt.ids))
+			require.NoError(t, err)
+			if tt.ids == nil {
+				require.False(t, changedMap)
+				require.False(t, changedRaw)
+				require.JSONEq(t, tt.wantRawString, string(rawBody))
+				return
+			}
+			require.True(t, changedMap)
+			require.True(t, changedRaw)
+
+			rawDecoded := map[string]any{}
+			require.NoError(t, json.Unmarshal(rawBody, &rawDecoded))
+			_, mapExists := mapBody["prompt_cache_key"]
+			_, rawExists := rawDecoded["prompt_cache_key"]
+			require.Equal(t, tt.wantExists, mapExists)
+			require.Equal(t, tt.wantExists, rawExists)
+			if tt.wantExists {
+				require.Equal(t, tt.wantCacheKey, mapBody["prompt_cache_key"])
+				require.Equal(t, tt.wantCacheKey, rawDecoded["prompt_cache_key"])
+			}
+		})
+	}
+}
+
 func TestApplyCodexFingerprintClientMetadataRaw_MatchesMapVariant(t *testing.T) {
 	embedded := `{\"installation_id\":\"real-install\",\"session_id\":\"real-session\",\"sandbox\":\"seatbelt\"}`
 	bodies := map[string]string{
@@ -504,7 +750,7 @@ func TestApplyCodexFingerprintClientMetadataRaw_MatchesMapVariant(t *testing.T)
 		"non_object_value":   `{"model":"gpt-5.6-sol","client_metadata":"bogus","stream":true}`,
 	}
 	for _, mode := range []codexFingerprintMode{codexFingerprintDevice, codexFingerprintSession, codexFingerprintFull} {
-		account := newTestOAuthAccount(4242, nil)
+		account := newTestOAuthAccount(4242, map[string]any{codexFingerprintModeExtraKey: string(mode)})
 		ids := resolveCodexFingerprintIDs(account, "client-sess-raw", mode)
 		require.NotNil(t, ids)
 		for name, body := range bodies {
@@ -517,7 +763,7 @@ func TestApplyCodexFingerprintClientMetadataRaw_MatchesMapVariant(t *testing.T)
 }
 
 func TestApplyCodexFingerprintClientMetadataRaw_PreservesUnrelatedFields(t *testing.T) {
-	account := newTestOAuthAccount(4243, nil)
+	account := newTestOAuthAccount(4243, map[string]any{codexFingerprintModeExtraKey: "session"})
 	ids := resolveCodexFingerprintIDs(account, "client-sess-preserve", codexFingerprintSession)
 	require.NotNil(t, ids)
 
@@ -562,7 +808,7 @@ func newFingerprintStageTestContext(t *testing.T) *gin.Context {
 
 func TestStageCodexFingerprintIDs_NilOverwritesPreviousAccount(t *testing.T) {
 	c := newFingerprintStageTestContext(t)
-	accountA := newTestOAuthAccount(1001, nil)
+	accountA := newTestOAuthAccount(1001, map[string]any{codexFingerprintModeExtraKey: "session"})
 	idsA := resolveCodexFingerprintIDs(accountA, "sess-x", codexFingerprintSession)
 	require.NotNil(t, idsA)
 	stageCodexFingerprintIDs(c, idsA)
@@ -578,9 +824,30 @@ func TestStageCodexFingerprintIDs_NilOverwritesPreviousAccount(t *testing.T) {
 	assert.Empty(t, h.Get("x-codex-installation-id"))
 }
 
+func TestApplyStagedCodexFingerprintRejectsDifferentOAuthAccount(t *testing.T) {
+	c := newFingerprintStageTestContext(t)
+	accountA := newTestOAuthAccount(1003, map[string]any{codexFingerprintModeExtraKey: "session"})
+	idsA := resolveCodexFingerprintIDs(accountA, "sess-a", codexFingerprintSession)
+	require.NotNil(t, idsA)
+	stageCodexFingerprintIDs(c, idsA)
+
+	accountB := newTestOAuthAccount(1004, map[string]any{codexFingerprintModeExtraKey: "session"})
+	h := make(http.Header)
+	h.Set("session-id", "account-b-session")
+	applyStagedCodexFingerprintHeaders(c, accountB, h)
+	assert.Equal(t, "account-b-session", h.Get("session-id"))
+	assert.Empty(t, h.Get("x-codex-installation-id"))
+
+	body := map[string]any{"client_metadata": map[string]any{"session_id": "account-b-session"}}
+	assert.False(t, applyStagedCodexFingerprintClientMetadata(c, accountB, body))
+	clientMetadata, ok := body["client_metadata"].(map[string]any)
+	require.True(t, ok)
+	assert.Equal(t, "account-b-session", clientMetadata["session_id"])
+}
+
 func TestApplyStagedCodexFingerprintHeaders_SkipsNonOAuthAccount(t *testing.T) {
 	c := newFingerprintStageTestContext(t)
-	oauthIDs := resolveCodexFingerprintIDs(newTestOAuthAccount(1003, nil), "sess-y", codexFingerprintSession)
+	oauthIDs := resolveCodexFingerprintIDs(newTestOAuthAccount(1003, map[string]any{codexFingerprintModeExtraKey: "session"}), "sess-y", codexFingerprintSession)
 	require.NotNil(t, oauthIDs)
 	stageCodexFingerprintIDs(c, oauthIDs)
 
@@ -643,12 +910,12 @@ func TestBuildUpstreamRequestOpenAIPassthrough_OffModeKeepsIsolatedSession(t *te
 	require.NoError(t, err)
 
 	assert.NotEmpty(t, req.Header.Get("session_id"))
-	assert.NotEqual(t, resolveConvergedSessionID(account), req.Header.Get("session_id"), "off 模式不得收敛 session_id")
+	assert.NotEqual(t, resolveConvergedSessionID(testCodexFingerprintSeed), req.Header.Get("session_id"), "off 模式不得收敛 session_id")
 	assert.Empty(t, req.Header.Get("x-codex-window-id"))
 }
 
 func TestApplyCodexFingerprintClientMetadataRaw_NonObjectBodyUntouched(t *testing.T) {
-	account := newTestOAuthAccount(4244, nil)
+	account := newTestOAuthAccount(4244, map[string]any{codexFingerprintModeExtraKey: "session"})
 	ids := resolveCodexFingerprintIDs(account, "client-sess-nonobj", codexFingerprintSession)
 	require.NotNil(t, ids)
 
diff --git a/backend/internal/service/openai_gateway_forward.go b/backend/internal/service/openai_gateway_forward.go
index a262fb60ccb3..6dab7fb702c3 100644
--- a/backend/internal/service/openai_gateway_forward.go
+++ b/backend/internal/service/openai_gateway_forward.go
@@ -421,6 +421,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 		if !isCompactRequest && applyCodexClientMetadata(decoded, account) {
 			markDecodedModified()
 		}
+		stageCodexFingerprintIDs(c, nil)
 		// 指纹收敛：一次性解析收敛 ID，请求体和出站头共享同一份 IDs（保证 turn_id 等随机字段一致）。
 		// fingerprintIDs 在此处解析，后续 buildUpstreamRequest 中使用同一份。
 		if !isCompactRequest {
@@ -442,7 +443,9 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 		if codexResult.NormalizedModel != "" {
 			upstreamModel = codexResult.NormalizedModel
 		}
-		if codexResult.PromptCacheKey != "" {
+		if currentPromptCacheKey, ok := decoded["prompt_cache_key"].(string); ok && currentPromptCacheKey != "" {
+			promptCacheKey = currentPromptCacheKey
+		} else if codexResult.PromptCacheKey != "" {
 			promptCacheKey = codexResult.PromptCacheKey
 		}
 	}
diff --git a/backend/internal/service/openai_gateway_passthrough.go b/backend/internal/service/openai_gateway_passthrough.go
index 291f8bcc9e93..4bfd2fa26c8b 100644
--- a/backend/internal/service/openai_gateway_passthrough.go
+++ b/backend/internal/service/openai_gateway_passthrough.go
@@ -81,6 +81,7 @@ func (s *OpenAIGatewayService) forwardOpenAIPassthrough(
 		}
 		reqStream = gjson.GetBytes(body, "stream").Bool()
 
+		stageCodexFingerprintIDs(c, nil)
 		// 指纹收敛：与非透传路径同门控（仅 OAuth、legacy compact 形态跳过）。
 		// 一次性解析收敛 ID：请求体 client_metadata 在此改写（raw 字节外科
 		// 手术，透传热路径禁全量 Unmarshal），出站头改写由请求构造器读取
diff --git a/backend/internal/service/openai_oauth_passthrough_test.go b/backend/internal/service/openai_oauth_passthrough_test.go
index d2c030c5c517..c7dc77b3aa31 100644
--- a/backend/internal/service/openai_oauth_passthrough_test.go
+++ b/backend/internal/service/openai_oauth_passthrough_test.go
@@ -1948,6 +1948,212 @@ func TestOpenAIGatewayService_OAuthPassthrough_CodexTuiIdentityUnified(t *testin
 	require.Equal(t, codexCLIVersion, upstream.lastReq.Header.Get("version"))
 }
 
+func TestOpenAIGatewayService_CodexFingerprintHTTPTransformedHeaderBodyParityAndDefaultCacheKey(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", bytes.NewReader(nil))
+	c.Request.Header.Set("User-Agent", "codex_cli_rs/0.144.1")
+	c.Request.Header.Set("originator", "codex_cli_rs")
+	c.Request.Header.Set("session-id", "header-session")
+	c.Request.Header.Set("x-codex-turn-metadata", `{"installation_id":"header-install","session_id":"header-session","thread_id":"header-thread","turn_id":"header-turn","window_id":"header-window","sandbox":"seatbelt"}`)
+
+	body := []byte(`{"model":"gpt-5.2","stream":false,"prompt_cache_key":"body-session","client_metadata":{"session_id":"body-session","x-codex-turn-metadata":"{\"installation_id\":\"body-install\",\"session_id\":\"body-session\",\"thread_id\":\"body-thread\",\"turn_id\":\"body-turn\",\"window_id\":\"body-window\",\"sandbox\":\"seatbelt\"}"},"input":[{"type":"message","role":"user","content":"hi"}]}`)
+
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid"}},
+		Body:       io.NopCloser(strings.NewReader("data: [DONE]\n\n")),
+	}}
+	svc := &OpenAIGatewayService{
+		cfg:           &config.Config{},
+		httpUpstream:  upstream,
+		toolCorrector: NewCodexToolCorrector(),
+	}
+	account := newTestOAuthAccount(4401, map[string]any{codexFingerprintModeExtraKey: "session"})
+	account.Name = "oauth-transformed"
+	account.Status = StatusActive
+	account.Schedulable = true
+	account.Concurrency = 1
+	account.Credentials = map[string]any{"access_token": "oauth-token", "chatgpt_account_id": "chatgpt-acc"}
+
+	_, err := svc.Forward(context.Background(), c, account, body)
+	require.NoError(t, err)
+	require.NotNil(t, upstream.lastReq)
+
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	wantInstall := resolveConvergedInstallationID(account, seed)
+	wantSession := resolveConvergedSessionID(seed)
+	wantThread := resolveConvergedThreadID(seed, "header-session")
+
+	require.Equal(t, wantInstall, upstream.lastReq.Header.Get("x-codex-installation-id"))
+	require.Equal(t, wantSession, upstream.lastReq.Header.Get("session-id"))
+	require.Equal(t, wantSession, upstream.lastReq.Header.Get("session_id"))
+	require.Equal(t, wantThread, upstream.lastReq.Header.Get("thread-id"))
+	require.Equal(t, wantThread, upstream.lastReq.Header.Get("x-client-request-id"))
+	require.Equal(t, wantThread+":0", upstream.lastReq.Header.Get("x-codex-window-id"))
+
+	require.Equal(t, wantSession, gjson.GetBytes(upstream.lastBody, "prompt_cache_key").String())
+	require.Equal(t, wantInstall, gjson.GetBytes(upstream.lastBody, "client_metadata.x-codex-installation-id").String())
+	require.Equal(t, wantSession, gjson.GetBytes(upstream.lastBody, "client_metadata.session_id").String())
+	require.Equal(t, wantThread, gjson.GetBytes(upstream.lastBody, "client_metadata.thread_id").String())
+	require.Equal(t, wantThread+":0", gjson.GetBytes(upstream.lastBody, "client_metadata.x-codex-window-id").String())
+
+	bodyTurnMetadata := gjson.GetBytes(upstream.lastBody, "client_metadata.x-codex-turn-metadata").String()
+	headerTurnMetadata := upstream.lastReq.Header.Get("x-codex-turn-metadata")
+	require.Equal(t, wantSession, gjson.Get(bodyTurnMetadata, "session_id").String())
+	require.Equal(t, wantSession, gjson.Get(headerTurnMetadata, "session_id").String())
+	require.Equal(t, gjson.Get(bodyTurnMetadata, "turn_id").String(), gjson.Get(headerTurnMetadata, "turn_id").String())
+}
+
+func TestOpenAIGatewayService_CodexFingerprintHTTPRawPassthroughHeaderBodyParityAndDefaultCacheKey(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", bytes.NewReader(nil))
+	c.Request.Header.Set("User-Agent", "codex_cli_rs/0.144.1")
+	c.Request.Header.Set("originator", "codex_cli_rs")
+	c.Request.Header.Set("session-id", "header-session")
+	c.Request.Header.Set("x-codex-turn-metadata", `{"installation_id":"header-install","session_id":"header-session","thread_id":"header-thread","turn_id":"header-turn","window_id":"header-window","sandbox":"seatbelt"}`)
+
+	body := []byte(`{"model":"gpt-5.6-sol","stream":false,"prompt_cache_key":"body-session","client_metadata":{"session_id":"body-session","x-codex-turn-metadata":"{\"installation_id\":\"body-install\",\"session_id\":\"body-session\",\"thread_id\":\"body-thread\",\"turn_id\":\"body-turn\",\"window_id\":\"body-window\",\"sandbox\":\"seatbelt\"}"},"input":[{"type":"message","role":"user","content":"hi"}]}`)
+
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid"}},
+		Body:       io.NopCloser(strings.NewReader("data: [DONE]\n\n")),
+	}}
+	svc := &OpenAIGatewayService{
+		cfg:          &config.Config{},
+		httpUpstream: upstream,
+	}
+	account := newTestOAuthAccount(4402, map[string]any{
+		codexFingerprintModeExtraKey: "session",
+		"openai_oauth_passthrough":   true,
+	})
+	account.Name = "oauth-raw"
+	account.Status = StatusActive
+	account.Schedulable = true
+	account.Concurrency = 1
+	account.Credentials = map[string]any{"access_token": "oauth-token", "chatgpt_account_id": "chatgpt-acc"}
+
+	_, err := svc.Forward(context.Background(), c, account, body)
+	require.NoError(t, err)
+	require.NotNil(t, upstream.lastReq)
+
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	wantInstall := resolveConvergedInstallationID(account, seed)
+	wantSession := resolveConvergedSessionID(seed)
+	wantThread := resolveConvergedThreadID(seed, "header-session")
+
+	require.Equal(t, wantInstall, upstream.lastReq.Header.Get("x-codex-installation-id"))
+	require.Equal(t, wantSession, upstream.lastReq.Header.Get("session-id"))
+	require.Equal(t, wantSession, upstream.lastReq.Header.Get("session_id"))
+	require.Equal(t, wantThread, upstream.lastReq.Header.Get("thread-id"))
+	require.Equal(t, wantThread, upstream.lastReq.Header.Get("x-client-request-id"))
+	require.Equal(t, wantThread+":0", upstream.lastReq.Header.Get("x-codex-window-id"))
+
+	require.Equal(t, wantSession, gjson.GetBytes(upstream.lastBody, "prompt_cache_key").String())
+	require.Equal(t, wantInstall, gjson.GetBytes(upstream.lastBody, "client_metadata.x-codex-installation-id").String())
+	require.Equal(t, wantSession, gjson.GetBytes(upstream.lastBody, "client_metadata.session_id").String())
+	require.Equal(t, wantThread, gjson.GetBytes(upstream.lastBody, "client_metadata.thread_id").String())
+	require.Equal(t, wantThread+":0", gjson.GetBytes(upstream.lastBody, "client_metadata.x-codex-window-id").String())
+
+	bodyTurnMetadata := gjson.GetBytes(upstream.lastBody, "client_metadata.x-codex-turn-metadata").String()
+	headerTurnMetadata := upstream.lastReq.Header.Get("x-codex-turn-metadata")
+	require.Equal(t, wantSession, gjson.Get(bodyTurnMetadata, "session_id").String())
+	require.Equal(t, wantSession, gjson.Get(headerTurnMetadata, "session_id").String())
+	require.Equal(t, gjson.Get(bodyTurnMetadata, "turn_id").String(), gjson.Get(headerTurnMetadata, "turn_id").String())
+}
+
+func TestOpenAIGatewayService_CodexFingerprintCompactDoesNotRewriteBodyCacheKeyOrMetadata(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses/compact", bytes.NewReader(nil))
+	c.Request.Header.Set("User-Agent", "codex_cli_rs/0.144.1")
+	c.Request.Header.Set("originator", "codex_cli_rs")
+	c.Request.Header.Set("session-id", "header-session")
+
+	body := []byte(`{"model":"gpt-5.4","stream":false,"prompt_cache_key":"body-session","client_metadata":{"session_id":"body-session"},"input":[{"type":"message","role":"user","content":"compress"}]}`)
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid"}},
+		Body:       io.NopCloser(strings.NewReader(compactProbeSSESuccessBody)),
+	}}
+	svc := &OpenAIGatewayService{
+		cfg:           &config.Config{},
+		httpUpstream:  upstream,
+		toolCorrector: NewCodexToolCorrector(),
+	}
+	account := newTestOAuthAccount(4403, map[string]any{codexFingerprintModeExtraKey: "session"})
+	account.Name = "oauth-compact"
+	account.Status = StatusActive
+	account.Schedulable = true
+	account.Concurrency = 1
+	account.Credentials = map[string]any{"access_token": "oauth-token", "chatgpt_account_id": "chatgpt-acc"}
+	staleIDs := resolveCodexFingerprintIDs(account, "stale-session", codexFingerprintSession)
+	require.NotNil(t, staleIDs)
+	stageCodexFingerprintIDs(c, staleIDs)
+
+	_, err := svc.Forward(context.Background(), c, account, body)
+	require.NoError(t, err)
+	require.NotNil(t, upstream.lastReq)
+
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	require.NotEqual(t, resolveConvergedSessionID(seed), gjson.GetBytes(upstream.lastBody, "prompt_cache_key").String())
+	require.Equal(t, "body-session", gjson.GetBytes(upstream.lastBody, "prompt_cache_key").String())
+	require.Equal(t, "body-session", gjson.GetBytes(upstream.lastBody, "client_metadata.session_id").String())
+	require.False(t, gjson.GetBytes(upstream.lastBody, "client_metadata.x-codex-installation-id").Exists())
+	require.Empty(t, upstream.lastReq.Header.Get("x-codex-window-id"))
+}
+
+func TestOpenAIGatewayService_CodexFingerprintMessagesBridgeDoesNotInjectBodyPromptCacheKey(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", bytes.NewReader(nil))
+	c.Request.Header.Set("User-Agent", "codex_cli_rs/0.144.1")
+	c.Request.Header.Set("originator", "codex_cli_rs")
+	c.Request.Header.Set("session-id", "header-session")
+
+	body := []byte(`{"model":"gpt-5.5","stream":true,"prompt_cache_key":"anthropic-metadata-session-1","client_metadata":{"session_id":"anthropic-metadata-session-1"},"input":[{"type":"message","role":"developer","content":[{"type":"input_text","text":"` + openAICompatClaudeCodeTodoGuardMarker + `"}]},{"type":"message","role":"user","content":"hello"}]}`)
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid"}},
+		Body:       io.NopCloser(strings.NewReader("data: [DONE]\n\n")),
+	}}
+	svc := &OpenAIGatewayService{
+		cfg:           &config.Config{},
+		httpUpstream:  upstream,
+		toolCorrector: NewCodexToolCorrector(),
+	}
+	account := newTestOAuthAccount(4404, map[string]any{codexFingerprintModeExtraKey: "session"})
+	account.Name = "oauth-messages-bridge"
+	account.Status = StatusActive
+	account.Schedulable = true
+	account.Concurrency = 1
+	account.Credentials = map[string]any{"access_token": "oauth-token", "chatgpt_account_id": "chatgpt-acc"}
+
+	_, err := svc.Forward(context.Background(), c, account, body)
+	require.NoError(t, err)
+	require.NotNil(t, upstream.lastReq)
+
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	wantSession := resolveConvergedSessionID(seed)
+	require.False(t, gjson.GetBytes(upstream.lastBody, "prompt_cache_key").Exists())
+	require.Equal(t, wantSession, gjson.GetBytes(upstream.lastBody, "client_metadata.session_id").String())
+	require.Equal(t, wantSession, upstream.lastReq.Header.Get("session_id"))
+}
+
 func TestOpenAIGatewayService_CodexCLIOnly_RejectsNonCodexClient(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 
diff --git a/backend/internal/service/openai_ws_forwarder_payload.go b/backend/internal/service/openai_ws_forwarder_payload.go
index d783c9810420..797361a3a646 100644
--- a/backend/internal/service/openai_ws_forwarder_payload.go
+++ b/backend/internal/service/openai_ws_forwarder_payload.go
@@ -93,7 +93,13 @@ func (s *OpenAIGatewayService) buildOpenAIWSHeaders(
 				headers.Add("x-codex-beta-features", value)
 			}
 		}
-		for _, name := range [...]string{"x-codex-window-id", "x-codex-installation-id"} {
+		for _, name := range [...]string{
+			"x-codex-window-id",
+			"x-codex-installation-id",
+			"session-id",
+			"thread-id",
+			"x-client-request-id",
+		} {
 			if value := c.Request.Header.Get(name); strings.TrimSpace(value) != "" {
 				headers.Set(name, value)
 			}
@@ -128,6 +134,7 @@ func (s *OpenAIGatewayService) buildOpenAIWSHeaders(
 	if metadata := strings.TrimSpace(turnMetadata); metadata != "" {
 		headers.Set(openAIWSTurnMetadataHeader, metadata)
 	}
+	applyStagedCodexFingerprintHeaders(c, account, headers)
 
 	if account != nil && account.Type == AccountTypeOAuth {
 		if err := resolveAndSetOpenAIChatGPTAccountHeaders(ctx, s.accountRepo, headers, account); err != nil {
diff --git a/backend/internal/service/openai_ws_forwarder_success_test.go b/backend/internal/service/openai_ws_forwarder_success_test.go
index d8270dfe4009..2c6977638297 100644
--- a/backend/internal/service/openai_ws_forwarder_success_test.go
+++ b/backend/internal/service/openai_ws_forwarder_success_test.go
@@ -400,6 +400,9 @@ func TestOpenAIGatewayService_BuildOpenAIWSHeadersPreservesCodexIdentity(t *test
 	c.Request.Header.Set("User-Agent", "codex_cli_rs/0.144.1")
 	c.Request.Header.Set("X-Codex-Window-ID", "window-ws")
 	c.Request.Header.Set("X-Codex-Installation-ID", "installation-ws")
+	c.Request.Header.Set("session-id", "session-ws")
+	c.Request.Header.Set("thread-id", "thread-ws")
+	c.Request.Header.Set("x-client-request-id", "client-request-ws")
 	c.Request.Header.Set("X-Test", "blocked")
 
 	svc := &OpenAIGatewayService{}
@@ -421,9 +424,53 @@ func TestOpenAIGatewayService_BuildOpenAIWSHeadersPreservesCodexIdentity(t *test
 	require.NoError(t, err)
 	require.Equal(t, "window-ws", headers.Get("X-Codex-Window-ID"))
 	require.Equal(t, "installation-ws", headers.Get("X-Codex-Installation-ID"))
+	require.Equal(t, "session-ws", headers.Get("session-id"))
+	require.Equal(t, "thread-ws", headers.Get("thread-id"))
+	require.Equal(t, "client-request-ws", headers.Get("x-client-request-id"))
 	require.Empty(t, headers.Get("X-Test"))
 }
 
+func TestOpenAIGatewayService_BuildOpenAIWSHeadersDeviceModePreservesClientSessionIdentity(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodGet, "/v1/responses", nil)
+	c.Request.Header.Set("User-Agent", "codex_cli_rs/0.144.1")
+	c.Request.Header.Set("X-Codex-Installation-ID", "client-installation")
+	c.Request.Header.Set("X-Codex-Window-ID", "client-window")
+	c.Request.Header.Set("session-id", "client-session")
+	c.Request.Header.Set("thread-id", "client-thread")
+	c.Request.Header.Set("x-client-request-id", "client-request")
+
+	account := newTestOAuthAccount(1300, map[string]any{codexFingerprintModeExtraKey: "device"})
+	ids := resolveCodexFingerprintIDsFromRequest(account, c.Request.Header)
+	require.NotNil(t, ids)
+	stageCodexFingerprintIDs(c, ids)
+
+	svc := &OpenAIGatewayService{}
+	headers, _, err := svc.buildOpenAIWSHeaders(
+		context.Background(),
+		c,
+		account,
+		"token",
+		OpenAIWSProtocolDecision{Transport: OpenAIUpstreamTransportResponsesWebsocketV2},
+		true,
+		"",
+		"",
+		"",
+		"",
+		"",
+	)
+
+	require.NoError(t, err)
+	require.Equal(t, ids.installationID, headers.Get("x-codex-installation-id"))
+	require.NotEqual(t, "client-installation", headers.Get("x-codex-installation-id"))
+	require.Equal(t, "client-window", headers.Get("x-codex-window-id"))
+	require.Equal(t, "client-session", headers.Get("session-id"))
+	require.Equal(t, "client-thread", headers.Get("thread-id"))
+	require.Equal(t, "client-request", headers.Get("x-client-request-id"))
+}
+
 func TestLogOpenAIWSBindResponseAccountWarn(t *testing.T) {
 	require.NotPanics(t, func() {
 		logOpenAIWSBindResponseAccountWarn(1, 2, "resp_ok", nil)
@@ -983,6 +1030,96 @@ func TestOpenAIGatewayService_Forward_WSv2_HeaderSessionFallbackFromPromptCacheK
 	require.True(t, gjson.Get(requestToJSONString(captureConn.lastWrite), "stream").Exists())
 }
 
+func TestOpenAIGatewayService_Forward_WSv2_CodexFingerprintHandshakeBodyParityAndDefaultCacheKey(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/openai/v1/responses", nil)
+	c.Request.Header.Set("User-Agent", "codex_cli_rs/0.144.1")
+	c.Request.Header.Set("originator", "codex_cli_rs")
+	c.Request.Header.Set("session-id", "header-session")
+	c.Request.Header.Set("x-codex-turn-metadata", `{"installation_id":"header-install","session_id":"header-session","thread_id":"header-thread","turn_id":"header-turn","window_id":"header-window","sandbox":"seatbelt"}`)
+
+	cfg := &config.Config{}
+	cfg.Security.URLAllowlist.Enabled = false
+	cfg.Security.URLAllowlist.AllowInsecureHTTP = true
+	cfg.Gateway.OpenAIWS.Enabled = true
+	cfg.Gateway.OpenAIWS.OAuthEnabled = true
+	cfg.Gateway.OpenAIWS.APIKeyEnabled = true
+	cfg.Gateway.OpenAIWS.ResponsesWebsocketsV2 = true
+	cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 1
+	cfg.Gateway.OpenAIWS.MinIdlePerAccount = 0
+	cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 1
+
+	captureConn := &openAIWSCaptureConn{
+		events: [][]byte{
+			[]byte(`{"type":"response.completed","response":{"id":"resp_ws_fingerprint","model":"gpt-5.2","usage":{"input_tokens":2,"output_tokens":1}}}`),
+		},
+	}
+	captureDialer := &openAIWSCaptureDialer{conn: captureConn}
+	pool := newOpenAIWSConnPool(cfg)
+	pool.setClientDialerForTest(captureDialer)
+
+	svc := &OpenAIGatewayService{
+		cfg:              cfg,
+		httpUpstream:     &httpUpstreamRecorder{},
+		cache:            &stubGatewayCache{},
+		openaiWSResolver: NewOpenAIWSProtocolResolver(cfg),
+		toolCorrector:    NewCodexToolCorrector(),
+		openaiWSPool:     pool,
+	}
+	account := newTestOAuthAccount(4405, map[string]any{
+		codexFingerprintModeExtraKey:      "session",
+		"responses_websockets_v2_enabled": true,
+	})
+	account.Name = "oauth-ws-fingerprint"
+	account.Status = StatusActive
+	account.Schedulable = true
+	account.Concurrency = 1
+	account.Credentials = map[string]any{"access_token": "oauth-token", "chatgpt_account_id": "chatgpt-acc"}
+
+	body := []byte(`{"model":"gpt-5.2","stream":true,"prompt_cache_key":"body-session","client_metadata":{"session_id":"body-session","x-codex-turn-metadata":"{\"installation_id\":\"body-install\",\"session_id\":\"body-session\",\"thread_id\":\"body-thread\",\"turn_id\":\"body-turn\",\"window_id\":\"body-window\",\"sandbox\":\"seatbelt\"}"},"input":[{"type":"input_text","text":"hi"}]}`)
+	result, err := svc.Forward(context.Background(), c, account, body)
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.Equal(t, "resp_ws_fingerprint", result.RequestID)
+	require.NotNil(t, captureConn.lastWrite)
+
+	seed, ok := codexFingerprintSeed(account.Extra)
+	require.True(t, ok)
+	wantInstall := resolveConvergedInstallationID(account, seed)
+	wantSession := resolveConvergedSessionID(seed)
+	wantThread := resolveConvergedThreadID(seed, "header-session")
+	payloadJSON := requestToJSONString(captureConn.lastWrite)
+
+	require.Equal(t, wantInstall, captureDialer.lastHeaders.Get("x-codex-installation-id"))
+	require.Equal(t, wantSession, captureDialer.lastHeaders.Get("session-id"))
+	require.Equal(t, wantSession, captureDialer.lastHeaders.Get("session_id"))
+	require.Equal(t, wantThread, captureDialer.lastHeaders.Get("thread-id"))
+	require.Equal(t, wantThread, captureDialer.lastHeaders.Get("x-client-request-id"))
+	require.Equal(t, wantThread+":0", captureDialer.lastHeaders.Get("x-codex-window-id"))
+
+	require.Equal(t, wantSession, gjson.Get(payloadJSON, "prompt_cache_key").String())
+	require.Equal(t, wantInstall, gjson.Get(payloadJSON, "client_metadata.x-codex-installation-id").String())
+	require.Equal(t, wantSession, gjson.Get(payloadJSON, "client_metadata.session_id").String())
+	require.Equal(t, wantThread, gjson.Get(payloadJSON, "client_metadata.thread_id").String())
+	require.Equal(t, wantThread+":0", gjson.Get(payloadJSON, "client_metadata.x-codex-window-id").String())
+
+	bodyTurnMetadata := gjson.Get(payloadJSON, "client_metadata.x-codex-turn-metadata").String()
+	headerTurnMetadata := captureDialer.lastHeaders.Get("x-codex-turn-metadata")
+	require.Equal(t, wantInstall, gjson.Get(bodyTurnMetadata, "installation_id").String())
+	require.Equal(t, wantSession, gjson.Get(bodyTurnMetadata, "session_id").String())
+	require.Equal(t, wantThread, gjson.Get(bodyTurnMetadata, "thread_id").String())
+	require.Equal(t, wantSession, gjson.Get(headerTurnMetadata, "session_id").String())
+	require.Equal(t, gjson.Get(bodyTurnMetadata, "turn_id").String(), gjson.Get(headerTurnMetadata, "turn_id").String())
+	require.NotZero(t, gjson.Get(bodyTurnMetadata, "turn_started_at_unix_ms").Int())
+	require.Equal(t,
+		gjson.Get(bodyTurnMetadata, "turn_started_at_unix_ms").Int(),
+		gjson.Get(headerTurnMetadata, "turn_started_at_unix_ms").Int(),
+	)
+}
+
 func TestOpenAIGatewayService_Forward_WSv2_ResponseDoneUsageParsed(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 
diff --git a/backend/internal/service/openai_ws_forwarder_v2.go b/backend/internal/service/openai_ws_forwarder_v2.go
index e0dbd58bd4f5..020d409d0b9c 100644
--- a/backend/internal/service/openai_ws_forwarder_v2.go
+++ b/backend/internal/service/openai_ws_forwarder_v2.go
@@ -62,6 +62,14 @@ func (s *OpenAIGatewayService) forwardOpenAIWSV2(
 
 	payload := s.buildOpenAIWSCreatePayload(reqBody, account)
 	payloadStrategy, removedKeys := applyOpenAIWSRetryPayloadStrategy(payload, attempt)
+	turnState := ""
+	turnMetadata := ""
+	if c != nil && c.Request != nil {
+		turnState = strings.TrimSpace(c.GetHeader(openAIWSTurnStateHeader))
+		turnMetadata = strings.TrimSpace(c.GetHeader(openAIWSTurnMetadataHeader))
+	}
+	setOpenAIWSTurnMetadata(payload, turnMetadata)
+	applyStagedCodexFingerprintClientMetadata(c, account, payload)
 	previousResponseID := openAIWSPayloadString(payload, "previous_response_id")
 	previousResponseIDKind := ClassifyOpenAIPreviousResponseIDKind(previousResponseID)
 	promptCacheKey := openAIWSPayloadString(payload, "prompt_cache_key")
@@ -79,13 +87,6 @@ func (s *OpenAIGatewayService) forwardOpenAIWSV2(
 	if raw, ok := payload["stream"]; ok {
 		streamValue = normalizeOpenAIWSLogValue(strings.TrimSpace(fmt.Sprintf("%v", raw)))
 	}
-	turnState := ""
-	turnMetadata := ""
-	if c != nil && c.Request != nil {
-		turnState = strings.TrimSpace(c.GetHeader(openAIWSTurnStateHeader))
-		turnMetadata = strings.TrimSpace(c.GetHeader(openAIWSTurnMetadataHeader))
-	}
-	setOpenAIWSTurnMetadata(payload, turnMetadata)
 	payloadEventType := openAIWSPayloadString(payload, "type")
 	if payloadEventType == "" {
 		payloadEventType = "response.create"
diff --git a/backend/internal/service/openai_ws_pool.go b/backend/internal/service/openai_ws_pool.go
index 048031d64066..2c62b6b4c07a 100644
--- a/backend/internal/service/openai_ws_pool.go
+++ b/backend/internal/service/openai_ws_pool.go
@@ -77,7 +77,13 @@ type openAIWSAcquireRequest struct {
 }
 
 type openAIWSHandshakeCompatibilityKey struct {
-	betaFeatures string
+	betaFeatures        string
+	codexInstallationID string
+	sessionIDHyphen     string
+	sessionIDUnderscore string
+	threadID            string
+	clientRequestID     string
+	codexWindowID       string
 }
 
 type openAIWSConnLease struct {
@@ -855,7 +861,7 @@ func (p *openAIWSConnPool) acquire(ctx context.Context, req openAIWSAcquireReque
 
 retryAcquire:
 	accountID := req.Account.ID
-	compatibility := normalizeOpenAIWSHandshakeCompatibility(req.Headers)
+	compatibility := normalizeOpenAIWSHandshakeCompatibility(req.Account, req.Headers)
 	routingAffinity := normalizeOpenAIWSRoutingAffinity(req.Headers)
 	effectiveMaxConns := p.effectiveMaxConnsByAccount(req.Account)
 	if effectiveMaxConns <= 0 {
@@ -1015,37 +1021,39 @@ retryAcquire:
 			p.ensureTargetIdleAsync(accountID)
 			return lease, nil
 		}
-		for _, conn := range ap.conns {
-			if conn == nil || conn == best || !conn.matchesHandshakeCompatibility(compatibility) || !conn.matchesRoutingAffinity(routingAffinity) {
-				continue
-			}
-			if conn.tryAcquire() {
-				connPick := time.Since(pickStartedAt)
-				p.recordConnPickDuration(connPick)
-				ap.mu.Unlock()
-				closeOpenAIWSConns(evicted)
-				if p.shouldHealthCheckConn(conn) {
-					if err := conn.pingWithTimeout(openAIWSConnHealthCheckTO); err != nil {
-						conn.close()
-						p.evictConn(accountID, conn.id)
-						if retry < 1 {
-							return p.acquire(ctx, req, retry+1)
+		if routingAffinity == "" || len(ap.conns)+ap.creating >= effectiveMaxConns {
+			for _, conn := range ap.conns {
+				if conn == nil || conn == best || !conn.matchesHandshakeCompatibility(compatibility) {
+					continue
+				}
+				if conn.tryAcquire() {
+					connPick := time.Since(pickStartedAt)
+					p.recordConnPickDuration(connPick)
+					ap.mu.Unlock()
+					closeOpenAIWSConns(evicted)
+					if p.shouldHealthCheckConn(conn) {
+						if err := conn.pingWithTimeout(openAIWSConnHealthCheckTO); err != nil {
+							conn.close()
+							p.evictConn(accountID, conn.id)
+							if retry < 1 {
+								return p.acquire(ctx, req, retry+1)
+							}
+							return nil, err
 						}
-						return nil, err
 					}
+					lease := &openAIWSConnLease{pool: p, accountID: accountID, conn: conn, connPick: connPick, reused: true}
+					p.metrics.acquireReuseTotal.Add(1)
+					p.recordLastSuccessfulAcquire(accountID, acquireGeneration, req)
+					p.ensureTargetIdleAsync(accountID)
+					return lease, nil
 				}
-				lease := &openAIWSConnLease{pool: p, accountID: accountID, conn: conn, connPick: connPick, reused: true}
-				p.metrics.acquireReuseTotal.Add(1)
-				p.recordLastSuccessfulAcquire(accountID, acquireGeneration, req)
-				p.ensureTargetIdleAsync(accountID)
-				return lease, nil
 			}
 		}
 	}
 
 	if !req.ForceNewConn && len(ap.conns)+ap.creating >= effectiveMaxConns {
 		affine := p.pickLeastBusyConnWithRoutingAffinityLocked(ap, compatibility, routingAffinity)
-		if idle := p.pickOldestIdleConnWithoutHandshakeCompatibilityOrRoutingAffinityLocked(ap, compatibility, routingAffinity); idle != nil {
+		if idle := p.pickOldestIdleConnWithoutHandshakeCompatibilityLocked(ap, compatibility); idle != nil {
 			delete(ap.conns, idle.id)
 			evicted = append(evicted, idle)
 			p.metrics.scaleDownTotal.Add(1)
@@ -1241,10 +1249,9 @@ func (p *openAIWSConnPool) pickOldestIdleConnLocked(ap *openAIWSAccountPool) *op
 	return oldest
 }
 
-func (p *openAIWSConnPool) pickOldestIdleConnWithoutHandshakeCompatibilityOrRoutingAffinityLocked(
+func (p *openAIWSConnPool) pickOldestIdleConnWithoutHandshakeCompatibilityLocked(
 	ap *openAIWSAccountPool,
 	compatibility openAIWSHandshakeCompatibilityKey,
-	routingAffinity string,
 ) *openAIWSConn {
 	if ap == nil || len(ap.conns) == 0 {
 		return nil
@@ -1252,7 +1259,7 @@ func (p *openAIWSConnPool) pickOldestIdleConnWithoutHandshakeCompatibilityOrRout
 	var oldest *openAIWSConn
 	for _, conn := range ap.conns {
 		if conn == nil ||
-			(conn.matchesHandshakeCompatibility(compatibility) && conn.matchesRoutingAffinity(routingAffinity)) ||
+			conn.matchesHandshakeCompatibility(compatibility) ||
 			conn.isLeased() || conn.waiters.Load() > 0 || p.isConnPinnedLocked(ap, conn.id) {
 			continue
 		}
@@ -1800,7 +1807,7 @@ func (p *openAIWSConnPool) dialConn(ctx context.Context, req openAIWSAcquireRequ
 	}
 	id := p.nextConnID(req.Account.ID)
 	pooledConn := newOpenAIWSConn(id, req.Account.ID, conn, handshakeHeaders)
-	pooledConn.handshakeCompatibility = normalizeOpenAIWSHandshakeCompatibility(req.Headers)
+	pooledConn.handshakeCompatibility = normalizeOpenAIWSHandshakeCompatibility(req.Account, req.Headers)
 	pooledConn.routingAffinity = normalizeOpenAIWSRoutingAffinity(req.Headers)
 	return pooledConn, nil
 }
@@ -1983,7 +1990,7 @@ func cloneOpenAIWSAcquireRequestPtr(req *openAIWSAcquireRequest) *openAIWSAcquir
 func sameOpenAIWSPrewarmTarget(a, b openAIWSAcquireRequest) bool {
 	return stringsTrim(a.WSURL) == stringsTrim(b.WSURL) &&
 		stringsTrim(a.ProxyURL) == stringsTrim(b.ProxyURL) &&
-		normalizeOpenAIWSHandshakeCompatibility(a.Headers) == normalizeOpenAIWSHandshakeCompatibility(b.Headers)
+		normalizeOpenAIWSHandshakeCompatibility(a.Account, a.Headers) == normalizeOpenAIWSHandshakeCompatibility(b.Account, b.Headers)
 }
 
 func normalizeOpenAIWSBetaFeatures(headers http.Header) string {
@@ -2011,10 +2018,41 @@ func normalizeOpenAIWSBetaFeatures(headers http.Header) string {
 	return strings.Join(normalized, ",")
 }
 
-func normalizeOpenAIWSHandshakeCompatibility(headers http.Header) openAIWSHandshakeCompatibilityKey {
-	return openAIWSHandshakeCompatibilityKey{
+func normalizeOpenAIWSHandshakeCompatibility(account *Account, headers http.Header) openAIWSHandshakeCompatibilityKey {
+	key := openAIWSHandshakeCompatibilityKey{
 		betaFeatures: normalizeOpenAIWSBetaFeatures(headers),
 	}
+	mode := activeCodexFingerprintMode(account)
+	if mode == codexFingerprintOff {
+		return key
+	}
+	key.codexInstallationID = normalizeOpenAIWSStableIdentityHeader(headers, "x-codex-installation-id")
+	if mode == codexFingerprintDevice {
+		return key
+	}
+	key.sessionIDHyphen = normalizeOpenAIWSStableIdentityHeader(headers, "session-id")
+	key.sessionIDUnderscore = normalizeOpenAIWSStableIdentityHeader(headers, "session_id")
+	key.threadID = normalizeOpenAIWSStableIdentityHeader(headers, "thread-id")
+	key.clientRequestID = normalizeOpenAIWSStableIdentityHeader(headers, "x-client-request-id")
+	key.codexWindowID = normalizeOpenAIWSStableIdentityHeader(headers, "x-codex-window-id")
+	return key
+}
+
+func activeCodexFingerprintMode(account *Account) codexFingerprintMode {
+	if account == nil || account.GetCodexFingerprintMode() == codexFingerprintOff {
+		return codexFingerprintOff
+	}
+	if _, ok := codexFingerprintSeed(account.Extra); !ok {
+		return codexFingerprintOff
+	}
+	return account.GetCodexFingerprintMode()
+}
+
+func normalizeOpenAIWSStableIdentityHeader(headers http.Header, name string) string {
+	if headers == nil {
+		return ""
+	}
+	return strings.TrimSpace(headers.Get(name))
 }
 
 func normalizeOpenAIWSRoutingAffinity(headers http.Header) string {
diff --git a/backend/internal/service/openai_ws_pool_test.go b/backend/internal/service/openai_ws_pool_test.go
index f04b81d40d1d..ada18af8d3b9 100644
--- a/backend/internal/service/openai_ws_pool_test.go
+++ b/backend/internal/service/openai_ws_pool_test.go
@@ -623,6 +623,206 @@ func TestOpenAIWSConnPool_AcquireReusesOnlyMatchingBetaFeatures(t *testing.T) {
 	require.Equal(t, 2, dialer.DialCount())
 }
 
+func activeCodexFingerprintPoolAccountForTest(id int64) *Account {
+	return &Account{
+		ID:       id,
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeOAuth,
+		Extra: map[string]any{
+			codexFingerprintModeExtraKey: "session",
+			codexFingerprintSeedExtraKey: "11111111-1111-4111-8111-111111111111",
+		},
+	}
+}
+
+func stableOpenAIWSIdentityHeadersForTest() http.Header {
+	headers := make(http.Header)
+	headers.Set("X-Codex-Beta-Features", "remote_compaction_v2,responses_websockets_v2")
+	headers.Set("X-Codex-Installation-ID", "install-a")
+	headers.Set("session-id", "session-hyphen-a")
+	headers.Set("session_id", "session-underscore-a")
+	headers.Set("thread-id", "thread-a")
+	headers.Set("x-client-request-id", "client-request-a")
+	headers.Set("x-codex-window-id", "window-a")
+	return headers
+}
+
+func TestOpenAIWSConnPool_AcquireReusesSameStableIdentityWithDifferentTurnMetadata(t *testing.T) {
+	cfg := &config.Config{}
+	cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 1
+	cfg.Gateway.OpenAIWS.MinIdlePerAccount = 0
+	cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 1
+
+	pool := newOpenAIWSConnPool(cfg)
+	dialer := &openAIWSCountingDialer{}
+	pool.setClientDialerForTest(dialer)
+	account := activeCodexFingerprintPoolAccountForTest(132)
+	headers := stableOpenAIWSIdentityHeadersForTest()
+	headers.Set("Authorization", "Bearer token-a")
+	headers.Set("x-codex-turn-metadata", `{"turn_id":"turn-a"}`)
+
+	first, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+		Account: account,
+		WSURL:   "wss://example.com/v1/responses",
+		Headers: headers,
+	})
+	require.NoError(t, err)
+	firstConnID := first.ConnID()
+	first.Release()
+
+	nextHeaders := stableOpenAIWSIdentityHeadersForTest()
+	nextHeaders.Set("Authorization", "Bearer token-b")
+	nextHeaders.Set("x-codex-turn-metadata", `{"turn_id":"turn-b"}`)
+	nextHeaders.Set(openAICodexRoutingHintHeader, "model=gpt-5.6-codex;tier=priority")
+	second, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+		Account: account,
+		WSURL:   "wss://example.com/v1/responses",
+		Headers: nextHeaders,
+	})
+	require.NoError(t, err)
+	require.True(t, second.Reused())
+	require.Equal(t, firstConnID, second.ConnID())
+	second.Release()
+	require.Equal(t, 1, dialer.DialCount(), "stable identity match should ignore auth, turn metadata, and soft routing hints")
+}
+
+func TestOpenAIWSConnPool_AcquireDoesNotReuseDifferentStableIdentity(t *testing.T) {
+	for _, tt := range []struct {
+		name   string
+		header string
+		value  string
+	}{
+		{name: "installation", header: "x-codex-installation-id", value: "install-b"},
+		{name: "session hyphen", header: "session-id", value: "session-hyphen-b"},
+		{name: "session underscore", header: "session_id", value: "session-underscore-b"},
+		{name: "thread", header: "thread-id", value: "thread-b"},
+		{name: "client request", header: "x-client-request-id", value: "client-request-b"},
+		{name: "window", header: "x-codex-window-id", value: "window-b"},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := &config.Config{}
+			cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 2
+			cfg.Gateway.OpenAIWS.MinIdlePerAccount = 0
+			cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 2
+
+			pool := newOpenAIWSConnPool(cfg)
+			dialer := &openAIWSCountingDialer{}
+			pool.setClientDialerForTest(dialer)
+			account := activeCodexFingerprintPoolAccountForTest(133)
+
+			first, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+				Account: account,
+				WSURL:   "wss://example.com/v1/responses",
+				Headers: stableOpenAIWSIdentityHeadersForTest(),
+			})
+			require.NoError(t, err)
+			firstConnID := first.ConnID()
+			first.Release()
+
+			nextHeaders := stableOpenAIWSIdentityHeadersForTest()
+			nextHeaders.Set(tt.header, tt.value)
+			second, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+				Account: account,
+				WSURL:   "wss://example.com/v1/responses",
+				Headers: nextHeaders,
+			})
+			require.NoError(t, err)
+			require.False(t, second.Reused())
+			require.NotEqual(t, firstConnID, second.ConnID())
+			second.Release()
+			require.Equal(t, 2, dialer.DialCount())
+		})
+	}
+}
+
+func TestOpenAIWSConnPool_AcquireRoutingHintRemainsSoftAffinity(t *testing.T) {
+	cfg := &config.Config{}
+	cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 1
+	cfg.Gateway.OpenAIWS.MinIdlePerAccount = 0
+	cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 1
+
+	pool := newOpenAIWSConnPool(cfg)
+	dialer := &openAIWSCountingDialer{}
+	pool.setClientDialerForTest(dialer)
+	account := activeCodexFingerprintPoolAccountForTest(134)
+
+	firstHeaders := stableOpenAIWSIdentityHeadersForTest()
+	firstHeaders.Set(openAICodexRoutingHintHeader, "model=gpt-5.6-codex")
+	first, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+		Account: account,
+		WSURL:   "wss://example.com/v1/responses",
+		Headers: firstHeaders,
+	})
+	require.NoError(t, err)
+	firstConnID := first.ConnID()
+	first.Release()
+
+	secondHeaders := stableOpenAIWSIdentityHeadersForTest()
+	secondHeaders.Set(openAICodexRoutingHintHeader, "model=gpt-5.6-codex;tier=priority")
+	second, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+		Account: account,
+		WSURL:   "wss://example.com/v1/responses",
+		Headers: secondHeaders,
+	})
+	require.NoError(t, err)
+	require.True(t, second.Reused())
+	require.Equal(t, firstConnID, second.ConnID())
+	second.Release()
+	require.Equal(t, 1, dialer.DialCount())
+}
+
+func TestOpenAIWSConnPool_DeviceModeKeysOnlyInstallationIdentity(t *testing.T) {
+	cfg := &config.Config{}
+	cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 2
+	cfg.Gateway.OpenAIWS.MinIdlePerAccount = 0
+	cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 2
+
+	pool := newOpenAIWSConnPool(cfg)
+	dialer := &openAIWSCountingDialer{}
+	pool.setClientDialerForTest(dialer)
+	account := activeCodexFingerprintPoolAccountForTest(135)
+	account.Extra[codexFingerprintModeExtraKey] = "device"
+
+	firstHeaders := stableOpenAIWSIdentityHeadersForTest()
+	first, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+		Account: account,
+		WSURL:   "wss://example.com/v1/responses",
+		Headers: firstHeaders,
+	})
+	require.NoError(t, err)
+	firstConnID := first.ConnID()
+	first.Release()
+
+	sessionChanged := stableOpenAIWSIdentityHeadersForTest()
+	sessionChanged.Set("session-id", "session-hyphen-b")
+	sessionChanged.Set("session_id", "session-underscore-b")
+	sessionChanged.Set("thread-id", "thread-b")
+	sessionChanged.Set("x-client-request-id", "client-request-b")
+	sessionChanged.Set("x-codex-window-id", "window-b")
+	second, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+		Account: account,
+		WSURL:   "wss://example.com/v1/responses",
+		Headers: sessionChanged,
+	})
+	require.NoError(t, err)
+	require.True(t, second.Reused())
+	require.Equal(t, firstConnID, second.ConnID())
+	second.Release()
+
+	installationChanged := sessionChanged.Clone()
+	installationChanged.Set("x-codex-installation-id", "install-b")
+	third, err := pool.Acquire(context.Background(), openAIWSAcquireRequest{
+		Account: account,
+		WSURL:   "wss://example.com/v1/responses",
+		Headers: installationChanged,
+	})
+	require.NoError(t, err)
+	require.False(t, third.Reused())
+	require.NotEqual(t, firstConnID, third.ConnID())
+	third.Release()
+	require.Equal(t, 2, dialer.DialCount())
+}
+
 func TestOpenAIWSConnPool_AcquireReplacesIdleConnWithDifferentBetaFeatures(t *testing.T) {
 	cfg := &config.Config{}
 	cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 1
diff --git a/backend/migrations/225_backfill_codex_fingerprint_seed.sql b/backend/migrations/225_backfill_codex_fingerprint_seed.sql
new file mode 100644
index 000000000000..5c7bdbb2f6c7
--- /dev/null
+++ b/backend/migrations/225_backfill_codex_fingerprint_seed.sql
@@ -0,0 +1,21 @@
+-- Backfill system-managed Codex fingerprint seeds for enabled OpenAI OAuth accounts.
+-- Idempotent: valid canonical seeds are preserved on rerun.
+UPDATE accounts
+SET extra = jsonb_set(
+    COALESCE(extra, '{}'::jsonb),
+    '{codex_fingerprint_seed}',
+    to_jsonb(gen_random_uuid()::text),
+    true
+)
+WHERE deleted_at IS NULL
+  AND platform = 'openai'
+  AND type = 'oauth'
+  AND COALESCE(extra->>'codex_fingerprint_mode', '') IN ('device', 'session', 'full')
+  AND (
+      extra->>'codex_fingerprint_seed' IS NULL
+      OR btrim(extra->>'codex_fingerprint_seed') = ''
+      OR NOT (
+          extra->>'codex_fingerprint_seed' ~ '^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$'
+          AND extra->>'codex_fingerprint_seed' <> '00000000-0000-0000-0000-000000000000'
+      )
+  );

From c46d07ca07ee28413fd1dfa20406722e554a1c7b Mon Sep 17 00:00:00 2001
From: Heatherm Huang <heathermhuang@gmail.com>
Date: Mon, 17 Aug 2026 21:15:12 +0800
Subject: [PATCH 43/68] fix: normalize Grok response model audit aliases

---
 .../service/upstream_response_model.go        | 25 ++++++-
 .../service/upstream_response_model_test.go   | 71 +++++++++++++++++++
 2 files changed, 95 insertions(+), 1 deletion(-)

diff --git a/backend/internal/service/upstream_response_model.go b/backend/internal/service/upstream_response_model.go
index 81869c07d99f..f4e3bd4fe0a5 100644
--- a/backend/internal/service/upstream_response_model.go
+++ b/backend/internal/service/upstream_response_model.go
@@ -167,10 +167,33 @@ func upstreamModelMismatch(sentModel, responseModel string) *bool {
 		return nil
 	}
 	sentModel = strings.TrimSpace(sentModel)
-	mismatch := sentModel == "" || !strings.EqualFold(sentModel, responseModel)
+	mismatch := sentModel == "" || !upstreamModelsMatchForAudit(sentModel, responseModel)
 	return &mismatch
 }
 
+func upstreamModelsMatchForAudit(sentModel, responseModel string) bool {
+	if strings.EqualFold(sentModel, responseModel) {
+		return true
+	}
+
+	// xAI reports the runtime build ID for these supported public aliases.
+	// Canonicalize only for mismatch auditing; keep the raw response model for
+	// observability and for the separate response-model billing safeguards.
+	sentGrokModel := canonicalGrokBuildRuntimeModel(sentModel)
+	return sentGrokModel != "" && sentGrokModel == canonicalGrokBuildRuntimeModel(responseModel)
+}
+
+func canonicalGrokBuildRuntimeModel(model string) string {
+	switch strings.ToLower(strings.TrimSpace(model)) {
+	case "grok-4.5", "grok-4.5-latest", "grok-4.5-build":
+		return "grok-4.5-build"
+	case "grok-4.6", "grok-4.6-latest", "grok-4.6-build":
+		return "grok-4.6-build"
+	default:
+		return ""
+	}
+}
+
 func upstreamSentModel(requestedModel, upstreamModel string) string {
 	sentModel := strings.TrimSpace(upstreamModel)
 	if sentModel == "" {
diff --git a/backend/internal/service/upstream_response_model_test.go b/backend/internal/service/upstream_response_model_test.go
index a831e74e65be..51374ec1f320 100644
--- a/backend/internal/service/upstream_response_model_test.go
+++ b/backend/internal/service/upstream_response_model_test.go
@@ -59,6 +59,77 @@ func TestUpstreamModelMismatchThreeStateAndCaseInsensitiveComparison(t *testing.
 	require.True(t, *mismatched)
 }
 
+func TestUpstreamModelMismatchTreatsGrokBuildRuntimeIDsAsAliases(t *testing.T) {
+	tests := []struct {
+		name          string
+		sentModel     string
+		responseModel string
+	}{
+		{
+			name:          "issue 5634 grok 4.6",
+			sentModel:     "grok-4.6",
+			responseModel: "grok-4.6-build",
+		},
+		{
+			name:          "grok 4.6 latest",
+			sentModel:     "grok-4.6-latest",
+			responseModel: "grok-4.6-build",
+		},
+		{
+			name:          "issue 5647 grok 4.5 latest",
+			sentModel:     "grok-4.5-latest",
+			responseModel: "grok-4.5-build",
+		},
+		{
+			name:          "grok 4.5 canonical",
+			sentModel:     "grok-4.5",
+			responseModel: "GROK-4.5-BUILD",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			mismatch := upstreamModelMismatch(tt.sentModel, tt.responseModel)
+
+			require.NotNil(t, mismatch)
+			require.False(t, *mismatch)
+		})
+	}
+}
+
+func TestUpstreamModelMismatchDoesNotCollapseDifferentModels(t *testing.T) {
+	tests := []struct {
+		name          string
+		sentModel     string
+		responseModel string
+	}{
+		{
+			name:          "different grok versions",
+			sentModel:     "grok-4.5",
+			responseModel: "grok-4.6-build",
+		},
+		{
+			name:          "unrelated build suffix",
+			sentModel:     "gpt-5.5",
+			responseModel: "gpt-5.5-build",
+		},
+		{
+			name:          "different grok runtime",
+			sentModel:     "grok-build-0.1",
+			responseModel: "grok-4.5-build",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			mismatch := upstreamModelMismatch(tt.sentModel, tt.responseModel)
+
+			require.NotNil(t, mismatch)
+			require.True(t, *mismatch)
+		})
+	}
+}
+
 func TestObserveOpenAISSEBodyIgnoresMalformedPayload(t *testing.T) {
 	observer := &upstreamResponseModelObserver{}
 	observeOpenAISSEBody(observer, "data: not-json\n\ndata: {\"type\":\"response.completed\",\"response\":{\"model\":\"gpt-5.4\"}}\n\n")

From ab0fcd1a0e6fa8f9974ba21c6c3478bebdd07be4 Mon Sep 17 00:00:00 2001
From: feeeei <g540203430@gmail.com>
Date: Mon, 17 Aug 2026 18:49:59 +0800
Subject: [PATCH 44/68] =?UTF-8?q?fix(gemini):=20Skipped=20=E9=94=99?=
 =?UTF-8?q?=E8=AF=AF=E7=AD=96=E7=95=A5=E5=AF=B9=E9=BD=90=20OpenAI=EF=BC=8C?=
 =?UTF-8?q?=E4=B8=8A=E6=B8=B8=204xx=20=E4=B8=8D=E5=86=8D=E7=A1=AC=E6=94=B9?=
 =?UTF-8?q?=20500?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

ErrorPolicySkipped（池模式、或自定义错误码未命中）原来在响应写出上
自成一派：v1beta 原生把上游 4xx 硬改 500 后原文透传，/v1/messages
硬传 500 进映射（客户端拿到 502）。下游网关据此把请求级错误当可重
试的服务端故障反复换号，耗尽后改写成 All available accounts
exhausted（2026-08-17 gemini 生产事故链）。现对齐 OpenAI 路径语义：

- Skipped 只豁免账号状态标记，不豁免换号：可 failover 状态码一律
  返回 UpstreamFailoverError（poolModeSkippedFailoverError 泛化为
  skippedErrorPolicyFailoverError，同账号重试标记仍仅池模式携带）
- 池模式的不可 failover 4xx 保真：v1beta 原码+原文透传（新
  writeGeminiNativeUpstreamError 与 ErrorPolicyNone 共用同一写出，
  并补记 ops 事件），/v1/messages 与 chat completions 按真实状态
  码映射
- 自定义错误码未命中且不可 failover：三路径统一 500 + "Upstream
  gateway error" 固定文案，上游细节仅记 ops 错误日志
- 400 属确定性请求错误：mapped 写出回传脱敏后的上游 message，客户
  端可据此定位非法字段
---
 .../gemini_chat_completions_compat_service.go |  12 +-
 .../gemini_error_policy_skipped_write_test.go | 213 ++++++++++++++++++
 .../service/gemini_error_policy_test.go       |  48 ++--
 .../service/gemini_messages_compat_service.go | 161 ++++++++-----
 4 files changed, 360 insertions(+), 74 deletions(-)
 create mode 100644 backend/internal/service/gemini_error_policy_skipped_write_test.go

diff --git a/backend/internal/service/gemini_chat_completions_compat_service.go b/backend/internal/service/gemini_chat_completions_compat_service.go
index e76fbcee15be..3ee7a4af9e4d 100644
--- a/backend/internal/service/gemini_chat_completions_compat_service.go
+++ b/backend/internal/service/gemini_chat_completions_compat_service.go
@@ -240,6 +240,11 @@ func (s *GeminiMessagesCompatService) forwardClaudeBodyAsChatCompletions(
 			}
 		}
 
+		if policy == ErrorPolicySkipped && account.IsCustomErrorCodesEnabled() {
+			return nil, s.writeGeminiCustomCodeSkippedError(c, account, resp.StatusCode, requestID, evBody, func() {
+				_ = s.writeChatCompletionsError(c, http.StatusInternalServerError, "api_error", geminiCustomCodeSkippedClientMessage)
+			})
+		}
 		return nil, s.writeGeminiChatCompletionsMappedError(c, account, resp.StatusCode, requestID, evBody)
 	}
 
@@ -856,8 +861,13 @@ func (s *GeminiMessagesCompatService) writeGeminiChatCompletionsMappedError(
 		if errType == "upstream_error" {
 			errType = "invalid_request_error"
 		}
+		// 400 是确定性的请求错误：回传上游 message（已脱敏），客户端据此定位非法字段。
 		if errMsg == "Upstream request failed" {
-			errMsg = "Invalid request"
+			if upstreamMsg != "" {
+				errMsg = upstreamMsg
+			} else {
+				errMsg = "Invalid request"
+			}
 		}
 	case http.StatusNotFound:
 		statusCode = http.StatusNotFound
diff --git a/backend/internal/service/gemini_error_policy_skipped_write_test.go b/backend/internal/service/gemini_error_policy_skipped_write_test.go
new file mode 100644
index 000000000000..41c37e24da3d
--- /dev/null
+++ b/backend/internal/service/gemini_error_policy_skipped_write_test.go
@@ -0,0 +1,213 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+// ---------------------------------------------------------------------------
+// ErrorPolicySkipped 的客户端写出契约（与 OpenAI 网关路径对齐）：
+//   - 池模式：不可 failover 的 4xx 按上游原始状态码/响应体保真写出，不改写成 5xx；
+//   - 自定义错误码未命中：统一 500 + 固定文案，上游细节只进 ops 错误日志；
+//   - 可 failover 的状态码（两种账号）一律换号，不透传。
+// ---------------------------------------------------------------------------
+
+const geminiSkippedTestUpstreamMsg = "antigravity executor: invalid Gemini function call history"
+
+func geminiSkippedTestUpstreamBody() string {
+	return `{"error":{"code":null,"message":"` + geminiSkippedTestUpstreamMsg + `","param":"","type":"invalid_request_error"}}`
+}
+
+func newGeminiSkippedWriteService(status int, body string) (*GeminiMessagesCompatService, *geminiCompatHTTPUpstreamStub) {
+	httpStub := &geminiCompatHTTPUpstreamStub{
+		response: &http.Response{
+			StatusCode: status,
+			Header:     http.Header{"Content-Type": []string{"application/json"}},
+			Body:       io.NopCloser(strings.NewReader(body)),
+		},
+	}
+	svc := &GeminiMessagesCompatService{
+		httpUpstream:     httpStub,
+		cfg:              &config.Config{},
+		rateLimitService: NewRateLimitService(&errorPolicyRepoStub{}, nil, &config.Config{}, nil, nil),
+	}
+	return svc, httpStub
+}
+
+func geminiPoolModeAPIKeyAccount() *Account {
+	return &Account{
+		ID:       700,
+		Platform: PlatformGemini,
+		Type:     AccountTypeAPIKey,
+		Credentials: map[string]any{
+			"api_key":   "test-key",
+			"pool_mode": true,
+		},
+	}
+}
+
+func geminiCustomCodesAPIKeyAccount() *Account {
+	return &Account{
+		ID:       701,
+		Platform: PlatformGemini,
+		Type:     AccountTypeAPIKey,
+		Credentials: map[string]any{
+			"api_key":                    "test-key",
+			"custom_error_codes_enabled": true,
+			"custom_error_codes":         []any{float64(429)},
+		},
+	}
+}
+
+func newGeminiNativeTestContext(t *testing.T) (*gin.Context, *httptest.ResponseRecorder) {
+	t.Helper()
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1beta/models/gemini-2.5-flash:generateContent", strings.NewReader("{}"))
+	return c, rec
+}
+
+func TestGeminiForwardNative_PoolModeSkipped400PassthroughRealStatus(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	upstreamBody := geminiSkippedTestUpstreamBody()
+	svc, _ := newGeminiSkippedWriteService(http.StatusBadRequest, upstreamBody)
+	c, rec := newGeminiNativeTestContext(t)
+
+	result, err := svc.ForwardNative(context.Background(), c, geminiPoolModeAPIKeyAccount(),
+		"gemini-2.5-flash", "generateContent", false, []byte(`{"contents":[{"role":"user","parts":[{"text":"hi"}]}]}`))
+
+	require.Nil(t, result)
+	require.Error(t, err)
+	var failoverErr *UpstreamFailoverError
+	require.False(t, errors.As(err, &failoverErr), "池模式 400 不应换号")
+	require.Contains(t, err.Error(), "gemini upstream error: 400")
+	require.Equal(t, http.StatusBadRequest, rec.Code, "状态码应保真为上游 400")
+	require.Equal(t, upstreamBody, rec.Body.String(), "响应体应原样透传")
+}
+
+func TestGeminiForwardNative_PoolModeSkipped503Failover(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc, _ := newGeminiSkippedWriteService(http.StatusServiceUnavailable, `{"error":{"message":"Upstream service temporarily unavailable"}}`)
+	c, rec := newGeminiNativeTestContext(t)
+
+	result, err := svc.ForwardNative(context.Background(), c, geminiPoolModeAPIKeyAccount(),
+		"gemini-2.5-flash", "generateContent", false, []byte(`{"contents":[{"role":"user","parts":[{"text":"hi"}]}]}`))
+
+	require.Nil(t, result)
+	var failoverErr *UpstreamFailoverError
+	require.True(t, errors.As(err, &failoverErr), "池模式 503 应换号")
+	require.Equal(t, http.StatusServiceUnavailable, failoverErr.StatusCode)
+	require.Zero(t, rec.Body.Len(), "换号场景不应写客户端响应")
+}
+
+func TestGeminiForwardNative_CustomCodesMiss400HiddenAs500(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc, _ := newGeminiSkippedWriteService(http.StatusBadRequest, geminiSkippedTestUpstreamBody())
+	c, rec := newGeminiNativeTestContext(t)
+
+	result, err := svc.ForwardNative(context.Background(), c, geminiCustomCodesAPIKeyAccount(),
+		"gemini-2.5-flash", "generateContent", false, []byte(`{"contents":[{"role":"user","parts":[{"text":"hi"}]}]}`))
+
+	require.Nil(t, result)
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "not in custom error codes")
+	require.Equal(t, http.StatusInternalServerError, rec.Code)
+
+	var got map[string]any
+	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &got))
+	errObj, ok := got["error"].(map[string]any)
+	require.True(t, ok)
+	require.Equal(t, geminiCustomCodeSkippedClientMessage, errObj["message"])
+	require.NotContains(t, rec.Body.String(), geminiSkippedTestUpstreamMsg, "上游细节不应透传给客户端")
+}
+
+func TestGeminiForwardNative_CustomCodesMiss500Failover(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc, _ := newGeminiSkippedWriteService(http.StatusInternalServerError, `{"error":{"message":"internal"}}`)
+	c, rec := newGeminiNativeTestContext(t)
+
+	result, err := svc.ForwardNative(context.Background(), c, geminiCustomCodesAPIKeyAccount(),
+		"gemini-2.5-flash", "generateContent", false, []byte(`{"contents":[{"role":"user","parts":[{"text":"hi"}]}]}`))
+
+	require.Nil(t, result)
+	var failoverErr *UpstreamFailoverError
+	require.True(t, errors.As(err, &failoverErr), "自定义错误码未命中的 500 应换号")
+	require.Equal(t, http.StatusInternalServerError, failoverErr.StatusCode)
+	require.False(t, failoverErr.RetryableOnSameAccount, "非池模式不应同账号重试")
+	require.Zero(t, rec.Body.Len())
+}
+
+func TestGeminiForwardAsChatCompletions_CustomCodesMiss400HiddenAs500(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc, _ := newGeminiSkippedWriteService(http.StatusBadRequest, geminiSkippedTestUpstreamBody())
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	body := []byte(`{"model":"gemini-2.5-flash","messages":[{"role":"user","content":"hi"}]}`)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(string(body)))
+
+	result, err := svc.ForwardAsChatCompletions(context.Background(), c, geminiCustomCodesAPIKeyAccount(), body)
+
+	require.Nil(t, result)
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "not in custom error codes")
+	require.Equal(t, http.StatusInternalServerError, rec.Code)
+
+	var got map[string]any
+	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &got))
+	errObj, ok := got["error"].(map[string]any)
+	require.True(t, ok)
+	require.Equal(t, "api_error", errObj["type"])
+	require.Equal(t, geminiCustomCodeSkippedClientMessage, errObj["message"])
+}
+
+func TestGeminiForwardAsChatCompletions_PoolMode400KeepsUpstreamMessage(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc, _ := newGeminiSkippedWriteService(http.StatusBadRequest, geminiSkippedTestUpstreamBody())
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	body := []byte(`{"model":"gemini-2.5-flash","messages":[{"role":"user","content":"hi"}]}`)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(string(body)))
+
+	result, err := svc.ForwardAsChatCompletions(context.Background(), c, geminiPoolModeAPIKeyAccount(), body)
+
+	require.Nil(t, result)
+	require.Error(t, err)
+	require.Equal(t, http.StatusBadRequest, rec.Code, "状态码应保真为上游 400")
+
+	var got map[string]any
+	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &got))
+	errObj, ok := got["error"].(map[string]any)
+	require.True(t, ok)
+	require.Equal(t, "invalid_request_error", errObj["type"])
+	require.Equal(t, geminiSkippedTestUpstreamMsg, errObj["message"], "应回传上游 message")
+}
+
+func TestWriteGeminiMappedError_400KeepsUpstreamMessage(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := &GeminiMessagesCompatService{cfg: &config.Config{}}
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
+
+	err := svc.writeGeminiMappedError(c, &Account{ID: 702, Platform: PlatformGemini}, http.StatusBadRequest, "req-1", []byte(geminiSkippedTestUpstreamBody()))
+
+	require.Error(t, err)
+	require.Equal(t, http.StatusBadRequest, rec.Code)
+	var got map[string]any
+	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &got))
+	errObj, ok := got["error"].(map[string]any)
+	require.True(t, ok)
+	require.Equal(t, geminiSkippedTestUpstreamMsg, errObj["message"], "应回传上游 message")
+}
diff --git a/backend/internal/service/gemini_error_policy_test.go b/backend/internal/service/gemini_error_policy_test.go
index 46024077a589..807ed7056b14 100644
--- a/backend/internal/service/gemini_error_policy_test.go
+++ b/backend/internal/service/gemini_error_policy_test.go
@@ -218,7 +218,7 @@ func TestGeminiErrorPolicyIntegration(t *testing.T) {
 			expectHandleError: true,
 		},
 		{
-			name: "custom_codes_skipped_500_no_failover",
+			name: "custom_codes_skipped_500_failover",
 			account: &Account{
 				ID:       201,
 				Type:     AccountTypeAPIKey,
@@ -230,6 +230,22 @@ func TestGeminiErrorPolicyIntegration(t *testing.T) {
 			},
 			statusCode:        500,
 			respBody:          []byte(`{"error":"internal"}`),
+			expectFailover:    true,
+			expectHandleError: false,
+		},
+		{
+			name: "custom_codes_skipped_400_no_failover",
+			account: &Account{
+				ID:       205,
+				Type:     AccountTypeAPIKey,
+				Platform: PlatformGemini,
+				Credentials: map[string]any{
+					"custom_error_codes_enabled": true,
+					"custom_error_codes":         []any{float64(429)},
+				},
+			},
+			statusCode:        400,
+			respBody:          []byte(`{"error":"bad request"}`),
 			expectFailover:    false,
 			expectHandleError: false,
 		},
@@ -311,9 +327,9 @@ func TestGeminiErrorPolicyIntegration(t *testing.T) {
 				policy := svc.rateLimitService.CheckErrorPolicy(ctx, account, statusCode, respBody, "gemini-2.5-pro")
 				switch policy {
 				case ErrorPolicySkipped:
-					// Skipped → return error directly (no handleGeminiUpstreamError, no failover)
-					gotFailover = false
+					// Skipped → 不标记账号状态；可 failover 的状态码仍换号
 					handleErrorCalled = false
+					gotFailover = svc.skippedErrorPolicyFailoverError(c, account, statusCode, respBody, "req-test") != nil
 					goto verify
 				case ErrorPolicyMatched:
 					svc.handleGeminiUpstreamError(ctx, account, statusCode, headers, respBody)
@@ -353,12 +369,12 @@ func TestGeminiErrorPolicyIntegration(t *testing.T) {
 }
 
 // ---------------------------------------------------------------------------
-// TestPoolModeSkippedFailoverError — pool-mode accounts hitting
-// ErrorPolicySkipped must failover (align with other platform forwards)
-// instead of passing the upstream error through to the client.
+// TestSkippedErrorPolicyFailoverError — ErrorPolicySkipped（池模式、或自定义
+// 错误码未命中）不豁免换号：可 failover 的状态码返回 UpstreamFailoverError，
+// 仅池模式账号可携带同账号重试标记。
 // ---------------------------------------------------------------------------
 
-func TestPoolModeSkippedFailoverError(t *testing.T) {
+func TestSkippedErrorPolicyFailoverError(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	svc := &GeminiMessagesCompatService{}
 
@@ -369,6 +385,13 @@ func TestPoolModeSkippedFailoverError(t *testing.T) {
 		}
 		return &Account{ID: 300, Type: AccountTypeAPIKey, Platform: PlatformGemini, Credentials: creds}
 	}
+	customCodesAccount := &Account{
+		ID: 301, Type: AccountTypeAPIKey, Platform: PlatformGemini,
+		Credentials: map[string]any{
+			"custom_error_codes_enabled": true,
+			"custom_error_codes":         []any{float64(429)},
+		},
+	}
 
 	tests := []struct {
 		name              string
@@ -383,13 +406,8 @@ func TestPoolModeSkippedFailoverError(t *testing.T) {
 			"pool_mode_retry_status_codes": []any{float64(500)},
 		}), 500, true, true},
 		{"pool_400_not_failover_worthy", poolAccount(nil), 400, false, false},
-		{"non_pool_account_keeps_passthrough", &Account{
-			ID: 301, Type: AccountTypeAPIKey, Platform: PlatformGemini,
-			Credentials: map[string]any{
-				"custom_error_codes_enabled": true,
-				"custom_error_codes":         []any{float64(429)},
-			},
-		}, 500, false, false},
+		{"custom_codes_miss_500_failover_no_same_account_retry", customCodesAccount, 500, true, false},
+		{"custom_codes_miss_400_not_failover_worthy", customCodesAccount, 400, false, false},
 	}
 
 	for _, tt := range tests {
@@ -399,7 +417,7 @@ func TestPoolModeSkippedFailoverError(t *testing.T) {
 			c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
 
 			body := []byte(`{"error":{"code":"bad_response_status_code","message":"openai_error"}}`)
-			failoverErr := svc.poolModeSkippedFailoverError(c, tt.account, tt.statusCode, body, "req-1")
+			failoverErr := svc.skippedErrorPolicyFailoverError(c, tt.account, tt.statusCode, body, "req-1")
 
 			if !tt.expectFailover {
 				require.Nil(t, failoverErr)
diff --git a/backend/internal/service/gemini_messages_compat_service.go b/backend/internal/service/gemini_messages_compat_service.go
index 0abd14937aff..82cf9a20f3bc 100644
--- a/backend/internal/service/gemini_messages_compat_service.go
+++ b/backend/internal/service/gemini_messages_compat_service.go
@@ -947,10 +947,16 @@ func (s *GeminiMessagesCompatService) Forward(ctx context.Context, c *gin.Contex
 				if upstreamReqID == "" {
 					upstreamReqID = resp.Header.Get("x-goog-request-id")
 				}
-				if failoverErr := s.poolModeSkippedFailoverError(c, account, resp.StatusCode, respBody, upstreamReqID); failoverErr != nil {
+				if failoverErr := s.skippedErrorPolicyFailoverError(c, account, resp.StatusCode, respBody, upstreamReqID); failoverErr != nil {
 					return nil, failoverErr
 				}
-				return nil, s.writeGeminiMappedError(c, account, http.StatusInternalServerError, upstreamReqID, respBody)
+				if account.IsCustomErrorCodesEnabled() {
+					return nil, s.writeGeminiCustomCodeSkippedError(c, account, resp.StatusCode, upstreamReqID, respBody, func() {
+						_ = s.writeClaudeError(c, http.StatusInternalServerError, "api_error", geminiCustomCodeSkippedClientMessage)
+					})
+				}
+				// 池模式：客户端写出与 ErrorPolicyNone 相同（按上游真实状态码映射），仅跳过账号状态标记。
+				return nil, s.writeGeminiMappedError(c, account, resp.StatusCode, upstreamReqID, respBody)
 			case ErrorPolicyMatched, ErrorPolicyTempUnscheduled:
 				if policy == ErrorPolicyMatched {
 					s.handleGeminiUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)
@@ -1460,17 +1466,16 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin.
 			policy := s.rateLimitService.CheckErrorPolicy(ctx, account, resp.StatusCode, respBody, mappedModel)
 			switch policy {
 			case ErrorPolicySkipped:
-				if failoverErr := s.poolModeSkippedFailoverError(c, account, resp.StatusCode, respBody, requestID); failoverErr != nil {
+				if failoverErr := s.skippedErrorPolicyFailoverError(c, account, resp.StatusCode, respBody, requestID); failoverErr != nil {
 					return nil, failoverErr
 				}
-				respBody = unwrapIfNeeded(isOAuth, respBody)
-				contentType := resp.Header.Get("Content-Type")
-				if contentType == "" {
-					contentType = "application/json"
+				if account.IsCustomErrorCodesEnabled() {
+					return nil, s.writeGeminiCustomCodeSkippedError(c, account, resp.StatusCode, requestID, respBody, func() {
+						_ = s.writeGoogleError(c, http.StatusInternalServerError, geminiCustomCodeSkippedClientMessage)
+					})
 				}
-				MarkResponseCommitted(c)
-				c.Data(http.StatusInternalServerError, contentType, respBody)
-				return nil, fmt.Errorf("gemini upstream error: %d (skipped by error policy)", resp.StatusCode)
+				// 池模式：客户端写出与 ErrorPolicyNone 相同（状态码/响应体保真），仅跳过账号状态标记。
+				return nil, s.writeGeminiNativeUpstreamError(c, account, resp, respBody, requestID, isOAuth)
 			case ErrorPolicyMatched, ErrorPolicyTempUnscheduled:
 				if policy == ErrorPolicyMatched {
 					s.handleGeminiUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)
@@ -1555,40 +1560,7 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin.
 			return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: evBody}
 		}
 
-		respBody = unwrapIfNeeded(isOAuth, respBody)
-		upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(respBody))
-		upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg)
-		upstreamDetail := ""
-		if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody {
-			maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes
-			if maxBytes <= 0 {
-				maxBytes = 2048
-			}
-			upstreamDetail = truncateString(string(respBody), maxBytes)
-			logger.LegacyPrintf("service.gemini_messages_compat", "[Gemini] native upstream error %d: %s", resp.StatusCode, truncateForLog(respBody, s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes))
-		}
-		setOpsUpstreamError(c, resp.StatusCode, upstreamMsg, upstreamDetail)
-		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
-			Platform:           account.Platform,
-			AccountID:          account.ID,
-			AccountName:        account.Name,
-			UpstreamStatusCode: resp.StatusCode,
-			UpstreamRequestID:  requestID,
-			Kind:               "http_error",
-			Message:            upstreamMsg,
-			Detail:             upstreamDetail,
-		})
-
-		contentType := resp.Header.Get("Content-Type")
-		if contentType == "" {
-			contentType = "application/json"
-		}
-		MarkResponseCommitted(c)
-		c.Data(resp.StatusCode, contentType, respBody)
-		if upstreamMsg == "" {
-			return nil, fmt.Errorf("gemini upstream error: %d", resp.StatusCode)
-		}
-		return nil, fmt.Errorf("gemini upstream error: %d message=%s", resp.StatusCode, upstreamMsg)
+		return nil, s.writeGeminiNativeUpstreamError(c, account, resp, respBody, requestID, isOAuth)
 	}
 
 	var usage *ClaudeUsage
@@ -1695,22 +1667,16 @@ func (s *GeminiMessagesCompatService) shouldFailoverGeminiUpstreamError(statusCo
 	}
 }
 
-// poolModeSkippedFailoverError 池模式账号命中 ErrorPolicySkipped 时构造 failover 错误：
-// 可 failover 的状态码返回 UpstreamFailoverError，交给 handler 层按 pool_mode_retry_count
-// 同账号重试后换号；返回 nil 表示不适用（非池模式或状态码不可 failover），由调用方透传。
-func (s *GeminiMessagesCompatService) poolModeSkippedFailoverError(c *gin.Context, account *Account, statusCode int, respBody []byte, upstreamRequestID string) *UpstreamFailoverError {
-	if !account.IsPoolMode() || !s.shouldFailoverGeminiUpstreamError(statusCode) {
+// skippedErrorPolicyFailoverError 命中 ErrorPolicySkipped（池模式、或自定义错误码未命中）
+// 时构造 failover 错误：可 failover 的状态码返回 UpstreamFailoverError，交给 handler 层换号
+// （池模式账号按 pool_mode_retry_count 先同账号重试）；返回 nil 表示状态码不可 failover，
+// 由调用方决定客户端写出。Skipped 只豁免账号状态标记，不豁免换号，与 OpenAI 网关路径一致。
+func (s *GeminiMessagesCompatService) skippedErrorPolicyFailoverError(c *gin.Context, account *Account, statusCode int, respBody []byte, upstreamRequestID string) *UpstreamFailoverError {
+	if !s.shouldFailoverGeminiUpstreamError(statusCode) {
 		return nil
 	}
 	upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(respBody)))
-	upstreamDetail := ""
-	if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody {
-		maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes
-		if maxBytes <= 0 {
-			maxBytes = 2048
-		}
-		upstreamDetail = truncateString(string(respBody), maxBytes)
-	}
+	upstreamDetail := s.upstreamErrorDetail(respBody)
 	appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 		Platform:           account.Platform,
 		AccountID:          account.ID,
@@ -1724,8 +1690,83 @@ func (s *GeminiMessagesCompatService) poolModeSkippedFailoverError(c *gin.Contex
 	return &UpstreamFailoverError{
 		StatusCode:             statusCode,
 		ResponseBody:           respBody,
-		RetryableOnSameAccount: account.IsPoolModeRetryableStatus(statusCode),
+		RetryableOnSameAccount: account.IsPoolMode() && account.IsPoolModeRetryableStatus(statusCode),
+	}
+}
+
+// geminiCustomCodeSkippedClientMessage 自定义错误码未命中时对客户端隐藏上游细节的固定文案，
+// 与 OpenAI 网关路径同场景的文案一致。
+const geminiCustomCodeSkippedClientMessage = "Upstream gateway error"
+
+// upstreamErrorDetail 按配置截断上游错误响应体，用于 ops 错误日志的 Detail 字段；
+// 未开启 LogUpstreamErrorBody 时返回空。
+func (s *GeminiMessagesCompatService) upstreamErrorDetail(body []byte) string {
+	if s.cfg == nil || !s.cfg.Gateway.LogUpstreamErrorBody {
+		return ""
+	}
+	maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes
+	if maxBytes <= 0 {
+		maxBytes = 2048
+	}
+	return truncateString(string(body), maxBytes)
+}
+
+// writeGeminiCustomCodeSkippedError 处理自定义错误码未命中且不可 failover 的上游错误：
+// 客户端统一收到 500 + 固定文案（由 write 按端点格式写出），不透传上游细节；
+// 上游真实状态码与错误信息仅记录到 ops 错误日志。
+func (s *GeminiMessagesCompatService) writeGeminiCustomCodeSkippedError(c *gin.Context, account *Account, upstreamStatus int, upstreamRequestID string, body []byte, write func()) error {
+	upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(body)))
+	upstreamDetail := s.upstreamErrorDetail(body)
+	setOpsUpstreamError(c, upstreamStatus, upstreamMsg, upstreamDetail)
+	appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+		Platform:           account.Platform,
+		AccountID:          account.ID,
+		AccountName:        account.Name,
+		UpstreamStatusCode: upstreamStatus,
+		UpstreamRequestID:  upstreamRequestID,
+		Kind:               "http_error",
+		Message:            upstreamMsg,
+		Detail:             upstreamDetail,
+	})
+	write()
+	if upstreamMsg == "" {
+		return fmt.Errorf("gemini upstream error: %d (not in custom error codes)", upstreamStatus)
+	}
+	return fmt.Errorf("gemini upstream error: %d (not in custom error codes) message=%s", upstreamStatus, upstreamMsg)
+}
+
+// writeGeminiNativeUpstreamError 将不可 failover 的上游错误按原始状态码与响应体透传给客户端，
+// 并记录 ops 错误事件。状态码保真：下游据此区分请求级错误与可重试的链路故障。
+func (s *GeminiMessagesCompatService) writeGeminiNativeUpstreamError(c *gin.Context, account *Account, resp *http.Response, respBody []byte, requestID string, isOAuth bool) error {
+	respBody = unwrapIfNeeded(isOAuth, respBody)
+	upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(respBody))
+	upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg)
+	upstreamDetail := s.upstreamErrorDetail(respBody)
+	if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody {
+		logger.LegacyPrintf("service.gemini_messages_compat", "[Gemini] native upstream error %d: %s", resp.StatusCode, truncateForLog(respBody, s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes))
+	}
+	setOpsUpstreamError(c, resp.StatusCode, upstreamMsg, upstreamDetail)
+	appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+		Platform:           account.Platform,
+		AccountID:          account.ID,
+		AccountName:        account.Name,
+		UpstreamStatusCode: resp.StatusCode,
+		UpstreamRequestID:  requestID,
+		Kind:               "http_error",
+		Message:            upstreamMsg,
+		Detail:             upstreamDetail,
+	})
+
+	contentType := resp.Header.Get("Content-Type")
+	if contentType == "" {
+		contentType = "application/json"
 	}
+	MarkResponseCommitted(c)
+	c.Data(resp.StatusCode, contentType, respBody)
+	if upstreamMsg == "" {
+		return fmt.Errorf("gemini upstream error: %d", resp.StatusCode)
+	}
+	return fmt.Errorf("gemini upstream error: %d message=%s", resp.StatusCode, upstreamMsg)
 }
 
 func sleepGeminiBackoff(attempt int) {
@@ -1827,6 +1868,10 @@ func (s *GeminiMessagesCompatService) writeGeminiMappedError(c *gin.Context, acc
 		if errType == "" {
 			errType = "invalid_request_error"
 		}
+		// 400 是确定性的请求错误：回传上游 message（已脱敏），客户端据此定位非法字段。
+		if errMsg == "" {
+			errMsg = upstreamMsg
+		}
 		if errMsg == "" {
 			errMsg = "Invalid request"
 		}

From 615e6901ea41530c5439679b57bf291e31948e95 Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 03:50:01 +0000
Subject: [PATCH 45/68] feat(channel-monitor): add quota mode schema (migration
 226 + ent)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 迁移 226：provider CHECK 扩到 8 平台（monitors + request_templates）、
  channel_monitors 加 check_mode/account_id（FK ON DELETE SET NULL）、
  channel_monitor_histories 加 quota JSONB、插入公开设置
  channel_monitor_show_quota（默认 false）
- ent schema 同步：provider 枚举 +4、endpoint 去 NotEmpty（quota 模式存空串）、
  新增 check_mode/account_id 字段、history 加 quota JSON 快照
- domain 新增归一化配额快照类型 MonitorQuotaSnapshot/Tier/Balance
  （放 domain 是因为 ent schema 需要引用，service 会造成 import cycle）
---
 backend/ent/channelmonitor.go                 |  31 ++-
 backend/ent/channelmonitor/channelmonitor.go  |  34 ++-
 backend/ent/channelmonitor/where.go           | 125 +++++++++
 backend/ent/channelmonitor_create.go          | 168 +++++++++++++
 backend/ent/channelmonitor_update.go          | 116 +++++++++
 backend/ent/channelmonitorhistory.go          |  17 ++
 .../channelmonitorhistory.go                  |   3 +
 backend/ent/channelmonitorhistory/where.go    |  10 +
 backend/ent/channelmonitorhistory_create.go   |  71 ++++++
 backend/ent/channelmonitorhistory_update.go   |  37 +++
 .../channelmonitorrequesttemplate.go          |  14 +-
 backend/ent/group.go                          |   2 +-
 backend/ent/migrate/schema.go                 |  28 ++-
 backend/ent/mutation.go                       | 238 +++++++++++++++++-
 backend/ent/runtime/runtime.go                |  46 ++--
 backend/ent/schema/channel_monitor.go         |  24 +-
 backend/ent/schema/channel_monitor_history.go |   6 +
 .../channel_monitor_request_template.go       |   3 +-
 .../internal/domain/channel_monitor_quota.go  |  55 ++++
 .../226_channel_monitor_quota_mode.sql        |  78 ++++++
 20 files changed, 1049 insertions(+), 57 deletions(-)
 create mode 100644 backend/internal/domain/channel_monitor_quota.go
 create mode 100644 backend/migrations/226_channel_monitor_quota_mode.sql

diff --git a/backend/ent/channelmonitor.go b/backend/ent/channelmonitor.go
index e5697863f0d1..e74b6528ccaf 100644
--- a/backend/ent/channelmonitor.go
+++ b/backend/ent/channelmonitor.go
@@ -27,9 +27,13 @@ type ChannelMonitor struct {
 	Name string `json:"name,omitempty"`
 	// Provider holds the value of the "provider" field.
 	Provider channelmonitor.Provider `json:"provider,omitempty"`
+	// probe = LLM probe (default); quota = account usage only; quota_probe = both
+	CheckMode string `json:"check_mode,omitempty"`
+	// AccountID holds the value of the "account_id" field.
+	AccountID *int64 `json:"account_id,omitempty"`
 	// OpenAI request protocol: chat_completions or responses; non-OpenAI uses chat_completions
 	APIMode string `json:"api_mode,omitempty"`
-	// Provider base origin, e.g. https://api.openai.com
+	// Provider base origin, e.g. https://api.openai.com; empty for quota-only monitors
 	Endpoint string `json:"endpoint,omitempty"`
 	// AES-256-GCM encrypted API key
 	APIKeyEncrypted string `json:"-"`
@@ -114,9 +118,9 @@ func (*ChannelMonitor) scanValues(columns []string) ([]any, error) {
 			values[i] = new([]byte)
 		case channelmonitor.FieldEnabled:
 			values[i] = new(sql.NullBool)
-		case channelmonitor.FieldID, channelmonitor.FieldIntervalSeconds, channelmonitor.FieldJitterSeconds, channelmonitor.FieldCreatedBy, channelmonitor.FieldTemplateID:
+		case channelmonitor.FieldID, channelmonitor.FieldAccountID, channelmonitor.FieldIntervalSeconds, channelmonitor.FieldJitterSeconds, channelmonitor.FieldCreatedBy, channelmonitor.FieldTemplateID:
 			values[i] = new(sql.NullInt64)
-		case channelmonitor.FieldName, channelmonitor.FieldProvider, channelmonitor.FieldAPIMode, channelmonitor.FieldEndpoint, channelmonitor.FieldAPIKeyEncrypted, channelmonitor.FieldPrimaryModel, channelmonitor.FieldGroupName, channelmonitor.FieldBodyOverrideMode:
+		case channelmonitor.FieldName, channelmonitor.FieldProvider, channelmonitor.FieldCheckMode, channelmonitor.FieldAPIMode, channelmonitor.FieldEndpoint, channelmonitor.FieldAPIKeyEncrypted, channelmonitor.FieldPrimaryModel, channelmonitor.FieldGroupName, channelmonitor.FieldBodyOverrideMode:
 			values[i] = new(sql.NullString)
 		case channelmonitor.FieldCreatedAt, channelmonitor.FieldUpdatedAt, channelmonitor.FieldLastCheckedAt:
 			values[i] = new(sql.NullTime)
@@ -165,6 +169,19 @@ func (_m *ChannelMonitor) assignValues(columns []string, values []any) error {
 			} else if value.Valid {
 				_m.Provider = channelmonitor.Provider(value.String)
 			}
+		case channelmonitor.FieldCheckMode:
+			if value, ok := values[i].(*sql.NullString); !ok {
+				return fmt.Errorf("unexpected type %T for field check_mode", values[i])
+			} else if value.Valid {
+				_m.CheckMode = value.String
+			}
+		case channelmonitor.FieldAccountID:
+			if value, ok := values[i].(*sql.NullInt64); !ok {
+				return fmt.Errorf("unexpected type %T for field account_id", values[i])
+			} else if value.Valid {
+				_m.AccountID = new(int64)
+				*_m.AccountID = value.Int64
+			}
 		case channelmonitor.FieldAPIMode:
 			if value, ok := values[i].(*sql.NullString); !ok {
 				return fmt.Errorf("unexpected type %T for field api_mode", values[i])
@@ -326,6 +343,14 @@ func (_m *ChannelMonitor) String() string {
 	builder.WriteString("provider=")
 	builder.WriteString(fmt.Sprintf("%v", _m.Provider))
 	builder.WriteString(", ")
+	builder.WriteString("check_mode=")
+	builder.WriteString(_m.CheckMode)
+	builder.WriteString(", ")
+	if v := _m.AccountID; v != nil {
+		builder.WriteString("account_id=")
+		builder.WriteString(fmt.Sprintf("%v", *v))
+	}
+	builder.WriteString(", ")
 	builder.WriteString("api_mode=")
 	builder.WriteString(_m.APIMode)
 	builder.WriteString(", ")
diff --git a/backend/ent/channelmonitor/channelmonitor.go b/backend/ent/channelmonitor/channelmonitor.go
index 711e6217e2ff..3224c6eb0474 100644
--- a/backend/ent/channelmonitor/channelmonitor.go
+++ b/backend/ent/channelmonitor/channelmonitor.go
@@ -23,6 +23,10 @@ const (
 	FieldName = "name"
 	// FieldProvider holds the string denoting the provider field in the database.
 	FieldProvider = "provider"
+	// FieldCheckMode holds the string denoting the check_mode field in the database.
+	FieldCheckMode = "check_mode"
+	// FieldAccountID holds the string denoting the account_id field in the database.
+	FieldAccountID = "account_id"
 	// FieldAPIMode holds the string denoting the api_mode field in the database.
 	FieldAPIMode = "api_mode"
 	// FieldEndpoint holds the string denoting the endpoint field in the database.
@@ -91,6 +95,8 @@ var Columns = []string{
 	FieldUpdatedAt,
 	FieldName,
 	FieldProvider,
+	FieldCheckMode,
+	FieldAccountID,
 	FieldAPIMode,
 	FieldEndpoint,
 	FieldAPIKeyEncrypted,
@@ -127,6 +133,10 @@ var (
 	UpdateDefaultUpdatedAt func() time.Time
 	// NameValidator is a validator for the "name" field. It is called by the builders before save.
 	NameValidator func(string) error
+	// DefaultCheckMode holds the default value on creation for the "check_mode" field.
+	DefaultCheckMode string
+	// CheckModeValidator is a validator for the "check_mode" field. It is called by the builders before save.
+	CheckModeValidator func(string) error
 	// DefaultAPIMode holds the default value on creation for the "api_mode" field.
 	DefaultAPIMode string
 	// APIModeValidator is a validator for the "api_mode" field. It is called by the builders before save.
@@ -164,10 +174,14 @@ type Provider string
 
 // Provider values.
 const (
-	ProviderOpenai    Provider = "openai"
-	ProviderAnthropic Provider = "anthropic"
-	ProviderGemini    Provider = "gemini"
-	ProviderGrok      Provider = "grok"
+	ProviderOpenai      Provider = "openai"
+	ProviderAnthropic   Provider = "anthropic"
+	ProviderGemini      Provider = "gemini"
+	ProviderGrok        Provider = "grok"
+	ProviderAntigravity Provider = "antigravity"
+	ProviderKimi        Provider = "kimi"
+	ProviderZhipu       Provider = "zhipu"
+	ProviderDeepseek    Provider = "deepseek"
 )
 
 func (pr Provider) String() string {
@@ -177,7 +191,7 @@ func (pr Provider) String() string {
 // ProviderValidator is a validator for the "provider" field enum values. It is called by the builders before save.
 func ProviderValidator(pr Provider) error {
 	switch pr {
-	case ProviderOpenai, ProviderAnthropic, ProviderGemini, ProviderGrok:
+	case ProviderOpenai, ProviderAnthropic, ProviderGemini, ProviderGrok, ProviderAntigravity, ProviderKimi, ProviderZhipu, ProviderDeepseek:
 		return nil
 	default:
 		return fmt.Errorf("channelmonitor: invalid enum value for provider field: %q", pr)
@@ -212,6 +226,16 @@ func ByProvider(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldProvider, opts...).ToFunc()
 }
 
+// ByCheckMode orders the results by the check_mode field.
+func ByCheckMode(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldCheckMode, opts...).ToFunc()
+}
+
+// ByAccountID orders the results by the account_id field.
+func ByAccountID(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldAccountID, opts...).ToFunc()
+}
+
 // ByAPIMode orders the results by the api_mode field.
 func ByAPIMode(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldAPIMode, opts...).ToFunc()
diff --git a/backend/ent/channelmonitor/where.go b/backend/ent/channelmonitor/where.go
index 6575aa5df6ad..9370a0b5ea2e 100644
--- a/backend/ent/channelmonitor/where.go
+++ b/backend/ent/channelmonitor/where.go
@@ -70,6 +70,16 @@ func Name(v string) predicate.ChannelMonitor {
 	return predicate.ChannelMonitor(sql.FieldEQ(FieldName, v))
 }
 
+// CheckMode applies equality check predicate on the "check_mode" field. It's identical to CheckModeEQ.
+func CheckMode(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldEQ(FieldCheckMode, v))
+}
+
+// AccountID applies equality check predicate on the "account_id" field. It's identical to AccountIDEQ.
+func AccountID(v int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldEQ(FieldAccountID, v))
+}
+
 // APIMode applies equality check predicate on the "api_mode" field. It's identical to APIModeEQ.
 func APIMode(v string) predicate.ChannelMonitor {
 	return predicate.ChannelMonitor(sql.FieldEQ(FieldAPIMode, v))
@@ -295,6 +305,121 @@ func ProviderNotIn(vs ...Provider) predicate.ChannelMonitor {
 	return predicate.ChannelMonitor(sql.FieldNotIn(FieldProvider, vs...))
 }
 
+// CheckModeEQ applies the EQ predicate on the "check_mode" field.
+func CheckModeEQ(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldEQ(FieldCheckMode, v))
+}
+
+// CheckModeNEQ applies the NEQ predicate on the "check_mode" field.
+func CheckModeNEQ(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldNEQ(FieldCheckMode, v))
+}
+
+// CheckModeIn applies the In predicate on the "check_mode" field.
+func CheckModeIn(vs ...string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldIn(FieldCheckMode, vs...))
+}
+
+// CheckModeNotIn applies the NotIn predicate on the "check_mode" field.
+func CheckModeNotIn(vs ...string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldNotIn(FieldCheckMode, vs...))
+}
+
+// CheckModeGT applies the GT predicate on the "check_mode" field.
+func CheckModeGT(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldGT(FieldCheckMode, v))
+}
+
+// CheckModeGTE applies the GTE predicate on the "check_mode" field.
+func CheckModeGTE(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldGTE(FieldCheckMode, v))
+}
+
+// CheckModeLT applies the LT predicate on the "check_mode" field.
+func CheckModeLT(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldLT(FieldCheckMode, v))
+}
+
+// CheckModeLTE applies the LTE predicate on the "check_mode" field.
+func CheckModeLTE(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldLTE(FieldCheckMode, v))
+}
+
+// CheckModeContains applies the Contains predicate on the "check_mode" field.
+func CheckModeContains(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldContains(FieldCheckMode, v))
+}
+
+// CheckModeHasPrefix applies the HasPrefix predicate on the "check_mode" field.
+func CheckModeHasPrefix(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldHasPrefix(FieldCheckMode, v))
+}
+
+// CheckModeHasSuffix applies the HasSuffix predicate on the "check_mode" field.
+func CheckModeHasSuffix(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldHasSuffix(FieldCheckMode, v))
+}
+
+// CheckModeEqualFold applies the EqualFold predicate on the "check_mode" field.
+func CheckModeEqualFold(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldEqualFold(FieldCheckMode, v))
+}
+
+// CheckModeContainsFold applies the ContainsFold predicate on the "check_mode" field.
+func CheckModeContainsFold(v string) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldContainsFold(FieldCheckMode, v))
+}
+
+// AccountIDEQ applies the EQ predicate on the "account_id" field.
+func AccountIDEQ(v int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldEQ(FieldAccountID, v))
+}
+
+// AccountIDNEQ applies the NEQ predicate on the "account_id" field.
+func AccountIDNEQ(v int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldNEQ(FieldAccountID, v))
+}
+
+// AccountIDIn applies the In predicate on the "account_id" field.
+func AccountIDIn(vs ...int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldIn(FieldAccountID, vs...))
+}
+
+// AccountIDNotIn applies the NotIn predicate on the "account_id" field.
+func AccountIDNotIn(vs ...int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldNotIn(FieldAccountID, vs...))
+}
+
+// AccountIDGT applies the GT predicate on the "account_id" field.
+func AccountIDGT(v int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldGT(FieldAccountID, v))
+}
+
+// AccountIDGTE applies the GTE predicate on the "account_id" field.
+func AccountIDGTE(v int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldGTE(FieldAccountID, v))
+}
+
+// AccountIDLT applies the LT predicate on the "account_id" field.
+func AccountIDLT(v int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldLT(FieldAccountID, v))
+}
+
+// AccountIDLTE applies the LTE predicate on the "account_id" field.
+func AccountIDLTE(v int64) predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldLTE(FieldAccountID, v))
+}
+
+// AccountIDIsNil applies the IsNil predicate on the "account_id" field.
+func AccountIDIsNil() predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldIsNull(FieldAccountID))
+}
+
+// AccountIDNotNil applies the NotNil predicate on the "account_id" field.
+func AccountIDNotNil() predicate.ChannelMonitor {
+	return predicate.ChannelMonitor(sql.FieldNotNull(FieldAccountID))
+}
+
 // APIModeEQ applies the EQ predicate on the "api_mode" field.
 func APIModeEQ(v string) predicate.ChannelMonitor {
 	return predicate.ChannelMonitor(sql.FieldEQ(FieldAPIMode, v))
diff --git a/backend/ent/channelmonitor_create.go b/backend/ent/channelmonitor_create.go
index 6c7e65491722..cfd63437ff99 100644
--- a/backend/ent/channelmonitor_create.go
+++ b/backend/ent/channelmonitor_create.go
@@ -65,6 +65,34 @@ func (_c *ChannelMonitorCreate) SetProvider(v channelmonitor.Provider) *ChannelM
 	return _c
 }
 
+// SetCheckMode sets the "check_mode" field.
+func (_c *ChannelMonitorCreate) SetCheckMode(v string) *ChannelMonitorCreate {
+	_c.mutation.SetCheckMode(v)
+	return _c
+}
+
+// SetNillableCheckMode sets the "check_mode" field if the given value is not nil.
+func (_c *ChannelMonitorCreate) SetNillableCheckMode(v *string) *ChannelMonitorCreate {
+	if v != nil {
+		_c.SetCheckMode(*v)
+	}
+	return _c
+}
+
+// SetAccountID sets the "account_id" field.
+func (_c *ChannelMonitorCreate) SetAccountID(v int64) *ChannelMonitorCreate {
+	_c.mutation.SetAccountID(v)
+	return _c
+}
+
+// SetNillableAccountID sets the "account_id" field if the given value is not nil.
+func (_c *ChannelMonitorCreate) SetNillableAccountID(v *int64) *ChannelMonitorCreate {
+	if v != nil {
+		_c.SetAccountID(*v)
+	}
+	return _c
+}
+
 // SetAPIMode sets the "api_mode" field.
 func (_c *ChannelMonitorCreate) SetAPIMode(v string) *ChannelMonitorCreate {
 	_c.mutation.SetAPIMode(v)
@@ -303,6 +331,10 @@ func (_c *ChannelMonitorCreate) defaults() {
 		v := channelmonitor.DefaultUpdatedAt()
 		_c.mutation.SetUpdatedAt(v)
 	}
+	if _, ok := _c.mutation.CheckMode(); !ok {
+		v := channelmonitor.DefaultCheckMode
+		_c.mutation.SetCheckMode(v)
+	}
 	if _, ok := _c.mutation.APIMode(); !ok {
 		v := channelmonitor.DefaultAPIMode
 		_c.mutation.SetAPIMode(v)
@@ -357,6 +389,14 @@ func (_c *ChannelMonitorCreate) check() error {
 			return &ValidationError{Name: "provider", err: fmt.Errorf(`ent: validator failed for field "ChannelMonitor.provider": %w`, err)}
 		}
 	}
+	if _, ok := _c.mutation.CheckMode(); !ok {
+		return &ValidationError{Name: "check_mode", err: errors.New(`ent: missing required field "ChannelMonitor.check_mode"`)}
+	}
+	if v, ok := _c.mutation.CheckMode(); ok {
+		if err := channelmonitor.CheckModeValidator(v); err != nil {
+			return &ValidationError{Name: "check_mode", err: fmt.Errorf(`ent: validator failed for field "ChannelMonitor.check_mode": %w`, err)}
+		}
+	}
 	if _, ok := _c.mutation.APIMode(); !ok {
 		return &ValidationError{Name: "api_mode", err: errors.New(`ent: missing required field "ChannelMonitor.api_mode"`)}
 	}
@@ -473,6 +513,14 @@ func (_c *ChannelMonitorCreate) createSpec() (*ChannelMonitor, *sqlgraph.CreateS
 		_spec.SetField(channelmonitor.FieldProvider, field.TypeEnum, value)
 		_node.Provider = value
 	}
+	if value, ok := _c.mutation.CheckMode(); ok {
+		_spec.SetField(channelmonitor.FieldCheckMode, field.TypeString, value)
+		_node.CheckMode = value
+	}
+	if value, ok := _c.mutation.AccountID(); ok {
+		_spec.SetField(channelmonitor.FieldAccountID, field.TypeInt64, value)
+		_node.AccountID = &value
+	}
 	if value, ok := _c.mutation.APIMode(); ok {
 		_spec.SetField(channelmonitor.FieldAPIMode, field.TypeString, value)
 		_node.APIMode = value
@@ -666,6 +714,42 @@ func (u *ChannelMonitorUpsert) UpdateProvider() *ChannelMonitorUpsert {
 	return u
 }
 
+// SetCheckMode sets the "check_mode" field.
+func (u *ChannelMonitorUpsert) SetCheckMode(v string) *ChannelMonitorUpsert {
+	u.Set(channelmonitor.FieldCheckMode, v)
+	return u
+}
+
+// UpdateCheckMode sets the "check_mode" field to the value that was provided on create.
+func (u *ChannelMonitorUpsert) UpdateCheckMode() *ChannelMonitorUpsert {
+	u.SetExcluded(channelmonitor.FieldCheckMode)
+	return u
+}
+
+// SetAccountID sets the "account_id" field.
+func (u *ChannelMonitorUpsert) SetAccountID(v int64) *ChannelMonitorUpsert {
+	u.Set(channelmonitor.FieldAccountID, v)
+	return u
+}
+
+// UpdateAccountID sets the "account_id" field to the value that was provided on create.
+func (u *ChannelMonitorUpsert) UpdateAccountID() *ChannelMonitorUpsert {
+	u.SetExcluded(channelmonitor.FieldAccountID)
+	return u
+}
+
+// AddAccountID adds v to the "account_id" field.
+func (u *ChannelMonitorUpsert) AddAccountID(v int64) *ChannelMonitorUpsert {
+	u.Add(channelmonitor.FieldAccountID, v)
+	return u
+}
+
+// ClearAccountID clears the value of the "account_id" field.
+func (u *ChannelMonitorUpsert) ClearAccountID() *ChannelMonitorUpsert {
+	u.SetNull(channelmonitor.FieldAccountID)
+	return u
+}
+
 // SetAPIMode sets the "api_mode" field.
 func (u *ChannelMonitorUpsert) SetAPIMode(v string) *ChannelMonitorUpsert {
 	u.Set(channelmonitor.FieldAPIMode, v)
@@ -975,6 +1059,48 @@ func (u *ChannelMonitorUpsertOne) UpdateProvider() *ChannelMonitorUpsertOne {
 	})
 }
 
+// SetCheckMode sets the "check_mode" field.
+func (u *ChannelMonitorUpsertOne) SetCheckMode(v string) *ChannelMonitorUpsertOne {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.SetCheckMode(v)
+	})
+}
+
+// UpdateCheckMode sets the "check_mode" field to the value that was provided on create.
+func (u *ChannelMonitorUpsertOne) UpdateCheckMode() *ChannelMonitorUpsertOne {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.UpdateCheckMode()
+	})
+}
+
+// SetAccountID sets the "account_id" field.
+func (u *ChannelMonitorUpsertOne) SetAccountID(v int64) *ChannelMonitorUpsertOne {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.SetAccountID(v)
+	})
+}
+
+// AddAccountID adds v to the "account_id" field.
+func (u *ChannelMonitorUpsertOne) AddAccountID(v int64) *ChannelMonitorUpsertOne {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.AddAccountID(v)
+	})
+}
+
+// UpdateAccountID sets the "account_id" field to the value that was provided on create.
+func (u *ChannelMonitorUpsertOne) UpdateAccountID() *ChannelMonitorUpsertOne {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.UpdateAccountID()
+	})
+}
+
+// ClearAccountID clears the value of the "account_id" field.
+func (u *ChannelMonitorUpsertOne) ClearAccountID() *ChannelMonitorUpsertOne {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.ClearAccountID()
+	})
+}
+
 // SetAPIMode sets the "api_mode" field.
 func (u *ChannelMonitorUpsertOne) SetAPIMode(v string) *ChannelMonitorUpsertOne {
 	return u.Update(func(s *ChannelMonitorUpsert) {
@@ -1487,6 +1613,48 @@ func (u *ChannelMonitorUpsertBulk) UpdateProvider() *ChannelMonitorUpsertBulk {
 	})
 }
 
+// SetCheckMode sets the "check_mode" field.
+func (u *ChannelMonitorUpsertBulk) SetCheckMode(v string) *ChannelMonitorUpsertBulk {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.SetCheckMode(v)
+	})
+}
+
+// UpdateCheckMode sets the "check_mode" field to the value that was provided on create.
+func (u *ChannelMonitorUpsertBulk) UpdateCheckMode() *ChannelMonitorUpsertBulk {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.UpdateCheckMode()
+	})
+}
+
+// SetAccountID sets the "account_id" field.
+func (u *ChannelMonitorUpsertBulk) SetAccountID(v int64) *ChannelMonitorUpsertBulk {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.SetAccountID(v)
+	})
+}
+
+// AddAccountID adds v to the "account_id" field.
+func (u *ChannelMonitorUpsertBulk) AddAccountID(v int64) *ChannelMonitorUpsertBulk {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.AddAccountID(v)
+	})
+}
+
+// UpdateAccountID sets the "account_id" field to the value that was provided on create.
+func (u *ChannelMonitorUpsertBulk) UpdateAccountID() *ChannelMonitorUpsertBulk {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.UpdateAccountID()
+	})
+}
+
+// ClearAccountID clears the value of the "account_id" field.
+func (u *ChannelMonitorUpsertBulk) ClearAccountID() *ChannelMonitorUpsertBulk {
+	return u.Update(func(s *ChannelMonitorUpsert) {
+		s.ClearAccountID()
+	})
+}
+
 // SetAPIMode sets the "api_mode" field.
 func (u *ChannelMonitorUpsertBulk) SetAPIMode(v string) *ChannelMonitorUpsertBulk {
 	return u.Update(func(s *ChannelMonitorUpsert) {
diff --git a/backend/ent/channelmonitor_update.go b/backend/ent/channelmonitor_update.go
index ccc6fa8cc2d8..f5b52e4e1dee 100644
--- a/backend/ent/channelmonitor_update.go
+++ b/backend/ent/channelmonitor_update.go
@@ -66,6 +66,47 @@ func (_u *ChannelMonitorUpdate) SetNillableProvider(v *channelmonitor.Provider)
 	return _u
 }
 
+// SetCheckMode sets the "check_mode" field.
+func (_u *ChannelMonitorUpdate) SetCheckMode(v string) *ChannelMonitorUpdate {
+	_u.mutation.SetCheckMode(v)
+	return _u
+}
+
+// SetNillableCheckMode sets the "check_mode" field if the given value is not nil.
+func (_u *ChannelMonitorUpdate) SetNillableCheckMode(v *string) *ChannelMonitorUpdate {
+	if v != nil {
+		_u.SetCheckMode(*v)
+	}
+	return _u
+}
+
+// SetAccountID sets the "account_id" field.
+func (_u *ChannelMonitorUpdate) SetAccountID(v int64) *ChannelMonitorUpdate {
+	_u.mutation.ResetAccountID()
+	_u.mutation.SetAccountID(v)
+	return _u
+}
+
+// SetNillableAccountID sets the "account_id" field if the given value is not nil.
+func (_u *ChannelMonitorUpdate) SetNillableAccountID(v *int64) *ChannelMonitorUpdate {
+	if v != nil {
+		_u.SetAccountID(*v)
+	}
+	return _u
+}
+
+// AddAccountID adds value to the "account_id" field.
+func (_u *ChannelMonitorUpdate) AddAccountID(v int64) *ChannelMonitorUpdate {
+	_u.mutation.AddAccountID(v)
+	return _u
+}
+
+// ClearAccountID clears the value of the "account_id" field.
+func (_u *ChannelMonitorUpdate) ClearAccountID() *ChannelMonitorUpdate {
+	_u.mutation.ClearAccountID()
+	return _u
+}
+
 // SetAPIMode sets the "api_mode" field.
 func (_u *ChannelMonitorUpdate) SetAPIMode(v string) *ChannelMonitorUpdate {
 	_u.mutation.SetAPIMode(v)
@@ -453,6 +494,11 @@ func (_u *ChannelMonitorUpdate) check() error {
 			return &ValidationError{Name: "provider", err: fmt.Errorf(`ent: validator failed for field "ChannelMonitor.provider": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.CheckMode(); ok {
+		if err := channelmonitor.CheckModeValidator(v); err != nil {
+			return &ValidationError{Name: "check_mode", err: fmt.Errorf(`ent: validator failed for field "ChannelMonitor.check_mode": %w`, err)}
+		}
+	}
 	if v, ok := _u.mutation.APIMode(); ok {
 		if err := channelmonitor.APIModeValidator(v); err != nil {
 			return &ValidationError{Name: "api_mode", err: fmt.Errorf(`ent: validator failed for field "ChannelMonitor.api_mode": %w`, err)}
@@ -517,6 +563,18 @@ func (_u *ChannelMonitorUpdate) sqlSave(ctx context.Context) (_node int, err err
 	if value, ok := _u.mutation.Provider(); ok {
 		_spec.SetField(channelmonitor.FieldProvider, field.TypeEnum, value)
 	}
+	if value, ok := _u.mutation.CheckMode(); ok {
+		_spec.SetField(channelmonitor.FieldCheckMode, field.TypeString, value)
+	}
+	if value, ok := _u.mutation.AccountID(); ok {
+		_spec.SetField(channelmonitor.FieldAccountID, field.TypeInt64, value)
+	}
+	if value, ok := _u.mutation.AddedAccountID(); ok {
+		_spec.AddField(channelmonitor.FieldAccountID, field.TypeInt64, value)
+	}
+	if _u.mutation.AccountIDCleared() {
+		_spec.ClearField(channelmonitor.FieldAccountID, field.TypeInt64)
+	}
 	if value, ok := _u.mutation.APIMode(); ok {
 		_spec.SetField(channelmonitor.FieldAPIMode, field.TypeString, value)
 	}
@@ -755,6 +813,47 @@ func (_u *ChannelMonitorUpdateOne) SetNillableProvider(v *channelmonitor.Provide
 	return _u
 }
 
+// SetCheckMode sets the "check_mode" field.
+func (_u *ChannelMonitorUpdateOne) SetCheckMode(v string) *ChannelMonitorUpdateOne {
+	_u.mutation.SetCheckMode(v)
+	return _u
+}
+
+// SetNillableCheckMode sets the "check_mode" field if the given value is not nil.
+func (_u *ChannelMonitorUpdateOne) SetNillableCheckMode(v *string) *ChannelMonitorUpdateOne {
+	if v != nil {
+		_u.SetCheckMode(*v)
+	}
+	return _u
+}
+
+// SetAccountID sets the "account_id" field.
+func (_u *ChannelMonitorUpdateOne) SetAccountID(v int64) *ChannelMonitorUpdateOne {
+	_u.mutation.ResetAccountID()
+	_u.mutation.SetAccountID(v)
+	return _u
+}
+
+// SetNillableAccountID sets the "account_id" field if the given value is not nil.
+func (_u *ChannelMonitorUpdateOne) SetNillableAccountID(v *int64) *ChannelMonitorUpdateOne {
+	if v != nil {
+		_u.SetAccountID(*v)
+	}
+	return _u
+}
+
+// AddAccountID adds value to the "account_id" field.
+func (_u *ChannelMonitorUpdateOne) AddAccountID(v int64) *ChannelMonitorUpdateOne {
+	_u.mutation.AddAccountID(v)
+	return _u
+}
+
+// ClearAccountID clears the value of the "account_id" field.
+func (_u *ChannelMonitorUpdateOne) ClearAccountID() *ChannelMonitorUpdateOne {
+	_u.mutation.ClearAccountID()
+	return _u
+}
+
 // SetAPIMode sets the "api_mode" field.
 func (_u *ChannelMonitorUpdateOne) SetAPIMode(v string) *ChannelMonitorUpdateOne {
 	_u.mutation.SetAPIMode(v)
@@ -1155,6 +1254,11 @@ func (_u *ChannelMonitorUpdateOne) check() error {
 			return &ValidationError{Name: "provider", err: fmt.Errorf(`ent: validator failed for field "ChannelMonitor.provider": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.CheckMode(); ok {
+		if err := channelmonitor.CheckModeValidator(v); err != nil {
+			return &ValidationError{Name: "check_mode", err: fmt.Errorf(`ent: validator failed for field "ChannelMonitor.check_mode": %w`, err)}
+		}
+	}
 	if v, ok := _u.mutation.APIMode(); ok {
 		if err := channelmonitor.APIModeValidator(v); err != nil {
 			return &ValidationError{Name: "api_mode", err: fmt.Errorf(`ent: validator failed for field "ChannelMonitor.api_mode": %w`, err)}
@@ -1236,6 +1340,18 @@ func (_u *ChannelMonitorUpdateOne) sqlSave(ctx context.Context) (_node *ChannelM
 	if value, ok := _u.mutation.Provider(); ok {
 		_spec.SetField(channelmonitor.FieldProvider, field.TypeEnum, value)
 	}
+	if value, ok := _u.mutation.CheckMode(); ok {
+		_spec.SetField(channelmonitor.FieldCheckMode, field.TypeString, value)
+	}
+	if value, ok := _u.mutation.AccountID(); ok {
+		_spec.SetField(channelmonitor.FieldAccountID, field.TypeInt64, value)
+	}
+	if value, ok := _u.mutation.AddedAccountID(); ok {
+		_spec.AddField(channelmonitor.FieldAccountID, field.TypeInt64, value)
+	}
+	if _u.mutation.AccountIDCleared() {
+		_spec.ClearField(channelmonitor.FieldAccountID, field.TypeInt64)
+	}
 	if value, ok := _u.mutation.APIMode(); ok {
 		_spec.SetField(channelmonitor.FieldAPIMode, field.TypeString, value)
 	}
diff --git a/backend/ent/channelmonitorhistory.go b/backend/ent/channelmonitorhistory.go
index 70dde5422a7d..fef5109528c8 100644
--- a/backend/ent/channelmonitorhistory.go
+++ b/backend/ent/channelmonitorhistory.go
@@ -3,6 +3,7 @@
 package ent
 
 import (
+	"encoding/json"
 	"fmt"
 	"strings"
 	"time"
@@ -11,6 +12,7 @@ import (
 	"entgo.io/ent/dialect/sql"
 	"github.com/Wei-Shaw/sub2api/ent/channelmonitor"
 	"github.com/Wei-Shaw/sub2api/ent/channelmonitorhistory"
+	"github.com/Wei-Shaw/sub2api/internal/domain"
 )
 
 // ChannelMonitorHistory is the model entity for the ChannelMonitorHistory schema.
@@ -30,6 +32,8 @@ type ChannelMonitorHistory struct {
 	PingLatencyMs *int `json:"ping_latency_ms,omitempty"`
 	// Message holds the value of the "message" field.
 	Message string `json:"message,omitempty"`
+	// Quota holds the value of the "quota" field.
+	Quota *domain.MonitorQuotaSnapshot `json:"quota,omitempty"`
 	// CheckedAt holds the value of the "checked_at" field.
 	CheckedAt time.Time `json:"checked_at,omitempty"`
 	// Edges holds the relations/edges for other nodes in the graph.
@@ -63,6 +67,8 @@ func (*ChannelMonitorHistory) scanValues(columns []string) ([]any, error) {
 	values := make([]any, len(columns))
 	for i := range columns {
 		switch columns[i] {
+		case channelmonitorhistory.FieldQuota:
+			values[i] = new([]byte)
 		case channelmonitorhistory.FieldID, channelmonitorhistory.FieldMonitorID, channelmonitorhistory.FieldLatencyMs, channelmonitorhistory.FieldPingLatencyMs:
 			values[i] = new(sql.NullInt64)
 		case channelmonitorhistory.FieldModel, channelmonitorhistory.FieldStatus, channelmonitorhistory.FieldMessage:
@@ -128,6 +134,14 @@ func (_m *ChannelMonitorHistory) assignValues(columns []string, values []any) er
 			} else if value.Valid {
 				_m.Message = value.String
 			}
+		case channelmonitorhistory.FieldQuota:
+			if value, ok := values[i].(*[]byte); !ok {
+				return fmt.Errorf("unexpected type %T for field quota", values[i])
+			} else if value != nil && len(*value) > 0 {
+				if err := json.Unmarshal(*value, &_m.Quota); err != nil {
+					return fmt.Errorf("unmarshal field quota: %w", err)
+				}
+			}
 		case channelmonitorhistory.FieldCheckedAt:
 			if value, ok := values[i].(*sql.NullTime); !ok {
 				return fmt.Errorf("unexpected type %T for field checked_at", values[i])
@@ -197,6 +211,9 @@ func (_m *ChannelMonitorHistory) String() string {
 	builder.WriteString("message=")
 	builder.WriteString(_m.Message)
 	builder.WriteString(", ")
+	builder.WriteString("quota=")
+	builder.WriteString(fmt.Sprintf("%v", _m.Quota))
+	builder.WriteString(", ")
 	builder.WriteString("checked_at=")
 	builder.WriteString(_m.CheckedAt.Format(time.ANSIC))
 	builder.WriteByte(')')
diff --git a/backend/ent/channelmonitorhistory/channelmonitorhistory.go b/backend/ent/channelmonitorhistory/channelmonitorhistory.go
index 6a9dc006703b..ceb11cee67d4 100644
--- a/backend/ent/channelmonitorhistory/channelmonitorhistory.go
+++ b/backend/ent/channelmonitorhistory/channelmonitorhistory.go
@@ -27,6 +27,8 @@ const (
 	FieldPingLatencyMs = "ping_latency_ms"
 	// FieldMessage holds the string denoting the message field in the database.
 	FieldMessage = "message"
+	// FieldQuota holds the string denoting the quota field in the database.
+	FieldQuota = "quota"
 	// FieldCheckedAt holds the string denoting the checked_at field in the database.
 	FieldCheckedAt = "checked_at"
 	// EdgeMonitor holds the string denoting the monitor edge name in mutations.
@@ -51,6 +53,7 @@ var Columns = []string{
 	FieldLatencyMs,
 	FieldPingLatencyMs,
 	FieldMessage,
+	FieldQuota,
 	FieldCheckedAt,
 }
 
diff --git a/backend/ent/channelmonitorhistory/where.go b/backend/ent/channelmonitorhistory/where.go
index afa73f35c875..f7afd7c319a4 100644
--- a/backend/ent/channelmonitorhistory/where.go
+++ b/backend/ent/channelmonitorhistory/where.go
@@ -365,6 +365,16 @@ func MessageContainsFold(v string) predicate.ChannelMonitorHistory {
 	return predicate.ChannelMonitorHistory(sql.FieldContainsFold(FieldMessage, v))
 }
 
+// QuotaIsNil applies the IsNil predicate on the "quota" field.
+func QuotaIsNil() predicate.ChannelMonitorHistory {
+	return predicate.ChannelMonitorHistory(sql.FieldIsNull(FieldQuota))
+}
+
+// QuotaNotNil applies the NotNil predicate on the "quota" field.
+func QuotaNotNil() predicate.ChannelMonitorHistory {
+	return predicate.ChannelMonitorHistory(sql.FieldNotNull(FieldQuota))
+}
+
 // CheckedAtEQ applies the EQ predicate on the "checked_at" field.
 func CheckedAtEQ(v time.Time) predicate.ChannelMonitorHistory {
 	return predicate.ChannelMonitorHistory(sql.FieldEQ(FieldCheckedAt, v))
diff --git a/backend/ent/channelmonitorhistory_create.go b/backend/ent/channelmonitorhistory_create.go
index 71034865c970..4d4222d6d355 100644
--- a/backend/ent/channelmonitorhistory_create.go
+++ b/backend/ent/channelmonitorhistory_create.go
@@ -13,6 +13,7 @@ import (
 	"entgo.io/ent/schema/field"
 	"github.com/Wei-Shaw/sub2api/ent/channelmonitor"
 	"github.com/Wei-Shaw/sub2api/ent/channelmonitorhistory"
+	"github.com/Wei-Shaw/sub2api/internal/domain"
 )
 
 // ChannelMonitorHistoryCreate is the builder for creating a ChannelMonitorHistory entity.
@@ -83,6 +84,12 @@ func (_c *ChannelMonitorHistoryCreate) SetNillableMessage(v *string) *ChannelMon
 	return _c
 }
 
+// SetQuota sets the "quota" field.
+func (_c *ChannelMonitorHistoryCreate) SetQuota(v *domain.MonitorQuotaSnapshot) *ChannelMonitorHistoryCreate {
+	_c.mutation.SetQuota(v)
+	return _c
+}
+
 // SetCheckedAt sets the "checked_at" field.
 func (_c *ChannelMonitorHistoryCreate) SetCheckedAt(v time.Time) *ChannelMonitorHistoryCreate {
 	_c.mutation.SetCheckedAt(v)
@@ -226,6 +233,10 @@ func (_c *ChannelMonitorHistoryCreate) createSpec() (*ChannelMonitorHistory, *sq
 		_spec.SetField(channelmonitorhistory.FieldMessage, field.TypeString, value)
 		_node.Message = value
 	}
+	if value, ok := _c.mutation.Quota(); ok {
+		_spec.SetField(channelmonitorhistory.FieldQuota, field.TypeJSON, value)
+		_node.Quota = value
+	}
 	if value, ok := _c.mutation.CheckedAt(); ok {
 		_spec.SetField(channelmonitorhistory.FieldCheckedAt, field.TypeTime, value)
 		_node.CheckedAt = value
@@ -401,6 +412,24 @@ func (u *ChannelMonitorHistoryUpsert) ClearMessage() *ChannelMonitorHistoryUpser
 	return u
 }
 
+// SetQuota sets the "quota" field.
+func (u *ChannelMonitorHistoryUpsert) SetQuota(v *domain.MonitorQuotaSnapshot) *ChannelMonitorHistoryUpsert {
+	u.Set(channelmonitorhistory.FieldQuota, v)
+	return u
+}
+
+// UpdateQuota sets the "quota" field to the value that was provided on create.
+func (u *ChannelMonitorHistoryUpsert) UpdateQuota() *ChannelMonitorHistoryUpsert {
+	u.SetExcluded(channelmonitorhistory.FieldQuota)
+	return u
+}
+
+// ClearQuota clears the value of the "quota" field.
+func (u *ChannelMonitorHistoryUpsert) ClearQuota() *ChannelMonitorHistoryUpsert {
+	u.SetNull(channelmonitorhistory.FieldQuota)
+	return u
+}
+
 // SetCheckedAt sets the "checked_at" field.
 func (u *ChannelMonitorHistoryUpsert) SetCheckedAt(v time.Time) *ChannelMonitorHistoryUpsert {
 	u.Set(channelmonitorhistory.FieldCheckedAt, v)
@@ -572,6 +601,27 @@ func (u *ChannelMonitorHistoryUpsertOne) ClearMessage() *ChannelMonitorHistoryUp
 	})
 }
 
+// SetQuota sets the "quota" field.
+func (u *ChannelMonitorHistoryUpsertOne) SetQuota(v *domain.MonitorQuotaSnapshot) *ChannelMonitorHistoryUpsertOne {
+	return u.Update(func(s *ChannelMonitorHistoryUpsert) {
+		s.SetQuota(v)
+	})
+}
+
+// UpdateQuota sets the "quota" field to the value that was provided on create.
+func (u *ChannelMonitorHistoryUpsertOne) UpdateQuota() *ChannelMonitorHistoryUpsertOne {
+	return u.Update(func(s *ChannelMonitorHistoryUpsert) {
+		s.UpdateQuota()
+	})
+}
+
+// ClearQuota clears the value of the "quota" field.
+func (u *ChannelMonitorHistoryUpsertOne) ClearQuota() *ChannelMonitorHistoryUpsertOne {
+	return u.Update(func(s *ChannelMonitorHistoryUpsert) {
+		s.ClearQuota()
+	})
+}
+
 // SetCheckedAt sets the "checked_at" field.
 func (u *ChannelMonitorHistoryUpsertOne) SetCheckedAt(v time.Time) *ChannelMonitorHistoryUpsertOne {
 	return u.Update(func(s *ChannelMonitorHistoryUpsert) {
@@ -909,6 +959,27 @@ func (u *ChannelMonitorHistoryUpsertBulk) ClearMessage() *ChannelMonitorHistoryU
 	})
 }
 
+// SetQuota sets the "quota" field.
+func (u *ChannelMonitorHistoryUpsertBulk) SetQuota(v *domain.MonitorQuotaSnapshot) *ChannelMonitorHistoryUpsertBulk {
+	return u.Update(func(s *ChannelMonitorHistoryUpsert) {
+		s.SetQuota(v)
+	})
+}
+
+// UpdateQuota sets the "quota" field to the value that was provided on create.
+func (u *ChannelMonitorHistoryUpsertBulk) UpdateQuota() *ChannelMonitorHistoryUpsertBulk {
+	return u.Update(func(s *ChannelMonitorHistoryUpsert) {
+		s.UpdateQuota()
+	})
+}
+
+// ClearQuota clears the value of the "quota" field.
+func (u *ChannelMonitorHistoryUpsertBulk) ClearQuota() *ChannelMonitorHistoryUpsertBulk {
+	return u.Update(func(s *ChannelMonitorHistoryUpsert) {
+		s.ClearQuota()
+	})
+}
+
 // SetCheckedAt sets the "checked_at" field.
 func (u *ChannelMonitorHistoryUpsertBulk) SetCheckedAt(v time.Time) *ChannelMonitorHistoryUpsertBulk {
 	return u.Update(func(s *ChannelMonitorHistoryUpsert) {
diff --git a/backend/ent/channelmonitorhistory_update.go b/backend/ent/channelmonitorhistory_update.go
index a85a8072a9e6..58d7214068ae 100644
--- a/backend/ent/channelmonitorhistory_update.go
+++ b/backend/ent/channelmonitorhistory_update.go
@@ -14,6 +14,7 @@ import (
 	"github.com/Wei-Shaw/sub2api/ent/channelmonitor"
 	"github.com/Wei-Shaw/sub2api/ent/channelmonitorhistory"
 	"github.com/Wei-Shaw/sub2api/ent/predicate"
+	"github.com/Wei-Shaw/sub2api/internal/domain"
 )
 
 // ChannelMonitorHistoryUpdate is the builder for updating ChannelMonitorHistory entities.
@@ -145,6 +146,18 @@ func (_u *ChannelMonitorHistoryUpdate) ClearMessage() *ChannelMonitorHistoryUpda
 	return _u
 }
 
+// SetQuota sets the "quota" field.
+func (_u *ChannelMonitorHistoryUpdate) SetQuota(v *domain.MonitorQuotaSnapshot) *ChannelMonitorHistoryUpdate {
+	_u.mutation.SetQuota(v)
+	return _u
+}
+
+// ClearQuota clears the value of the "quota" field.
+func (_u *ChannelMonitorHistoryUpdate) ClearQuota() *ChannelMonitorHistoryUpdate {
+	_u.mutation.ClearQuota()
+	return _u
+}
+
 // SetCheckedAt sets the "checked_at" field.
 func (_u *ChannelMonitorHistoryUpdate) SetCheckedAt(v time.Time) *ChannelMonitorHistoryUpdate {
 	_u.mutation.SetCheckedAt(v)
@@ -267,6 +280,12 @@ func (_u *ChannelMonitorHistoryUpdate) sqlSave(ctx context.Context) (_node int,
 	if _u.mutation.MessageCleared() {
 		_spec.ClearField(channelmonitorhistory.FieldMessage, field.TypeString)
 	}
+	if value, ok := _u.mutation.Quota(); ok {
+		_spec.SetField(channelmonitorhistory.FieldQuota, field.TypeJSON, value)
+	}
+	if _u.mutation.QuotaCleared() {
+		_spec.ClearField(channelmonitorhistory.FieldQuota, field.TypeJSON)
+	}
 	if value, ok := _u.mutation.CheckedAt(); ok {
 		_spec.SetField(channelmonitorhistory.FieldCheckedAt, field.TypeTime, value)
 	}
@@ -435,6 +454,18 @@ func (_u *ChannelMonitorHistoryUpdateOne) ClearMessage() *ChannelMonitorHistoryU
 	return _u
 }
 
+// SetQuota sets the "quota" field.
+func (_u *ChannelMonitorHistoryUpdateOne) SetQuota(v *domain.MonitorQuotaSnapshot) *ChannelMonitorHistoryUpdateOne {
+	_u.mutation.SetQuota(v)
+	return _u
+}
+
+// ClearQuota clears the value of the "quota" field.
+func (_u *ChannelMonitorHistoryUpdateOne) ClearQuota() *ChannelMonitorHistoryUpdateOne {
+	_u.mutation.ClearQuota()
+	return _u
+}
+
 // SetCheckedAt sets the "checked_at" field.
 func (_u *ChannelMonitorHistoryUpdateOne) SetCheckedAt(v time.Time) *ChannelMonitorHistoryUpdateOne {
 	_u.mutation.SetCheckedAt(v)
@@ -587,6 +618,12 @@ func (_u *ChannelMonitorHistoryUpdateOne) sqlSave(ctx context.Context) (_node *C
 	if _u.mutation.MessageCleared() {
 		_spec.ClearField(channelmonitorhistory.FieldMessage, field.TypeString)
 	}
+	if value, ok := _u.mutation.Quota(); ok {
+		_spec.SetField(channelmonitorhistory.FieldQuota, field.TypeJSON, value)
+	}
+	if _u.mutation.QuotaCleared() {
+		_spec.ClearField(channelmonitorhistory.FieldQuota, field.TypeJSON)
+	}
 	if value, ok := _u.mutation.CheckedAt(); ok {
 		_spec.SetField(channelmonitorhistory.FieldCheckedAt, field.TypeTime, value)
 	}
diff --git a/backend/ent/channelmonitorrequesttemplate/channelmonitorrequesttemplate.go b/backend/ent/channelmonitorrequesttemplate/channelmonitorrequesttemplate.go
index 5989d0e74386..f0e9dbf16e16 100644
--- a/backend/ent/channelmonitorrequesttemplate/channelmonitorrequesttemplate.go
+++ b/backend/ent/channelmonitorrequesttemplate/channelmonitorrequesttemplate.go
@@ -100,10 +100,14 @@ type Provider string
 
 // Provider values.
 const (
-	ProviderOpenai    Provider = "openai"
-	ProviderAnthropic Provider = "anthropic"
-	ProviderGemini    Provider = "gemini"
-	ProviderGrok      Provider = "grok"
+	ProviderOpenai      Provider = "openai"
+	ProviderAnthropic   Provider = "anthropic"
+	ProviderGemini      Provider = "gemini"
+	ProviderGrok        Provider = "grok"
+	ProviderAntigravity Provider = "antigravity"
+	ProviderKimi        Provider = "kimi"
+	ProviderZhipu       Provider = "zhipu"
+	ProviderDeepseek    Provider = "deepseek"
 )
 
 func (pr Provider) String() string {
@@ -113,7 +117,7 @@ func (pr Provider) String() string {
 // ProviderValidator is a validator for the "provider" field enum values. It is called by the builders before save.
 func ProviderValidator(pr Provider) error {
 	switch pr {
-	case ProviderOpenai, ProviderAnthropic, ProviderGemini, ProviderGrok:
+	case ProviderOpenai, ProviderAnthropic, ProviderGemini, ProviderGrok, ProviderAntigravity, ProviderKimi, ProviderZhipu, ProviderDeepseek:
 		return nil
 	default:
 		return fmt.Errorf("channelmonitorrequesttemplate: invalid enum value for provider field: %q", pr)
diff --git a/backend/ent/group.go b/backend/ent/group.go
index 110f06742a38..c9795fb58f04 100644
--- a/backend/ent/group.go
+++ b/backend/ent/group.go
@@ -97,7 +97,7 @@ type Group struct {
 	AudioTtsPricePerMillionChars *float64 `json:"audio_tts_price_per_million_chars,omitempty"`
 	// STT 每小时价格（USD）
 	AudioSttPricePerHour *float64 `json:"audio_stt_price_per_hour,omitempty"`
-	// 是否按上下文长度应用模型阶梯价格
+	// 是否按上下文长度应用模型阶梯价格；默认开启以保持官方/渠道长上下文价
 	LongContextPricingEnabled bool `json:"long_context_pricing_enabled,omitempty"`
 	// 分组逐模型定价；优先级高于渠道和内置定价
 	ModelPricing json.RawMessage `json:"model_pricing,omitempty"`
diff --git a/backend/ent/migrate/schema.go b/backend/ent/migrate/schema.go
index 316870e5a55b..f667c3ffbf0e 100644
--- a/backend/ent/migrate/schema.go
+++ b/backend/ent/migrate/schema.go
@@ -623,7 +623,9 @@ var (
 		{Name: "created_at", Type: field.TypeTime, SchemaType: map[string]string{"postgres": "timestamptz"}},
 		{Name: "updated_at", Type: field.TypeTime, SchemaType: map[string]string{"postgres": "timestamptz"}},
 		{Name: "name", Type: field.TypeString, Size: 100},
-		{Name: "provider", Type: field.TypeEnum, Enums: []string{"openai", "anthropic", "gemini", "grok"}},
+		{Name: "provider", Type: field.TypeEnum, Enums: []string{"openai", "anthropic", "gemini", "grok", "antigravity", "kimi", "zhipu", "deepseek"}},
+		{Name: "check_mode", Type: field.TypeString, Size: 32, Default: "probe"},
+		{Name: "account_id", Type: field.TypeInt64, Nullable: true},
 		{Name: "api_mode", Type: field.TypeString, Size: 32, Default: "chat_completions"},
 		{Name: "endpoint", Type: field.TypeString, Size: 500},
 		{Name: "api_key_encrypted", Type: field.TypeString},
@@ -648,7 +650,7 @@ var (
 		ForeignKeys: []*schema.ForeignKey{
 			{
 				Symbol:     "channel_monitors_channel_monitor_request_templates_request_template",
-				Columns:    []*schema.Column{ChannelMonitorsColumns[19]},
+				Columns:    []*schema.Column{ChannelMonitorsColumns[21]},
 				RefColumns: []*schema.Column{ChannelMonitorRequestTemplatesColumns[0]},
 				OnDelete:   schema.SetNull,
 			},
@@ -657,7 +659,7 @@ var (
 			{
 				Name:    "channelmonitor_enabled_last_checked_at",
 				Unique:  false,
-				Columns: []*schema.Column{ChannelMonitorsColumns[11], ChannelMonitorsColumns[14]},
+				Columns: []*schema.Column{ChannelMonitorsColumns[13], ChannelMonitorsColumns[16]},
 			},
 			{
 				Name:    "channelmonitor_provider",
@@ -667,17 +669,22 @@ var (
 			{
 				Name:    "channelmonitor_provider_api_mode",
 				Unique:  false,
-				Columns: []*schema.Column{ChannelMonitorsColumns[4], ChannelMonitorsColumns[5]},
+				Columns: []*schema.Column{ChannelMonitorsColumns[4], ChannelMonitorsColumns[7]},
 			},
 			{
 				Name:    "channelmonitor_group_name",
 				Unique:  false,
-				Columns: []*schema.Column{ChannelMonitorsColumns[10]},
+				Columns: []*schema.Column{ChannelMonitorsColumns[12]},
 			},
 			{
 				Name:    "channelmonitor_template_id",
 				Unique:  false,
-				Columns: []*schema.Column{ChannelMonitorsColumns[19]},
+				Columns: []*schema.Column{ChannelMonitorsColumns[21]},
+			},
+			{
+				Name:    "channelmonitor_account_id",
+				Unique:  false,
+				Columns: []*schema.Column{ChannelMonitorsColumns[6]},
 			},
 		},
 	}
@@ -733,6 +740,7 @@ var (
 		{Name: "latency_ms", Type: field.TypeInt, Nullable: true},
 		{Name: "ping_latency_ms", Type: field.TypeInt, Nullable: true},
 		{Name: "message", Type: field.TypeString, Nullable: true, Size: 500, Default: ""},
+		{Name: "quota", Type: field.TypeJSON, Nullable: true},
 		{Name: "checked_at", Type: field.TypeTime},
 		{Name: "monitor_id", Type: field.TypeInt64},
 	}
@@ -744,7 +752,7 @@ var (
 		ForeignKeys: []*schema.ForeignKey{
 			{
 				Symbol:     "channel_monitor_histories_channel_monitors_history",
-				Columns:    []*schema.Column{ChannelMonitorHistoriesColumns[7]},
+				Columns:    []*schema.Column{ChannelMonitorHistoriesColumns[8]},
 				RefColumns: []*schema.Column{ChannelMonitorsColumns[0]},
 				OnDelete:   schema.Cascade,
 			},
@@ -753,12 +761,12 @@ var (
 			{
 				Name:    "channelmonitorhistory_monitor_id_model_checked_at",
 				Unique:  false,
-				Columns: []*schema.Column{ChannelMonitorHistoriesColumns[7], ChannelMonitorHistoriesColumns[1], ChannelMonitorHistoriesColumns[6]},
+				Columns: []*schema.Column{ChannelMonitorHistoriesColumns[8], ChannelMonitorHistoriesColumns[1], ChannelMonitorHistoriesColumns[7]},
 			},
 			{
 				Name:    "channelmonitorhistory_checked_at",
 				Unique:  false,
-				Columns: []*schema.Column{ChannelMonitorHistoriesColumns[6]},
+				Columns: []*schema.Column{ChannelMonitorHistoriesColumns[7]},
 			},
 		},
 	}
@@ -768,7 +776,7 @@ var (
 		{Name: "created_at", Type: field.TypeTime, SchemaType: map[string]string{"postgres": "timestamptz"}},
 		{Name: "updated_at", Type: field.TypeTime, SchemaType: map[string]string{"postgres": "timestamptz"}},
 		{Name: "name", Type: field.TypeString, Size: 100},
-		{Name: "provider", Type: field.TypeEnum, Enums: []string{"openai", "anthropic", "gemini", "grok"}},
+		{Name: "provider", Type: field.TypeEnum, Enums: []string{"openai", "anthropic", "gemini", "grok", "antigravity", "kimi", "zhipu", "deepseek"}},
 		{Name: "api_mode", Type: field.TypeString, Size: 32, Default: "chat_completions"},
 		{Name: "description", Type: field.TypeString, Nullable: true, Size: 500, Default: ""},
 		{Name: "extra_headers", Type: field.TypeJSON},
diff --git a/backend/ent/mutation.go b/backend/ent/mutation.go
index da0500d9be6e..ddf50a1c4263 100644
--- a/backend/ent/mutation.go
+++ b/backend/ent/mutation.go
@@ -14608,6 +14608,9 @@ type ChannelMonitorMutation struct {
 	updated_at              *time.Time
 	name                    *string
 	provider                *channelmonitor.Provider
+	check_mode              *string
+	account_id              *int64
+	addaccount_id           *int64
 	api_mode                *string
 	endpoint                *string
 	api_key_encrypted       *string
@@ -14882,6 +14885,112 @@ func (m *ChannelMonitorMutation) ResetProvider() {
 	m.provider = nil
 }
 
+// SetCheckMode sets the "check_mode" field.
+func (m *ChannelMonitorMutation) SetCheckMode(s string) {
+	m.check_mode = &s
+}
+
+// CheckMode returns the value of the "check_mode" field in the mutation.
+func (m *ChannelMonitorMutation) CheckMode() (r string, exists bool) {
+	v := m.check_mode
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldCheckMode returns the old "check_mode" field's value of the ChannelMonitor entity.
+// If the ChannelMonitor object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *ChannelMonitorMutation) OldCheckMode(ctx context.Context) (v string, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldCheckMode is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldCheckMode requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldCheckMode: %w", err)
+	}
+	return oldValue.CheckMode, nil
+}
+
+// ResetCheckMode resets all changes to the "check_mode" field.
+func (m *ChannelMonitorMutation) ResetCheckMode() {
+	m.check_mode = nil
+}
+
+// SetAccountID sets the "account_id" field.
+func (m *ChannelMonitorMutation) SetAccountID(i int64) {
+	m.account_id = &i
+	m.addaccount_id = nil
+}
+
+// AccountID returns the value of the "account_id" field in the mutation.
+func (m *ChannelMonitorMutation) AccountID() (r int64, exists bool) {
+	v := m.account_id
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldAccountID returns the old "account_id" field's value of the ChannelMonitor entity.
+// If the ChannelMonitor object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *ChannelMonitorMutation) OldAccountID(ctx context.Context) (v *int64, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldAccountID is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldAccountID requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldAccountID: %w", err)
+	}
+	return oldValue.AccountID, nil
+}
+
+// AddAccountID adds i to the "account_id" field.
+func (m *ChannelMonitorMutation) AddAccountID(i int64) {
+	if m.addaccount_id != nil {
+		*m.addaccount_id += i
+	} else {
+		m.addaccount_id = &i
+	}
+}
+
+// AddedAccountID returns the value that was added to the "account_id" field in this mutation.
+func (m *ChannelMonitorMutation) AddedAccountID() (r int64, exists bool) {
+	v := m.addaccount_id
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// ClearAccountID clears the value of the "account_id" field.
+func (m *ChannelMonitorMutation) ClearAccountID() {
+	m.account_id = nil
+	m.addaccount_id = nil
+	m.clearedFields[channelmonitor.FieldAccountID] = struct{}{}
+}
+
+// AccountIDCleared returns if the "account_id" field was cleared in this mutation.
+func (m *ChannelMonitorMutation) AccountIDCleared() bool {
+	_, ok := m.clearedFields[channelmonitor.FieldAccountID]
+	return ok
+}
+
+// ResetAccountID resets all changes to the "account_id" field.
+func (m *ChannelMonitorMutation) ResetAccountID() {
+	m.account_id = nil
+	m.addaccount_id = nil
+	delete(m.clearedFields, channelmonitor.FieldAccountID)
+}
+
 // SetAPIMode sets the "api_mode" field.
 func (m *ChannelMonitorMutation) SetAPIMode(s string) {
 	m.api_mode = &s
@@ -15731,7 +15840,7 @@ func (m *ChannelMonitorMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *ChannelMonitorMutation) Fields() []string {
-	fields := make([]string, 0, 19)
+	fields := make([]string, 0, 21)
 	if m.created_at != nil {
 		fields = append(fields, channelmonitor.FieldCreatedAt)
 	}
@@ -15744,6 +15853,12 @@ func (m *ChannelMonitorMutation) Fields() []string {
 	if m.provider != nil {
 		fields = append(fields, channelmonitor.FieldProvider)
 	}
+	if m.check_mode != nil {
+		fields = append(fields, channelmonitor.FieldCheckMode)
+	}
+	if m.account_id != nil {
+		fields = append(fields, channelmonitor.FieldAccountID)
+	}
 	if m.api_mode != nil {
 		fields = append(fields, channelmonitor.FieldAPIMode)
 	}
@@ -15805,6 +15920,10 @@ func (m *ChannelMonitorMutation) Field(name string) (ent.Value, bool) {
 		return m.Name()
 	case channelmonitor.FieldProvider:
 		return m.Provider()
+	case channelmonitor.FieldCheckMode:
+		return m.CheckMode()
+	case channelmonitor.FieldAccountID:
+		return m.AccountID()
 	case channelmonitor.FieldAPIMode:
 		return m.APIMode()
 	case channelmonitor.FieldEndpoint:
@@ -15852,6 +15971,10 @@ func (m *ChannelMonitorMutation) OldField(ctx context.Context, name string) (ent
 		return m.OldName(ctx)
 	case channelmonitor.FieldProvider:
 		return m.OldProvider(ctx)
+	case channelmonitor.FieldCheckMode:
+		return m.OldCheckMode(ctx)
+	case channelmonitor.FieldAccountID:
+		return m.OldAccountID(ctx)
 	case channelmonitor.FieldAPIMode:
 		return m.OldAPIMode(ctx)
 	case channelmonitor.FieldEndpoint:
@@ -15919,6 +16042,20 @@ func (m *ChannelMonitorMutation) SetField(name string, value ent.Value) error {
 		}
 		m.SetProvider(v)
 		return nil
+	case channelmonitor.FieldCheckMode:
+		v, ok := value.(string)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetCheckMode(v)
+		return nil
+	case channelmonitor.FieldAccountID:
+		v, ok := value.(int64)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetAccountID(v)
+		return nil
 	case channelmonitor.FieldAPIMode:
 		v, ok := value.(string)
 		if !ok {
@@ -16032,6 +16169,9 @@ func (m *ChannelMonitorMutation) SetField(name string, value ent.Value) error {
 // this mutation.
 func (m *ChannelMonitorMutation) AddedFields() []string {
 	var fields []string
+	if m.addaccount_id != nil {
+		fields = append(fields, channelmonitor.FieldAccountID)
+	}
 	if m.addinterval_seconds != nil {
 		fields = append(fields, channelmonitor.FieldIntervalSeconds)
 	}
@@ -16049,6 +16189,8 @@ func (m *ChannelMonitorMutation) AddedFields() []string {
 // was not set, or was not defined in the schema.
 func (m *ChannelMonitorMutation) AddedField(name string) (ent.Value, bool) {
 	switch name {
+	case channelmonitor.FieldAccountID:
+		return m.AddedAccountID()
 	case channelmonitor.FieldIntervalSeconds:
 		return m.AddedIntervalSeconds()
 	case channelmonitor.FieldJitterSeconds:
@@ -16064,6 +16206,13 @@ func (m *ChannelMonitorMutation) AddedField(name string) (ent.Value, bool) {
 // type.
 func (m *ChannelMonitorMutation) AddField(name string, value ent.Value) error {
 	switch name {
+	case channelmonitor.FieldAccountID:
+		v, ok := value.(int64)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.AddAccountID(v)
+		return nil
 	case channelmonitor.FieldIntervalSeconds:
 		v, ok := value.(int)
 		if !ok {
@@ -16093,6 +16242,9 @@ func (m *ChannelMonitorMutation) AddField(name string, value ent.Value) error {
 // mutation.
 func (m *ChannelMonitorMutation) ClearedFields() []string {
 	var fields []string
+	if m.FieldCleared(channelmonitor.FieldAccountID) {
+		fields = append(fields, channelmonitor.FieldAccountID)
+	}
 	if m.FieldCleared(channelmonitor.FieldGroupName) {
 		fields = append(fields, channelmonitor.FieldGroupName)
 	}
@@ -16119,6 +16271,9 @@ func (m *ChannelMonitorMutation) FieldCleared(name string) bool {
 // error if the field is not defined in the schema.
 func (m *ChannelMonitorMutation) ClearField(name string) error {
 	switch name {
+	case channelmonitor.FieldAccountID:
+		m.ClearAccountID()
+		return nil
 	case channelmonitor.FieldGroupName:
 		m.ClearGroupName()
 		return nil
@@ -16151,6 +16306,12 @@ func (m *ChannelMonitorMutation) ResetField(name string) error {
 	case channelmonitor.FieldProvider:
 		m.ResetProvider()
 		return nil
+	case channelmonitor.FieldCheckMode:
+		m.ResetCheckMode()
+		return nil
+	case channelmonitor.FieldAccountID:
+		m.ResetAccountID()
+		return nil
 	case channelmonitor.FieldAPIMode:
 		m.ResetAPIMode()
 		return nil
@@ -17756,6 +17917,7 @@ type ChannelMonitorHistoryMutation struct {
 	ping_latency_ms    *int
 	addping_latency_ms *int
 	message            *string
+	quota              **domain.MonitorQuotaSnapshot
 	checked_at         *time.Time
 	clearedFields      map[string]struct{}
 	monitor            *int64
@@ -18160,6 +18322,55 @@ func (m *ChannelMonitorHistoryMutation) ResetMessage() {
 	delete(m.clearedFields, channelmonitorhistory.FieldMessage)
 }
 
+// SetQuota sets the "quota" field.
+func (m *ChannelMonitorHistoryMutation) SetQuota(dqs *domain.MonitorQuotaSnapshot) {
+	m.quota = &dqs
+}
+
+// Quota returns the value of the "quota" field in the mutation.
+func (m *ChannelMonitorHistoryMutation) Quota() (r *domain.MonitorQuotaSnapshot, exists bool) {
+	v := m.quota
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldQuota returns the old "quota" field's value of the ChannelMonitorHistory entity.
+// If the ChannelMonitorHistory object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *ChannelMonitorHistoryMutation) OldQuota(ctx context.Context) (v *domain.MonitorQuotaSnapshot, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldQuota is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldQuota requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldQuota: %w", err)
+	}
+	return oldValue.Quota, nil
+}
+
+// ClearQuota clears the value of the "quota" field.
+func (m *ChannelMonitorHistoryMutation) ClearQuota() {
+	m.quota = nil
+	m.clearedFields[channelmonitorhistory.FieldQuota] = struct{}{}
+}
+
+// QuotaCleared returns if the "quota" field was cleared in this mutation.
+func (m *ChannelMonitorHistoryMutation) QuotaCleared() bool {
+	_, ok := m.clearedFields[channelmonitorhistory.FieldQuota]
+	return ok
+}
+
+// ResetQuota resets all changes to the "quota" field.
+func (m *ChannelMonitorHistoryMutation) ResetQuota() {
+	m.quota = nil
+	delete(m.clearedFields, channelmonitorhistory.FieldQuota)
+}
+
 // SetCheckedAt sets the "checked_at" field.
 func (m *ChannelMonitorHistoryMutation) SetCheckedAt(t time.Time) {
 	m.checked_at = &t
@@ -18257,7 +18468,7 @@ func (m *ChannelMonitorHistoryMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *ChannelMonitorHistoryMutation) Fields() []string {
-	fields := make([]string, 0, 7)
+	fields := make([]string, 0, 8)
 	if m.monitor != nil {
 		fields = append(fields, channelmonitorhistory.FieldMonitorID)
 	}
@@ -18276,6 +18487,9 @@ func (m *ChannelMonitorHistoryMutation) Fields() []string {
 	if m.message != nil {
 		fields = append(fields, channelmonitorhistory.FieldMessage)
 	}
+	if m.quota != nil {
+		fields = append(fields, channelmonitorhistory.FieldQuota)
+	}
 	if m.checked_at != nil {
 		fields = append(fields, channelmonitorhistory.FieldCheckedAt)
 	}
@@ -18299,6 +18513,8 @@ func (m *ChannelMonitorHistoryMutation) Field(name string) (ent.Value, bool) {
 		return m.PingLatencyMs()
 	case channelmonitorhistory.FieldMessage:
 		return m.Message()
+	case channelmonitorhistory.FieldQuota:
+		return m.Quota()
 	case channelmonitorhistory.FieldCheckedAt:
 		return m.CheckedAt()
 	}
@@ -18322,6 +18538,8 @@ func (m *ChannelMonitorHistoryMutation) OldField(ctx context.Context, name strin
 		return m.OldPingLatencyMs(ctx)
 	case channelmonitorhistory.FieldMessage:
 		return m.OldMessage(ctx)
+	case channelmonitorhistory.FieldQuota:
+		return m.OldQuota(ctx)
 	case channelmonitorhistory.FieldCheckedAt:
 		return m.OldCheckedAt(ctx)
 	}
@@ -18375,6 +18593,13 @@ func (m *ChannelMonitorHistoryMutation) SetField(name string, value ent.Value) e
 		}
 		m.SetMessage(v)
 		return nil
+	case channelmonitorhistory.FieldQuota:
+		v, ok := value.(*domain.MonitorQuotaSnapshot)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetQuota(v)
+		return nil
 	case channelmonitorhistory.FieldCheckedAt:
 		v, ok := value.(time.Time)
 		if !ok {
@@ -18448,6 +18673,9 @@ func (m *ChannelMonitorHistoryMutation) ClearedFields() []string {
 	if m.FieldCleared(channelmonitorhistory.FieldMessage) {
 		fields = append(fields, channelmonitorhistory.FieldMessage)
 	}
+	if m.FieldCleared(channelmonitorhistory.FieldQuota) {
+		fields = append(fields, channelmonitorhistory.FieldQuota)
+	}
 	return fields
 }
 
@@ -18471,6 +18699,9 @@ func (m *ChannelMonitorHistoryMutation) ClearField(name string) error {
 	case channelmonitorhistory.FieldMessage:
 		m.ClearMessage()
 		return nil
+	case channelmonitorhistory.FieldQuota:
+		m.ClearQuota()
+		return nil
 	}
 	return fmt.Errorf("unknown ChannelMonitorHistory nullable field %s", name)
 }
@@ -18497,6 +18728,9 @@ func (m *ChannelMonitorHistoryMutation) ResetField(name string) error {
 	case channelmonitorhistory.FieldMessage:
 		m.ResetMessage()
 		return nil
+	case channelmonitorhistory.FieldQuota:
+		m.ResetQuota()
+		return nil
 	case channelmonitorhistory.FieldCheckedAt:
 		m.ResetCheckedAt()
 		return nil
diff --git a/backend/ent/runtime/runtime.go b/backend/ent/runtime/runtime.go
index 4be8d374d6fc..03a6894e0dea 100644
--- a/backend/ent/runtime/runtime.go
+++ b/backend/ent/runtime/runtime.go
@@ -635,36 +635,28 @@ func init() {
 			return nil
 		}
 	}()
+	// channelmonitorDescCheckMode is the schema descriptor for check_mode field.
+	channelmonitorDescCheckMode := channelmonitorFields[2].Descriptor()
+	// channelmonitor.DefaultCheckMode holds the default value on creation for the check_mode field.
+	channelmonitor.DefaultCheckMode = channelmonitorDescCheckMode.Default.(string)
+	// channelmonitor.CheckModeValidator is a validator for the "check_mode" field. It is called by the builders before save.
+	channelmonitor.CheckModeValidator = channelmonitorDescCheckMode.Validators[0].(func(string) error)
 	// channelmonitorDescAPIMode is the schema descriptor for api_mode field.
-	channelmonitorDescAPIMode := channelmonitorFields[2].Descriptor()
+	channelmonitorDescAPIMode := channelmonitorFields[4].Descriptor()
 	// channelmonitor.DefaultAPIMode holds the default value on creation for the api_mode field.
 	channelmonitor.DefaultAPIMode = channelmonitorDescAPIMode.Default.(string)
 	// channelmonitor.APIModeValidator is a validator for the "api_mode" field. It is called by the builders before save.
 	channelmonitor.APIModeValidator = channelmonitorDescAPIMode.Validators[0].(func(string) error)
 	// channelmonitorDescEndpoint is the schema descriptor for endpoint field.
-	channelmonitorDescEndpoint := channelmonitorFields[3].Descriptor()
+	channelmonitorDescEndpoint := channelmonitorFields[5].Descriptor()
 	// channelmonitor.EndpointValidator is a validator for the "endpoint" field. It is called by the builders before save.
-	channelmonitor.EndpointValidator = func() func(string) error {
-		validators := channelmonitorDescEndpoint.Validators
-		fns := [...]func(string) error{
-			validators[0].(func(string) error),
-			validators[1].(func(string) error),
-		}
-		return func(endpoint string) error {
-			for _, fn := range fns {
-				if err := fn(endpoint); err != nil {
-					return err
-				}
-			}
-			return nil
-		}
-	}()
+	channelmonitor.EndpointValidator = channelmonitorDescEndpoint.Validators[0].(func(string) error)
 	// channelmonitorDescAPIKeyEncrypted is the schema descriptor for api_key_encrypted field.
-	channelmonitorDescAPIKeyEncrypted := channelmonitorFields[4].Descriptor()
+	channelmonitorDescAPIKeyEncrypted := channelmonitorFields[6].Descriptor()
 	// channelmonitor.APIKeyEncryptedValidator is a validator for the "api_key_encrypted" field. It is called by the builders before save.
 	channelmonitor.APIKeyEncryptedValidator = channelmonitorDescAPIKeyEncrypted.Validators[0].(func(string) error)
 	// channelmonitorDescPrimaryModel is the schema descriptor for primary_model field.
-	channelmonitorDescPrimaryModel := channelmonitorFields[5].Descriptor()
+	channelmonitorDescPrimaryModel := channelmonitorFields[7].Descriptor()
 	// channelmonitor.PrimaryModelValidator is a validator for the "primary_model" field. It is called by the builders before save.
 	channelmonitor.PrimaryModelValidator = func() func(string) error {
 		validators := channelmonitorDescPrimaryModel.Validators
@@ -682,35 +674,35 @@ func init() {
 		}
 	}()
 	// channelmonitorDescExtraModels is the schema descriptor for extra_models field.
-	channelmonitorDescExtraModels := channelmonitorFields[6].Descriptor()
+	channelmonitorDescExtraModels := channelmonitorFields[8].Descriptor()
 	// channelmonitor.DefaultExtraModels holds the default value on creation for the extra_models field.
 	channelmonitor.DefaultExtraModels = channelmonitorDescExtraModels.Default.([]string)
 	// channelmonitorDescGroupName is the schema descriptor for group_name field.
-	channelmonitorDescGroupName := channelmonitorFields[7].Descriptor()
+	channelmonitorDescGroupName := channelmonitorFields[9].Descriptor()
 	// channelmonitor.DefaultGroupName holds the default value on creation for the group_name field.
 	channelmonitor.DefaultGroupName = channelmonitorDescGroupName.Default.(string)
 	// channelmonitor.GroupNameValidator is a validator for the "group_name" field. It is called by the builders before save.
 	channelmonitor.GroupNameValidator = channelmonitorDescGroupName.Validators[0].(func(string) error)
 	// channelmonitorDescEnabled is the schema descriptor for enabled field.
-	channelmonitorDescEnabled := channelmonitorFields[8].Descriptor()
+	channelmonitorDescEnabled := channelmonitorFields[10].Descriptor()
 	// channelmonitor.DefaultEnabled holds the default value on creation for the enabled field.
 	channelmonitor.DefaultEnabled = channelmonitorDescEnabled.Default.(bool)
 	// channelmonitorDescIntervalSeconds is the schema descriptor for interval_seconds field.
-	channelmonitorDescIntervalSeconds := channelmonitorFields[9].Descriptor()
+	channelmonitorDescIntervalSeconds := channelmonitorFields[11].Descriptor()
 	// channelmonitor.IntervalSecondsValidator is a validator for the "interval_seconds" field. It is called by the builders before save.
 	channelmonitor.IntervalSecondsValidator = channelmonitorDescIntervalSeconds.Validators[0].(func(int) error)
 	// channelmonitorDescJitterSeconds is the schema descriptor for jitter_seconds field.
-	channelmonitorDescJitterSeconds := channelmonitorFields[10].Descriptor()
+	channelmonitorDescJitterSeconds := channelmonitorFields[12].Descriptor()
 	// channelmonitor.DefaultJitterSeconds holds the default value on creation for the jitter_seconds field.
 	channelmonitor.DefaultJitterSeconds = channelmonitorDescJitterSeconds.Default.(int)
 	// channelmonitor.JitterSecondsValidator is a validator for the "jitter_seconds" field. It is called by the builders before save.
 	channelmonitor.JitterSecondsValidator = channelmonitorDescJitterSeconds.Validators[0].(func(int) error)
 	// channelmonitorDescExtraHeaders is the schema descriptor for extra_headers field.
-	channelmonitorDescExtraHeaders := channelmonitorFields[14].Descriptor()
+	channelmonitorDescExtraHeaders := channelmonitorFields[16].Descriptor()
 	// channelmonitor.DefaultExtraHeaders holds the default value on creation for the extra_headers field.
 	channelmonitor.DefaultExtraHeaders = channelmonitorDescExtraHeaders.Default.(map[string]string)
 	// channelmonitorDescBodyOverrideMode is the schema descriptor for body_override_mode field.
-	channelmonitorDescBodyOverrideMode := channelmonitorFields[15].Descriptor()
+	channelmonitorDescBodyOverrideMode := channelmonitorFields[17].Descriptor()
 	// channelmonitor.DefaultBodyOverrideMode holds the default value on creation for the body_override_mode field.
 	channelmonitor.DefaultBodyOverrideMode = channelmonitorDescBodyOverrideMode.Default.(string)
 	// channelmonitor.BodyOverrideModeValidator is a validator for the "body_override_mode" field. It is called by the builders before save.
@@ -808,7 +800,7 @@ func init() {
 	// channelmonitorhistory.MessageValidator is a validator for the "message" field. It is called by the builders before save.
 	channelmonitorhistory.MessageValidator = channelmonitorhistoryDescMessage.Validators[0].(func(string) error)
 	// channelmonitorhistoryDescCheckedAt is the schema descriptor for checked_at field.
-	channelmonitorhistoryDescCheckedAt := channelmonitorhistoryFields[6].Descriptor()
+	channelmonitorhistoryDescCheckedAt := channelmonitorhistoryFields[7].Descriptor()
 	// channelmonitorhistory.DefaultCheckedAt holds the default value on creation for the checked_at field.
 	channelmonitorhistory.DefaultCheckedAt = channelmonitorhistoryDescCheckedAt.Default.(func() time.Time)
 	channelmonitorrequesttemplateMixin := schema.ChannelMonitorRequestTemplate{}.Mixin()
diff --git a/backend/ent/schema/channel_monitor.go b/backend/ent/schema/channel_monitor.go
index cb62079316e9..e14416b76541 100644
--- a/backend/ent/schema/channel_monitor.go
+++ b/backend/ent/schema/channel_monitor.go
@@ -35,15 +35,32 @@ func (ChannelMonitor) Fields() []ent.Field {
 			NotEmpty().
 			MaxLen(100),
 		field.Enum("provider").
-			Values("openai", "anthropic", "gemini", "grok"),
+			Values("openai", "anthropic", "gemini", "grok",
+				"antigravity", "kimi", "zhipu", "deepseek"),
+		// check_mode: 'probe' | 'quota' | 'quota_probe'
+		//   probe       - LLM 探活（默认，原有行为）
+		//   quota       - 仅查关联账号的用量/余额（零 LLM 成本；endpoint/api_key 可空）
+		//   quota_probe - 探活 + 配额并存（配额快照挂到主模型历史行）
+		// antigravity 无探活 adapter，仅允许 quota。
+		field.String("check_mode").
+			Default("probe").
+			MaxLen(32).
+			Comment("probe = LLM probe (default); quota = account usage only; quota_probe = both"),
+		// account_id: 配额模式的数据源账号（复用账号侧用量服务，不直接对接上游）。
+		// 普通字段而非 edge（FK 由 SQL 迁移管理）；账号删除时数据库置空，
+		// 监控保留并报「账号未关联」。
+		field.Int64("account_id").
+			Optional().
+			Nillable(),
 		field.String("api_mode").
 			Default("chat_completions").
 			MaxLen(32).
 			Comment("OpenAI request protocol: chat_completions or responses; non-OpenAI uses chat_completions"),
+		// endpoint: 探活模式必填（service 层校验）；quota 模式存空串
+		// （列保持 NOT NULL，去掉 NotEmpty 校验器即可）。
 		field.String("endpoint").
-			NotEmpty().
 			MaxLen(500).
-			Comment("Provider base origin, e.g. https://api.openai.com"),
+			Comment("Provider base origin, e.g. https://api.openai.com; empty for quota-only monitors"),
 		field.String("api_key_encrypted").
 			NotEmpty().
 			Sensitive().
@@ -115,5 +132,6 @@ func (ChannelMonitor) Indexes() []ent.Index {
 		index.Fields("provider", "api_mode"),
 		index.Fields("group_name"),
 		index.Fields("template_id"),
+		index.Fields("account_id"),
 	}
 }
diff --git a/backend/ent/schema/channel_monitor_history.go b/backend/ent/schema/channel_monitor_history.go
index 4366e79a672f..c2ec4dc25c14 100644
--- a/backend/ent/schema/channel_monitor_history.go
+++ b/backend/ent/schema/channel_monitor_history.go
@@ -3,6 +3,8 @@ package schema
 import (
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/domain"
+
 	"entgo.io/ent"
 	"entgo.io/ent/dialect/entsql"
 	"entgo.io/ent/schema"
@@ -43,6 +45,10 @@ func (ChannelMonitorHistory) Fields() []ent.Field {
 			Optional().
 			Default("").
 			MaxLen(500),
+		// quota: 配额模式（check_mode = quota / quota_probe）检测时附带的
+		// 归一化配额快照（domain.MonitorQuotaSnapshot，JSONB）；探活模式为 NULL。
+		field.JSON("quota", &domain.MonitorQuotaSnapshot{}).
+			Optional(),
 		field.Time("checked_at").
 			Default(time.Now),
 	}
diff --git a/backend/ent/schema/channel_monitor_request_template.go b/backend/ent/schema/channel_monitor_request_template.go
index cf7fe0515803..332a01007c0a 100644
--- a/backend/ent/schema/channel_monitor_request_template.go
+++ b/backend/ent/schema/channel_monitor_request_template.go
@@ -39,7 +39,8 @@ func (ChannelMonitorRequestTemplate) Fields() []ent.Field {
 			NotEmpty().
 			MaxLen(100),
 		field.Enum("provider").
-			Values("openai", "anthropic", "gemini", "grok"),
+			Values("openai", "anthropic", "gemini", "grok",
+				"antigravity", "kimi", "zhipu", "deepseek"),
 		field.String("api_mode").
 			Default("chat_completions").
 			MaxLen(32).
diff --git a/backend/internal/domain/channel_monitor_quota.go b/backend/internal/domain/channel_monitor_quota.go
new file mode 100644
index 000000000000..656163dddf6f
--- /dev/null
+++ b/backend/internal/domain/channel_monitor_quota.go
@@ -0,0 +1,55 @@
+package domain
+
+import "time"
+
+// 渠道监控「配额模式」的归一化配额快照类型。
+//
+// 配额模式监控不直接对接上游，而是关联一个已有账号，复用账号侧的用量服务
+// （AccountUsageService / CNProviderQuotaService / CNProviderBalanceService），
+// 把各平台形态各异的用量数据归一成 MonitorQuotaSnapshot，随检测历史持久化
+// 到 channel_monitor_histories.quota（JSONB），供管理端与用户端渲染。
+//
+// 类型放在 domain 包是因为 ent schema（internal/domain 的下游）需要引用它做
+// field.JSON 序列化；service 不能被 ent import（会造成循环依赖）。
+
+// MonitorQuotaTier 单个用量窗口的快照。
+//
+// Window 取值约定（与前端 monitorCommon.quota.windows.* 标签一一对应）：
+//   - "5h"         5 小时滚动窗口（Claude/Codex/Kimi/Zhipu coding plan）
+//   - "7d"         7 天窗口（Claude/Codex）
+//   - "7d-sonnet"  Claude 7 天 Sonnet 独立额度
+//   - "7d-fable"   Claude 7 天 Fable 独立额度
+//   - "weekly"     周窗口（Kimi/Zhipu coding plan）
+//   - "30d"        30 天窗口（Grok 月度）
+//   - "total"      无窗口语义的总量额度（Gemini/Antigravity 等）
+type MonitorQuotaTier struct {
+	Window      string  `json:"window"`
+	UsedPercent float64 `json:"used_percent"` // 0-100；仅有绝对值时按 used/limit 计算
+	Used        float64 `json:"used,omitempty"`
+	Limit       float64 `json:"limit,omitempty"`
+	ResetAt     string  `json:"reset_at,omitempty"` // RFC3339；未知时留空
+}
+
+// MonitorQuotaSnapshot 一次配额查询的完整快照。
+//
+// Source 取值：
+//   - "usage"      海外平台（AccountUsageService.GetUsage）
+//   - "cn_quota"   国产 Coding Plan（CNProviderQuotaService.QueryUsage）
+//   - "cn_balance" 国产按量付费余额（CNProviderBalanceService.QueryBalance）
+type MonitorQuotaSnapshot struct {
+	Source    string             `json:"source"`
+	Success   bool               `json:"success"`
+	Tiers     []MonitorQuotaTier `json:"tiers,omitempty"`
+	Balance   *float64           `json:"balance,omitempty"`           // cn_balance 主余额
+	Balances  []MonitorBalance   `json:"balances,omitempty"`          // 多币种余额（如 DeepSeek CNY+USD）
+	Currency  string             `json:"currency,omitempty"`          // 主余额币种
+	PlanLevel string             `json:"plan_level,omitempty"`        // 套餐等级（如智谱 level）
+	Error     string             `json:"error,omitempty"`             // Success=false 时的错误摘要
+	FetchedAt time.Time          `json:"fetched_at"`
+}
+
+// MonitorBalance 单币种余额条目。
+type MonitorBalance struct {
+	Currency string  `json:"currency"`
+	Balance  float64 `json:"balance"`
+}
diff --git a/backend/migrations/226_channel_monitor_quota_mode.sql b/backend/migrations/226_channel_monitor_quota_mode.sql
new file mode 100644
index 000000000000..a0774e3dbf6e
--- /dev/null
+++ b/backend/migrations/226_channel_monitor_quota_mode.sql
@@ -0,0 +1,78 @@
+-- Migration: 226_channel_monitor_quota_mode
+-- 渠道监控配额模式：
+--   1. provider 扩容到全部 8 平台（antigravity/kimi/zhipu/deepseek）
+--      （antigravity 仅支持配额模式，无探活 adapter；国产 3 家复用 OpenAI 兼容探活）
+--   2. check_mode：probe（默认，现状探活）/ quota（仅查关联账号用量，零 LLM 成本）
+--      / quota_probe（探活 + 配额并存）
+--   3. account_id 关联已有账号（配额模式的数据源，复用账号侧用量服务）；
+--      账号删除时置空，监控保留并报「账号未关联」
+--   4. channel_monitor_histories.quota 持久化归一化配额快照（JSONB）
+--   5. 新增公开设置 channel_monitor_show_quota（默认关闭）：
+--      控制用户端监控页是否展示配额/余额；管理端始终可见
+
+DO $$
+DECLARE
+    monitor_constraint_def TEXT;
+    template_constraint_def TEXT;
+BEGIN
+    SELECT pg_get_constraintdef(c.oid)
+      INTO monitor_constraint_def
+      FROM pg_constraint c
+      JOIN pg_class t ON t.oid = c.conrelid
+     WHERE t.relname = 'channel_monitors'
+       AND c.conname = 'channel_monitors_provider_check';
+
+    IF monitor_constraint_def IS NULL OR position('kimi' IN monitor_constraint_def) = 0 THEN
+        ALTER TABLE channel_monitors
+            DROP CONSTRAINT IF EXISTS channel_monitors_provider_check;
+        ALTER TABLE channel_monitors
+            ADD CONSTRAINT channel_monitors_provider_check
+            CHECK (provider IN ('openai', 'anthropic', 'gemini', 'grok',
+                                'antigravity', 'kimi', 'zhipu', 'deepseek'));
+    END IF;
+
+    SELECT pg_get_constraintdef(c.oid)
+      INTO template_constraint_def
+      FROM pg_constraint c
+      JOIN pg_class t ON t.oid = c.conrelid
+     WHERE t.relname = 'channel_monitor_request_templates'
+       AND c.conname = 'channel_monitor_request_templates_provider_check';
+
+    IF template_constraint_def IS NULL OR position('kimi' IN template_constraint_def) = 0 THEN
+        ALTER TABLE channel_monitor_request_templates
+            DROP CONSTRAINT IF EXISTS channel_monitor_request_templates_provider_check;
+        ALTER TABLE channel_monitor_request_templates
+            ADD CONSTRAINT channel_monitor_request_templates_provider_check
+            CHECK (provider IN ('openai', 'anthropic', 'gemini', 'grok',
+                                'antigravity', 'kimi', 'zhipu', 'deepseek'));
+    END IF;
+END $$;
+
+ALTER TABLE channel_monitors
+    ADD COLUMN IF NOT EXISTS check_mode VARCHAR(32) NOT NULL DEFAULT 'probe';
+
+ALTER TABLE channel_monitors
+    ADD CONSTRAINT channel_monitors_check_mode_check
+    CHECK (check_mode IN ('probe', 'quota', 'quota_probe'));
+
+ALTER TABLE channel_monitors
+    ADD COLUMN IF NOT EXISTS account_id BIGINT REFERENCES accounts(id) ON DELETE SET NULL;
+
+CREATE INDEX IF NOT EXISTS idx_channel_monitors_account_id ON channel_monitors(account_id);
+
+COMMENT ON COLUMN channel_monitors.check_mode IS
+    'probe = LLM 探活（默认）；quota = 仅查关联账号用量；quota_probe = 探活 + 配额';
+COMMENT ON COLUMN channel_monitors.account_id IS
+    '配额模式关联的账号 ID（数据源）；账号删除时置空';
+
+ALTER TABLE channel_monitor_histories
+    ADD COLUMN IF NOT EXISTS quota JSONB;
+
+COMMENT ON COLUMN channel_monitor_histories.quota IS
+    '配额模式监控的归一化配额快照（domain.MonitorQuotaSnapshot）；探活模式为 NULL';
+
+-- 用户端是否展示配额/余额（默认关闭，fail-closed 解析：仅 "true" 视为开启）。
+-- 管理端不受此开关影响。
+INSERT INTO settings (key, value)
+VALUES ('channel_monitor_show_quota', 'false')
+ON CONFLICT (key) DO NOTHING;

From c44711ac98dd7f2cafacff18aa9b3bfa9e7bcbb1 Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 04:03:13 +0000
Subject: [PATCH 46/68] feat(channel-monitor): quota mode service layer
 (fetcher + dispatch + repo)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 常量：MonitorProvider{Antigravity,Kimi,Zhipu,Deepseek}、MonitorCheckMode 三态、
  配额缓存/告警阈值参数；新增 check_mode 相关业务错误
- 校验：monitorProviders(8)/probeCapableProviders(7) 集合替代 adapter 表判定，
  validateCheckMode 矩阵（antigravity 仅 quota）；quota 模式 primary_model
  默认占位 "quota"，endpoint/api_key 条件必填
- checker：kimi/deepseek 复用 /v1/chat/completions、zhipu 走
  /api/paas/v4/chat/completions；merge 黑名单与 replace 校验同步扩容
- ChannelMonitorQuotaFetcher：窄接口聚合账号侧三个用量服务 + 账号加载，
  归一为 domain.MonitorQuotaSnapshot；成功快照 5min TTL 缓存防配额端点风暴；
  Fetch 永不返回 error（失败降级 Success=false 快照）；状态推导
  operational/degraded(≥90% 或余额耗尽/账号未关联)/failed(401/403)/error
- RunCheck 按 check_mode 分派：quota 单行结果、quota_probe 探活+配额挂主行、
  probe 不变；Update 组合校验 + 关联账号复核（probe 自动解绑失效账号）；
  Duplicate 拷贝新字段且 quota 模式空明文重加密（防 runner 误停摆）
- repo：Create/Update/entToServiceMonitor 透传 check_mode/account_id；
  InsertHistoryBatch 写 quota；ListLatestForMonitorIDs 裸 SQL 加 quota 列 +
  scanMonitorQuota JSONB 解包（NULL 安全）；ListHistory 透传 quota
---
 .../internal/domain/channel_monitor_quota.go  |  17 +-
 .../repository/channel_monitor_repo.go        |  53 ++-
 .../service/channel_monitor_aggregator.go     |   3 +
 .../service/channel_monitor_checker.go        |  36 +-
 .../internal/service/channel_monitor_const.go |  50 +-
 .../service/channel_monitor_quota_fetcher.go  | 429 ++++++++++++++++++
 .../service/channel_monitor_service.go        | 238 +++++++++-
 .../internal/service/channel_monitor_types.go |  29 +-
 .../service/channel_monitor_validate.go       |  81 +++-
 9 files changed, 900 insertions(+), 36 deletions(-)
 create mode 100644 backend/internal/service/channel_monitor_quota_fetcher.go

diff --git a/backend/internal/domain/channel_monitor_quota.go b/backend/internal/domain/channel_monitor_quota.go
index 656163dddf6f..fed21dedb797 100644
--- a/backend/internal/domain/channel_monitor_quota.go
+++ b/backend/internal/domain/channel_monitor_quota.go
@@ -20,11 +20,17 @@ import "time"
 //   - "7d-sonnet"  Claude 7 天 Sonnet 独立额度
 //   - "7d-fable"   Claude 7 天 Fable 独立额度
 //   - "weekly"     周窗口（Kimi/Zhipu coding plan）
+//   - "daily"      日窗口（Gemini 日配额 / Grok 日请求）
 //   - "30d"        30 天窗口（Grok 月度）
-//   - "total"      无窗口语义的总量额度（Gemini/Antigravity 等）
+//   - "total"      无窗口语义的总量额度（Antigravity per-model 等）
+//
+// 同一 Window 可能出现多条（Gemini 多档日配额、Antigravity per-model、
+// Grok requests/tokens），用 Label 区分：Label 是机器 token（requests/tokens/
+// shared/pro/flash 或模型名），前端已知 token 走 i18n，未知原样展示。
 type MonitorQuotaTier struct {
 	Window      string  `json:"window"`
-	UsedPercent float64 `json:"used_percent"` // 0-100；仅有绝对值时按 used/limit 计算
+	Label       string  `json:"label,omitempty"`
+	UsedPercent float64 `json:"used_percent"` // 0-100+；仅有绝对值时按 used/limit 计算
 	Used        float64 `json:"used,omitempty"`
 	Limit       float64 `json:"limit,omitempty"`
 	ResetAt     string  `json:"reset_at,omitempty"` // RFC3339；未知时留空
@@ -44,8 +50,11 @@ type MonitorQuotaSnapshot struct {
 	Balances  []MonitorBalance   `json:"balances,omitempty"`          // 多币种余额（如 DeepSeek CNY+USD）
 	Currency  string             `json:"currency,omitempty"`          // 主余额币种
 	PlanLevel string             `json:"plan_level,omitempty"`        // 套餐等级（如智谱 level）
-	Error     string             `json:"error,omitempty"`             // Success=false 时的错误摘要
-	FetchedAt time.Time          `json:"fetched_at"`
+	// CredentialInvalid 上游 401/403 鉴权失败（区别于网络/解析错误），
+	// 检测状态据此推导 failed 而非 error。
+	CredentialInvalid bool      `json:"credential_invalid,omitempty"`
+	Error             string    `json:"error,omitempty"` // Success=false 时的错误摘要
+	FetchedAt         time.Time `json:"fetched_at"`
 }
 
 // MonitorBalance 单币种余额条目。
diff --git a/backend/internal/repository/channel_monitor_repo.go b/backend/internal/repository/channel_monitor_repo.go
index aa8b12ff8bea..c4586bc38c27 100644
--- a/backend/internal/repository/channel_monitor_repo.go
+++ b/backend/internal/repository/channel_monitor_repo.go
@@ -3,6 +3,7 @@ package repository
 import (
 	"context"
 	"database/sql"
+	"encoding/json"
 	"fmt"
 	"strings"
 	"time"
@@ -10,6 +11,7 @@ import (
 	dbent "github.com/Wei-Shaw/sub2api/ent"
 	"github.com/Wei-Shaw/sub2api/ent/channelmonitor"
 	"github.com/Wei-Shaw/sub2api/ent/channelmonitorhistory"
+	"github.com/Wei-Shaw/sub2api/internal/domain"
 	"github.com/Wei-Shaw/sub2api/internal/service"
 	"github.com/lib/pq"
 
@@ -51,10 +53,14 @@ func (r *channelMonitorRepository) Create(ctx context.Context, m *service.Channe
 		SetJitterSeconds(m.JitterSeconds).
 		SetCreatedBy(m.CreatedBy).
 		SetExtraHeaders(channelMonitorHeadersForPersistence(m)).
-		SetBodyOverrideMode(defaultBodyModeRepo(m.BodyOverrideMode))
+		SetBodyOverrideMode(defaultBodyModeRepo(m.BodyOverrideMode)).
+		SetCheckMode(defaultCheckModeRepo(m.CheckMode))
 	if m.TemplateID != nil {
 		builder = builder.SetTemplateID(*m.TemplateID)
 	}
+	if m.AccountID != nil {
+		builder = builder.SetAccountID(*m.AccountID)
+	}
 	if m.BodyOverride != nil {
 		builder = builder.SetBodyOverride(m.BodyOverride)
 	}
@@ -118,12 +124,18 @@ func (r *channelMonitorRepository) Update(ctx context.Context, m *service.Channe
 		SetIntervalSeconds(m.IntervalSeconds).
 		SetJitterSeconds(m.JitterSeconds).
 		SetExtraHeaders(channelMonitorHeadersForPersistence(m)).
-		SetBodyOverrideMode(defaultBodyModeRepo(m.BodyOverrideMode))
+		SetBodyOverrideMode(defaultBodyModeRepo(m.BodyOverrideMode)).
+		SetCheckMode(defaultCheckModeRepo(m.CheckMode))
 	if m.TemplateID != nil {
 		updater = updater.SetTemplateID(*m.TemplateID)
 	} else {
 		updater = updater.ClearTemplateID()
 	}
+	if m.AccountID != nil {
+		updater = updater.SetAccountID(*m.AccountID)
+	} else {
+		updater = updater.ClearAccountID()
+	}
 	if m.BodyOverride != nil {
 		updater = updater.SetBodyOverride(m.BodyOverride)
 	} else {
@@ -237,6 +249,9 @@ func (r *channelMonitorRepository) InsertHistoryBatch(ctx context.Context, rows
 		if row.PingLatencyMs != nil {
 			c = c.SetPingLatencyMs(*row.PingLatencyMs)
 		}
+		if row.Quota != nil {
+			c = c.SetQuota(row.Quota)
+		}
 		bulk = append(bulk, c)
 	}
 	if _, err := client.ChannelMonitorHistory.CreateBulk(bulk...).Save(ctx); err != nil {
@@ -276,6 +291,7 @@ func (r *channelMonitorRepository) ListHistory(ctx context.Context, monitorID in
 			PingLatencyMs: row.PingLatencyMs,
 			Message:       row.Message,
 			CheckedAt:     row.CheckedAt,
+			Quota:         row.Quota,
 		}
 		out = append(out, entry)
 	}
@@ -324,6 +340,20 @@ func assignNullInt(dst **int, n sql.NullInt64) {
 	*dst = &v
 }
 
+// scanMonitorQuota 把裸 SQL 读出的 JSONB quota 列解包为配额快照。
+// NULL（探活模式旧行）返回 nil；解析失败也返回 nil 并由调用方日志感知，
+// 不阻断列表渲染（与聚合层"失败仅日志"的原则一致）。
+func scanMonitorQuota(data []byte) *domain.MonitorQuotaSnapshot {
+	if len(data) == 0 {
+		return nil
+	}
+	snapshot := &domain.MonitorQuotaSnapshot{}
+	if err := json.Unmarshal(data, snapshot); err != nil {
+		return nil
+	}
+	return snapshot
+}
+
 // ComputeAvailability 计算指定窗口内每个模型的可用率与平均延迟。
 // "可用" = status IN (operational, degraded)。
 //
@@ -396,7 +426,7 @@ func (r *channelMonitorRepository) ListLatestForMonitorIDs(ctx context.Context,
 	}
 	const q = `
 		SELECT DISTINCT ON (monitor_id, model)
-		    monitor_id, model, status, latency_ms, ping_latency_ms, checked_at
+		    monitor_id, model, status, latency_ms, ping_latency_ms, checked_at, quota
 		FROM channel_monitor_histories
 		WHERE monitor_id = ANY($1)
 		ORDER BY monitor_id, model, checked_at DESC
@@ -411,11 +441,13 @@ func (r *channelMonitorRepository) ListLatestForMonitorIDs(ctx context.Context,
 		var monitorID int64
 		l := &service.ChannelMonitorLatest{}
 		var latency, ping sql.NullInt64
-		if err := rows.Scan(&monitorID, &l.Model, &l.Status, &latency, &ping, &l.CheckedAt); err != nil {
+		var quota []byte
+		if err := rows.Scan(&monitorID, &l.Model, &l.Status, &latency, &ping, &l.CheckedAt, &quota); err != nil {
 			return nil, fmt.Errorf("scan latest batch row: %w", err)
 		}
 		assignNullInt(&l.LatencyMs, latency)
 		assignNullInt(&l.PingLatencyMs, ping)
+		l.Quota = scanMonitorQuota(quota)
 		out[monitorID] = append(out[monitorID], l)
 	}
 	if err := rows.Err(); err != nil {
@@ -757,12 +789,17 @@ func entToServiceMonitor(row *dbent.ChannelMonitor) *service.ChannelMonitor {
 		ExtraHeaders:         headers,
 		BodyOverrideMode:     row.BodyOverrideMode,
 		BodyOverride:         row.BodyOverride,
+		CheckMode:            defaultCheckModeRepo(row.CheckMode),
 		DuplicateOperationID: duplicateOperationID,
 	}
 	if row.TemplateID != nil {
 		id := *row.TemplateID
 		out.TemplateID = &id
 	}
+	if row.AccountID != nil {
+		id := *row.AccountID
+		out.AccountID = &id
+	}
 	return out
 }
 
@@ -807,6 +844,14 @@ func defaultAPIModeRepo(apiMode string) string {
 	return apiMode
 }
 
+// defaultCheckModeRepo 空串归一为 probe（存量行有列默认值，这里兜底防御）。
+func defaultCheckModeRepo(checkMode string) string {
+	if checkMode == "" {
+		return "probe"
+	}
+	return checkMode
+}
+
 func emptySliceIfNil(in []string) []string {
 	if in == nil {
 		return []string{}
diff --git a/backend/internal/service/channel_monitor_aggregator.go b/backend/internal/service/channel_monitor_aggregator.go
index 09020f5fa427..b6f632bf4b17 100644
--- a/backend/internal/service/channel_monitor_aggregator.go
+++ b/backend/internal/service/channel_monitor_aggregator.go
@@ -204,6 +204,8 @@ func buildStatusSummary(
 		if l, ok := latestByModel[primary]; ok {
 			summary.PrimaryStatus = l.Status
 			summary.PrimaryLatencyMs = l.LatencyMs
+			// 配额快照只挂主模型行（quota 模式唯一行 / quota_probe 的主行）。
+			summary.LatestQuota = l.Quota
 		}
 		if a, ok := availByModel[primary]; ok {
 			summary.Availability7d = a.AvailabilityPct
@@ -242,6 +244,7 @@ func buildUserViewFromSummary(
 	}
 	if primaryLatest != nil {
 		view.PrimaryPingLatencyMs = primaryLatest.PingLatencyMs
+		view.LatestQuota = primaryLatest.Quota
 	}
 	return view
 }
diff --git a/backend/internal/service/channel_monitor_checker.go b/backend/internal/service/channel_monitor_checker.go
index 910a6e80d99b..bc8aa7064584 100644
--- a/backend/internal/service/channel_monitor_checker.go
+++ b/backend/internal/service/channel_monitor_checker.go
@@ -170,6 +170,11 @@ type providerAdapter struct {
 var providerAdapters = map[string]providerAdapter{
 	MonitorProviderOpenAI: providerOpenAIChatAdapter,
 	MonitorProviderGrok:   providerGrokChatAdapter,
+	// 国产 3 家（配额模式引入）：均为 OpenAI 兼容 Chat Completions，
+	// 仅智谱路径前缀不同（/api/paas/v4/chat/completions）。
+	MonitorProviderKimi:     providerKimiChatAdapter,
+	MonitorProviderZhipu:    providerZhipuChatAdapter,
+	MonitorProviderDeepseek: providerDeepseekChatAdapter,
 	MonitorProviderAnthropic: {
 		buildPath: func(string) string { return providerAnthropicPath },
 		buildBody: func(model, prompt string) ([]byte, error) {
@@ -212,6 +217,15 @@ var providerOpenAIChatAdapter = newOpenAICompatibleChatAdapter(providerOpenAIPat
 //nolint:gochecknoglobals // 适配器表是只读静态数据，初始化后不变更。
 var providerGrokChatAdapter = newOpenAICompatibleChatAdapter(providerGrokPath)
 
+//nolint:gochecknoglobals // 适配器表是只读静态数据，初始化后不变更。
+var providerKimiChatAdapter = newOpenAICompatibleChatAdapter(providerOpenAIPath)
+
+//nolint:gochecknoglobals // 适配器表是只读静态数据，初始化后不变更。
+var providerZhipuChatAdapter = newOpenAICompatibleChatAdapter(providerZhipuPath)
+
+//nolint:gochecknoglobals // 适配器表是只读静态数据，初始化后不变更。
+var providerDeepseekChatAdapter = newOpenAICompatibleChatAdapter(providerOpenAIPath)
+
 func newOpenAICompatibleChatAdapter(path string) providerAdapter {
 	return providerAdapter{
 		buildPath: func(string) string { return path },
@@ -257,8 +271,8 @@ func providerAdapterFor(provider, apiMode string) (providerAdapter, string, bool
 	return adapter, MonitorAPIModeChatCompletions, ok
 }
 
-// isSupportedProvider 校验 provider 字符串是否在 adapter 表中。
-// 供 validate.go 的 validateProvider 复用，避免两份 switch 漂移。
+// isSupportedProvider 校验 provider 字符串是否在探活 adapter 表中。
+// 注意：完整的 provider 合法性（含仅配额模式的 antigravity）走 validate.go 的 monitorProviders。
 func isSupportedProvider(p string) bool {
 	_, ok := providerAdapters[p]
 	return ok
@@ -447,6 +461,10 @@ var bodyMergeKeyDenyList = map[string]map[string]bool{
 	MonitorProviderGrok:      {"model": true, "messages": true, "stream": true},
 	MonitorProviderAnthropic: {"model": true, "messages": true},
 	MonitorProviderGemini:    {"contents": true},
+	// 国产 3 家与 OpenAI Chat Completions 同构。
+	MonitorProviderKimi:     {"model": true, "messages": true, "stream": true},
+	MonitorProviderZhipu:    {"model": true, "messages": true, "stream": true},
+	MonitorProviderDeepseek: {"model": true, "messages": true, "stream": true},
 }
 
 func checkAPIMode(opts *CheckOptions) string {
@@ -463,8 +481,20 @@ func bodyMergeDenyKey(provider, apiMode string) string {
 	return provider
 }
 
+// isOpenAICompatibleChatProvider 该 provider 的探活请求是否为 OpenAI Chat
+// Completions 同构（replace 模式的 body 校验按 messages 必填处理）。
+func isOpenAICompatibleChatProvider(provider string) bool {
+	switch provider {
+	case MonitorProviderOpenAI, MonitorProviderGrok,
+		MonitorProviderKimi, MonitorProviderZhipu, MonitorProviderDeepseek:
+		return true
+	default:
+		return false
+	}
+}
+
 func validateReplaceRequestBody(provider, apiMode string, body map[string]any) error {
-	if provider != MonitorProviderOpenAI && provider != MonitorProviderGrok {
+	if !isOpenAICompatibleChatProvider(provider) {
 		return nil
 	}
 	switch defaultAPIMode(apiMode) {
diff --git a/backend/internal/service/channel_monitor_const.go b/backend/internal/service/channel_monitor_const.go
index 6a41add2980e..4ed6a026ea6d 100644
--- a/backend/internal/service/channel_monitor_const.go
+++ b/backend/internal/service/channel_monitor_const.go
@@ -45,10 +45,12 @@ const (
 	monitorChallengeMin = 1
 	monitorChallengeMax = 50
 
-	// providerOpenAIPath OpenAI Chat Completions 路径。
+	// providerOpenAIPath OpenAI Chat Completions 路径（Kimi / DeepSeek 同为 OpenAI 兼容）。
 	providerOpenAIPath = "/v1/chat/completions"
 	// providerGrokPath Grok OpenAI-compatible Chat Completions 路径。
 	providerGrokPath = "/v1/chat/completions"
+	// providerZhipuPath 智谱 OpenAI 兼容 Chat Completions 路径（前缀与官方不同）。
+	providerZhipuPath = "/api/paas/v4/chat/completions"
 	// providerOpenAIResponsesPath OpenAI Responses API 路径。
 	providerOpenAIResponsesPath = "/v1/responses"
 	// providerAnthropicPath Anthropic Messages 路径。
@@ -56,11 +58,36 @@ const (
 	// providerGeminiPathTemplate Gemini generateContent 路径模板（含 model 占位）。
 	providerGeminiPathTemplate = "/v1beta/models/%s:generateContent"
 
-	// MonitorProviderOpenAI / Anthropic / Gemini / Grok provider 字符串常量（也是 ent enum 的实际值）。
-	MonitorProviderOpenAI    = "openai"
-	MonitorProviderAnthropic = "anthropic"
-	MonitorProviderGemini    = "gemini"
-	MonitorProviderGrok      = "grok"
+	// MonitorProviderOpenAI 等 provider 字符串常量（也是 ent enum 的实际值）。
+	// 后 4 个 provider（antigravity/kimi/zhipu/deepseek）为配额模式引入：
+	// antigravity 无探活 adapter（仅配额），其余 3 个复用 OpenAI 兼容探活。
+	MonitorProviderOpenAI     = "openai"
+	MonitorProviderAnthropic  = "anthropic"
+	MonitorProviderGemini     = "gemini"
+	MonitorProviderGrok       = "grok"
+	MonitorProviderAntigravity = "antigravity"
+	MonitorProviderKimi       = "kimi"
+	MonitorProviderZhipu      = "zhipu"
+	MonitorProviderDeepseek   = "deepseek"
+
+	// MonitorCheckMode 检测模式（channel_monitors.check_mode）。
+	//   probe       - LLM 探活（默认，原有行为）
+	//   quota       - 仅查关联账号用量/余额，零 LLM 成本
+	//   quota_probe - 探活 + 配额并存（配额快照挂到主模型历史行）
+	MonitorCheckModeProbe     = "probe"
+	MonitorCheckModeQuota     = "quota"
+	MonitorCheckModeQuotaProbe = "quota_probe"
+
+	// MonitorDefaultQuotaModel 是 quota 模式监控未显式指定模型时占位的虚拟模型名
+	// （primary_model 列 NotEmpty，用 "quota" 让历史行/时间线机制无需特判）。
+	MonitorDefaultQuotaModel = "quota"
+
+	// monitorQuotaFetchCacheTTL 配额快照缓存时长。多个监控可能关联同一账号，
+	// 而 interval 最小 15s 且国产配额服务无缓存，TTL 防止打爆上游配额端点。
+	monitorQuotaFetchCacheTTL = 5 * time.Minute
+	// monitorQuotaDegradedUsedPercent 任一用量窗口使用率超过该阈值时，
+	// 配额检查状态记为 degraded（对齐账号页展示阈值）。
+	monitorQuotaDegradedUsedPercent = 90.0
 
 	// MonitorDefaultGrokModel 是新增 Grok 监控未显式指定模型时使用的轻量测活模型。
 	MonitorDefaultGrokModel = "grok-4.5"
@@ -118,7 +145,16 @@ var (
 		"CHANNEL_MONITOR_NOT_FOUND", "channel monitor not found",
 	)
 	ErrChannelMonitorInvalidProvider = infraerrors.BadRequest(
-		"CHANNEL_MONITOR_INVALID_PROVIDER", "provider must be one of openai/anthropic/gemini/grok",
+		"CHANNEL_MONITOR_INVALID_PROVIDER", "provider must be one of openai/anthropic/gemini/grok/antigravity/kimi/zhipu/deepseek",
+	)
+	ErrChannelMonitorInvalidCheckMode = infraerrors.BadRequest(
+		"CHANNEL_MONITOR_INVALID_CHECK_MODE", "check_mode must be one of probe/quota/quota_probe; antigravity only supports quota",
+	)
+	ErrChannelMonitorAccountRequired = infraerrors.BadRequest(
+		"CHANNEL_MONITOR_ACCOUNT_REQUIRED", "account_id is required for quota-based check_mode",
+	)
+	ErrChannelMonitorProviderIncompatible = infraerrors.BadRequest(
+		"CHANNEL_MONITOR_PROVIDER_INCOMPATIBLE", "monitor provider must match the linked account platform",
 	)
 	ErrChannelMonitorInvalidAPIMode = infraerrors.BadRequest(
 		"CHANNEL_MONITOR_INVALID_API_MODE", "api_mode must be chat_completions or responses; responses is only supported for openai",
diff --git a/backend/internal/service/channel_monitor_quota_fetcher.go b/backend/internal/service/channel_monitor_quota_fetcher.go
new file mode 100644
index 000000000000..83d8e0f45527
--- /dev/null
+++ b/backend/internal/service/channel_monitor_quota_fetcher.go
@@ -0,0 +1,429 @@
+package service
+
+import (
+	"context"
+	"fmt"
+	"log/slog"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/domain"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/xai"
+)
+
+// 渠道监控「配额模式」的配额抓取器。
+//
+// 不直接对接上游，而是把账号侧现成的用量服务归一成 domain.MonitorQuotaSnapshot：
+//   - 海外 5 家（anthropic/openai/gemini/antigravity/grok）→ AccountUsageService.GetUsage
+//   - 国产 coding plan（kimi/zhipu/deepseek）→ CNProviderQuotaService.QueryUsage
+//   - 国产 payg（kimi/deepseek）→ CNProviderBalanceService.QueryBalance
+//     （zhipu payg 无公开余额端点，QueryBalance 会返回该错误，原样透出）
+//
+// Fetch 永不返回 error：所有失败都降级为 Success=false 的快照照常入库，
+// 由 deriveQuotaCheckResult 推导为 failed/error 状态。
+//
+// 多个监控可能关联同一账号，而 interval 最小 15s 且国产配额服务自身无缓存，
+// 所以成功快照统一带 monitorQuotaFetchCacheTTL 缓存，防止打爆上游配额端点。
+
+// monitorUsageSource 海外平台账号用量查询（AccountUsageService 天然满足）。
+type monitorUsageSource interface {
+	GetUsage(ctx context.Context, accountID int64, force ...bool) (*UsageInfo, error)
+}
+
+// monitorCNQuotaSource 国产 coding plan 滚动窗口额度探测（CNProviderQuotaService 天然满足）。
+type monitorCNQuotaSource interface {
+	QueryUsage(ctx context.Context, accountID int64) (*CNProviderQuotaProbeResult, error)
+}
+
+// monitorCNBalanceSource 国产 payg 余额探测（CNProviderBalanceService 天然满足）。
+type monitorCNBalanceSource interface {
+	QueryBalance(ctx context.Context, accountID int64) (*CNProviderBalanceResult, error)
+}
+
+// monitorAccountSource 账号加载（AccountRepository 天然满足）。
+type monitorAccountSource interface {
+	GetByID(ctx context.Context, id int64) (*Account, error)
+}
+
+// ChannelMonitorQuotaFetcher 配额抓取器（带成功快照 TTL 缓存）。
+type ChannelMonitorQuotaFetcher struct {
+	usage     monitorUsageSource
+	cnQuota   monitorCNQuotaSource
+	cnBalance monitorCNBalanceSource
+	accounts  monitorAccountSource
+
+	mu    sync.Mutex
+	cache map[int64]monitorQuotaCacheEntry
+}
+
+type monitorQuotaCacheEntry struct {
+	snapshot *domain.MonitorQuotaSnapshot
+	expiry   time.Time
+}
+
+// NewChannelMonitorQuotaFetcher 构造配额抓取器。
+func NewChannelMonitorQuotaFetcher(
+	usage monitorUsageSource,
+	cnQuota monitorCNQuotaSource,
+	cnBalance monitorCNBalanceSource,
+	accounts monitorAccountSource,
+) *ChannelMonitorQuotaFetcher {
+	return &ChannelMonitorQuotaFetcher{
+		usage:     usage,
+		cnQuota:   cnQuota,
+		cnBalance: cnBalance,
+		accounts:  accounts,
+		cache:     make(map[int64]monitorQuotaCacheEntry),
+	}
+}
+
+// LoadAccount 加载账号（不走缓存）。供 Create/Update 时校验
+// provider 与 account.platform 一致；账号不存在时返回错误。
+func (f *ChannelMonitorQuotaFetcher) LoadAccount(ctx context.Context, id int64) (*Account, error) {
+	if f == nil || f.accounts == nil {
+		return nil, fmt.Errorf("quota fetcher is not configured")
+	}
+	return f.accounts.GetByID(ctx, id)
+}
+
+// Fetch 抓取账号的最新配额快照。永不返回 error：失败降级为
+// Success=false 快照（Error 带摘要），保证检测历史的时间线连续。
+func (f *ChannelMonitorQuotaFetcher) Fetch(ctx context.Context, accountID int64) *domain.MonitorQuotaSnapshot {
+	now := time.Now()
+
+	if cached, ok := f.cachedSnapshot(accountID, now); ok {
+		return cached
+	}
+
+	snapshot := f.fetchUncached(ctx, accountID, now)
+	if snapshot.Success {
+		f.storeSnapshot(accountID, snapshot, now.Add(monitorQuotaFetchCacheTTL))
+	}
+	return snapshot
+}
+
+func (f *ChannelMonitorQuotaFetcher) cachedSnapshot(accountID int64, now time.Time) (*domain.MonitorQuotaSnapshot, bool) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	entry, ok := f.cache[accountID]
+	if !ok || now.After(entry.expiry) {
+		return nil, false
+	}
+	return entry.snapshot, true
+}
+
+func (f *ChannelMonitorQuotaFetcher) storeSnapshot(accountID int64, snapshot *domain.MonitorQuotaSnapshot, expiry time.Time) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.cache[accountID] = monitorQuotaCacheEntry{snapshot: snapshot, expiry: expiry}
+}
+
+func (f *ChannelMonitorQuotaFetcher) fetchUncached(ctx context.Context, accountID int64, now time.Time) *domain.MonitorQuotaSnapshot {
+	if f == nil {
+		return quotaErrorSnapshot("usage", "quota fetcher is not configured", now)
+	}
+
+	account, err := f.LoadAccount(ctx, accountID)
+	if err != nil || account == nil {
+		// FK ON DELETE SET NULL 后 account_id 可能为空/失效；显式报「账号未关联」，
+		// 推导为 degraded（配置问题，不是渠道故障）。
+		slog.Warn("channel_monitor: load linked account failed",
+			"account_id", accountID, "error", err)
+		return quotaErrorSnapshot("usage", "linked account not found", now)
+	}
+
+	switch account.Platform {
+	case domain.PlatformKimi, domain.PlatformZhipu, domain.PlatformDeepseek:
+		if account.IsCodingPlan() {
+			return f.fetchCNQuota(ctx, accountID, now)
+		}
+		return f.fetchCNBalance(ctx, accountID, now)
+	default:
+		return f.fetchUsage(ctx, accountID, now)
+	}
+}
+
+// fetchUsage 海外平台：AccountUsageService.GetUsage → 快照。
+func (f *ChannelMonitorQuotaFetcher) fetchUsage(ctx context.Context, accountID int64, now time.Time) *domain.MonitorQuotaSnapshot {
+	if f.usage == nil {
+		return quotaErrorSnapshot("usage", "usage service is not configured", now)
+	}
+	usage, err := f.usage.GetUsage(ctx, accountID)
+	if err != nil {
+		msg := truncateMessage(sanitizeErrorMessage(err.Error()))
+		return &domain.MonitorQuotaSnapshot{
+			Source:            "usage",
+			Success:           false,
+			CredentialInvalid: isCredentialErrorMessage(msg),
+			Error:             msg,
+			FetchedAt:         now,
+		}
+	}
+	snapshot := &domain.MonitorQuotaSnapshot{
+		Source:          "usage",
+		Success:         true,
+		PlanLevel:       usage.SubscriptionTier,
+		Tiers:           usageQuotaTiers(usage),
+		FetchedAt:       now,
+	}
+	if snapshot.PlanLevel == "" {
+		snapshot.PlanLevel = usage.SubscriptionTierRaw
+	}
+	return snapshot
+}
+
+// usageQuotaTiers 把 UsageInfo 的各平台窗口归一为 tier 列表（无数据的窗口跳过）。
+func usageQuotaTiers(usage *UsageInfo) []domain.MonitorQuotaTier {
+	if usage == nil {
+		return nil
+	}
+	tiers := make([]domain.MonitorQuotaTier, 0, 8)
+	appendProgressTier(&tiers, "5h", "", usage.FiveHour)
+	appendProgressTier(&tiers, "7d", "", usage.SevenDay)
+	appendProgressTier(&tiers, "7d-sonnet", "", usage.SevenDaySonnet)
+	appendProgressTier(&tiers, "7d-fable", "", usage.SevenDayFable)
+	appendProgressTier(&tiers, "30d", "", usage.ThirtyDay)
+	// Gemini 多档日配额：同 Window 不同 Label。
+	appendProgressTier(&tiers, "daily", "shared", usage.GeminiSharedDaily)
+	appendProgressTier(&tiers, "daily", "pro", usage.GeminiProDaily)
+	appendProgressTier(&tiers, "daily", "flash", usage.GeminiFlashDaily)
+	// Grok requests/tokens 两个日窗口 + 月度计费窗口。
+	appendQuotaWindowTier(&tiers, "daily", "requests", usage.GrokRequestQuota)
+	appendQuotaWindowTier(&tiers, "daily", "tokens", usage.GrokTokenQuota)
+	// Antigravity per-model 总量额度，Label = 模型名（按名排序保证输出稳定）。
+	for _, model := range sortedQuotaModelNames(usage.AntigravityQuota) {
+		q := usage.AntigravityQuota[model]
+		if q == nil {
+			continue
+		}
+		tiers = append(tiers, domain.MonitorQuotaTier{
+			Window:      "total",
+			Label:       model,
+			UsedPercent: float64(q.Utilization),
+			ResetAt:     q.ResetTime,
+		})
+	}
+	if len(tiers) == 0 {
+		return nil
+	}
+	return tiers
+}
+
+func appendProgressTier(tiers *[]domain.MonitorQuotaTier, window, label string, p *UsageProgress) {
+	if p == nil {
+		return
+	}
+	tier := domain.MonitorQuotaTier{
+		Window:      window,
+		Label:       label,
+		UsedPercent: p.Utilization,
+	}
+	if p.ResetsAt != nil {
+		tier.ResetAt = p.ResetsAt.UTC().Format(time.RFC3339)
+	}
+	if p.LimitRequests > 0 {
+		tier.Used = float64(p.UsedRequests)
+		tier.Limit = float64(p.LimitRequests)
+	}
+	*tiers = append(*tiers, tier)
+}
+
+func appendQuotaWindowTier(tiers *[]domain.MonitorQuotaTier, window, label string, q *xai.QuotaWindow) {
+	if q == nil || q.Limit == nil || *q.Limit <= 0 {
+		return
+	}
+	used := *q.Limit
+	if q.Remaining != nil {
+		used = *q.Limit - *q.Remaining
+		if used < 0 {
+			used = 0
+		}
+	}
+	tier := domain.MonitorQuotaTier{
+		Window:      window,
+		Label:       label,
+		Used:        float64(used),
+		Limit:       float64(*q.Limit),
+		UsedPercent: float64(used) / float64(*q.Limit) * 100,
+	}
+	if q.ResetAt != "" {
+		tier.ResetAt = q.ResetAt
+	} else if q.ResetUnix != nil && *q.ResetUnix > 0 {
+		tier.ResetAt = time.Unix(*q.ResetUnix, 0).UTC().Format(time.RFC3339)
+	}
+	*tiers = append(*tiers, tier)
+}
+
+func sortedQuotaModelNames(quotas map[string]*AntigravityModelQuota) []string {
+	names := make([]string, 0, len(quotas))
+	for name := range quotas {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	return names
+}
+
+// fetchCNQuota 国产 coding plan：CNProviderQuotaService.QueryUsage → 快照。
+func (f *ChannelMonitorQuotaFetcher) fetchCNQuota(ctx context.Context, accountID int64, now time.Time) *domain.MonitorQuotaSnapshot {
+	if f.cnQuota == nil {
+		return quotaErrorSnapshot("cn_quota", "cn quota service is not configured", now)
+	}
+	result, err := f.cnQuota.QueryUsage(ctx, accountID)
+	if err != nil {
+		msg := truncateMessage(sanitizeErrorMessage(err.Error()))
+		return &domain.MonitorQuotaSnapshot{
+			Source:            "cn_quota",
+			Success:           false,
+			CredentialInvalid: isCredentialErrorMessage(msg),
+			Error:             msg,
+			FetchedAt:         now,
+		}
+	}
+	snapshot := &domain.MonitorQuotaSnapshot{
+		Source:    "cn_quota",
+		Success:   result.Success,
+		PlanLevel: result.PlanLevel,
+		Error:     result.Error,
+		FetchedAt: now,
+	}
+	if !result.Success && !result.CredentialValid {
+		snapshot.CredentialInvalid = true
+	}
+	if len(result.Tiers) > 0 {
+		snapshot.Tiers = make([]domain.MonitorQuotaTier, 0, len(result.Tiers))
+		for _, t := range result.Tiers {
+			snapshot.Tiers = append(snapshot.Tiers, domain.MonitorQuotaTier{
+				Window:      t.Window,
+				UsedPercent: t.UsedPercent,
+				ResetAt:     t.ResetAt,
+			})
+		}
+	}
+	if !snapshot.Success {
+		snapshot.Error = firstNonEmpty(snapshot.Error, "cn quota probe failed")
+	}
+	return snapshot
+}
+
+// fetchCNBalance 国产 payg：CNProviderBalanceService.QueryBalance → 快照。
+func (f *ChannelMonitorQuotaFetcher) fetchCNBalance(ctx context.Context, accountID int64, now time.Time) *domain.MonitorQuotaSnapshot {
+	if f.cnBalance == nil {
+		return quotaErrorSnapshot("cn_balance", "cn balance service is not configured", now)
+	}
+	result, err := f.cnBalance.QueryBalance(ctx, accountID)
+	if err != nil {
+		msg := truncateMessage(sanitizeErrorMessage(err.Error()))
+		return &domain.MonitorQuotaSnapshot{
+			Source:            "cn_balance",
+			Success:           false,
+			CredentialInvalid: isCredentialErrorMessage(msg),
+			Error:             msg,
+			FetchedAt:         now,
+		}
+	}
+	snapshot := &domain.MonitorQuotaSnapshot{
+		Source:    "cn_balance",
+		Success:   result.Success,
+		Currency:  result.Currency,
+		Error:     result.Error,
+		FetchedAt: now,
+	}
+	if result.Success {
+		balance := result.Balance
+		snapshot.Balance = &balance
+	} else if result.StatusCode == 401 || result.StatusCode == 403 {
+		snapshot.CredentialInvalid = true
+	}
+	if len(result.Balances) > 0 {
+		snapshot.Balances = make([]domain.MonitorBalance, 0, len(result.Balances))
+		for _, b := range result.Balances {
+			snapshot.Balances = append(snapshot.Balances, domain.MonitorBalance{
+				Currency: b.Currency,
+				Balance:  b.Balance,
+			})
+		}
+	}
+	if !snapshot.Success {
+		snapshot.Error = firstNonEmpty(snapshot.Error, "cn balance probe failed")
+	}
+	return snapshot
+}
+
+// quotaErrorSnapshot 构造统一错误快照。
+func quotaErrorSnapshot(source, message string, now time.Time) *domain.MonitorQuotaSnapshot {
+	return &domain.MonitorQuotaSnapshot{
+		Source:    source,
+		Success:   false,
+		Error:     truncateMessage(sanitizeErrorMessage(message)),
+		FetchedAt: now,
+	}
+}
+
+// isCredentialErrorMessage 上游 401/403 鉴权失败的启发式识别
+// （海外 GetUsage 的错误没有结构化状态码，只能看文本）。
+func isCredentialErrorMessage(msg string) bool {
+	msg = strings.ToLower(msg)
+	return strings.Contains(msg, "401") ||
+		strings.Contains(msg, "403") ||
+		strings.Contains(msg, "unauthorized") ||
+		strings.Contains(msg, "forbidden") ||
+		strings.Contains(msg, "invalid_api_key") ||
+		strings.Contains(msg, "authentication")
+}
+
+// deriveQuotaCheckResult 把配额快照推导为检测状态（复用既有 status 枚举，
+// 时间线/可用率机制自动生效）：
+//   - 查询成功且无告警        → operational
+//   - 任一窗口使用率 >= 阈值或余额耗尽 → degraded
+//   - 账号未关联（配置问题）    → degraded
+//   - 凭据失效（401/403）     → failed
+//   - 网络/解析等其他错误      → error
+func deriveQuotaCheckResult(snapshot *domain.MonitorQuotaSnapshot, model string, checkedAt time.Time) *CheckResult {
+	res := &CheckResult{Model: model, CheckedAt: checkedAt}
+	if snapshot == nil {
+		res.Status = MonitorStatusError
+		res.Message = "quota snapshot missing"
+		return res
+	}
+
+	switch {
+	case !snapshot.Success && snapshot.CredentialInvalid:
+		res.Status = MonitorStatusFailed
+		res.Message = snapshot.Error
+	case !snapshot.Success && strings.Contains(snapshot.Error, "linked account not found"):
+		res.Status = MonitorStatusDegraded
+		res.Message = snapshot.Error
+	case !snapshot.Success:
+		res.Status = MonitorStatusError
+		res.Message = snapshot.Error
+	default:
+		if hint := quotaDegradedHint(snapshot); hint != "" {
+			res.Status = MonitorStatusDegraded
+			res.Message = hint
+		} else {
+			res.Status = MonitorStatusOperational
+		}
+	}
+	return res
+}
+
+// quotaDegradedHint 生成 degraded 的 message（指出触发告警的窗口/余额）；
+// 空串表示无告警。
+func quotaDegradedHint(snapshot *domain.MonitorQuotaSnapshot) string {
+	for _, tier := range snapshot.Tiers {
+		if tier.UsedPercent >= monitorQuotaDegradedUsedPercent {
+			name := tier.Window
+			if tier.Label != "" {
+				name = tier.Label + "/" + tier.Window
+			}
+			return fmt.Sprintf("quota high: %s at %s%%", name, strconv.FormatFloat(tier.UsedPercent, 'f', 1, 64))
+		}
+	}
+	if snapshot.Balance != nil && *snapshot.Balance <= 0 {
+		return fmt.Sprintf("balance depleted (%s)", firstNonEmpty(snapshot.Currency, "?"))
+	}
+	return ""
+}
diff --git a/backend/internal/service/channel_monitor_service.go b/backend/internal/service/channel_monitor_service.go
index 248b72133dfc..9b0bb0763980 100644
--- a/backend/internal/service/channel_monitor_service.go
+++ b/backend/internal/service/channel_monitor_service.go
@@ -11,6 +11,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/domain"
 	"golang.org/x/sync/errgroup"
 )
 
@@ -77,6 +78,10 @@ type ChannelMonitorService struct {
 	// scheduler 由 wire 通过 SetScheduler 注入；CRUD 后调用对应钩子即时同步任务。
 	// 测试或未注入场景下保持 nil，所有钩子调用变为 no-op。
 	scheduler MonitorScheduler
+	// quotaFetcher 由 wire 通过 SetQuotaFetcher 注入（accountUsage/CN 服务在本服务
+	// 之后构造，构造参数注入会破坏既有依赖顺序）。nil 时 fail-closed：
+	// 配额模式的检测产出「未配置」错误快照，Create/Update 关联账号直接报错。
+	quotaFetcher *ChannelMonitorQuotaFetcher
 }
 
 const maxChannelMonitorNameRunes = 100
@@ -150,6 +155,10 @@ func (s *ChannelMonitorService) Create(ctx context.Context, p ChannelMonitorCrea
 	if err := validateExtraHeaders(p.ExtraHeaders); err != nil {
 		return nil, err
 	}
+	if err := s.validateLinkedAccount(ctx, p.Provider, p.AccountID); err != nil {
+		return nil, err
+	}
+	checkMode := defaultCheckMode(p.CheckMode)
 	encrypted, err := s.encryptor.Encrypt(p.APIKey)
 	if err != nil {
 		return nil, fmt.Errorf("encrypt api key: %w", err)
@@ -160,7 +169,7 @@ func (s *ChannelMonitorService) Create(ctx context.Context, p ChannelMonitorCrea
 		APIMode:          defaultAPIMode(p.APIMode),
 		Endpoint:         normalizeEndpoint(p.Endpoint),
 		APIKey:           encrypted, // 注意：传入 repository 时该字段为密文
-		PrimaryModel:     normalizeMonitorPrimaryModel(p.Provider, p.PrimaryModel),
+		PrimaryModel:     normalizeMonitorPrimaryModel(p.Provider, checkMode, p.PrimaryModel),
 		ExtraModels:      normalizeModels(p.ExtraModels),
 		GroupName:        strings.TrimSpace(p.GroupName),
 		Enabled:          p.Enabled,
@@ -171,6 +180,8 @@ func (s *ChannelMonitorService) Create(ctx context.Context, p ChannelMonitorCrea
 		ExtraHeaders:     emptyHeadersIfNil(p.ExtraHeaders),
 		BodyOverrideMode: defaultBodyMode(p.BodyOverrideMode),
 		BodyOverride:     p.BodyOverride,
+		CheckMode:        checkMode,
+		AccountID:        cloneInt64Pointer(p.AccountID),
 	}
 	if err := s.repo.Create(ctx, m); err != nil {
 		return nil, fmt.Errorf("create channel monitor: %w", err)
@@ -236,6 +247,8 @@ func (s *ChannelMonitorService) Duplicate(
 		ExtraHeaders:         cloneChannelMonitorHeaders(source.ExtraHeaders),
 		BodyOverrideMode:     source.BodyOverrideMode,
 		BodyOverride:         bodyOverride,
+		CheckMode:            defaultCheckMode(source.CheckMode),
+		AccountID:            cloneInt64Pointer(source.AccountID),
 		DuplicateOperationID: operationID,
 	}
 	if err := s.repo.Create(ctx, duplicate); err != nil {
@@ -290,11 +303,22 @@ func (s *ChannelMonitorService) decryptAPIKeyForDuplicate(source *ChannelMonitor
 		return "", ErrChannelMonitorAPIKeyDecryptFailed
 	}
 	plain, err := s.encryptor.Decrypt(source.APIKey)
-	if err != nil || strings.TrimSpace(plain) == "" {
+	if err != nil {
 		slog.Warn("channel_monitor: decrypt api key for duplicate failed",
 			"monitor_id", source.ID, "error", err)
 		return "", ErrChannelMonitorAPIKeyDecryptFailed
 	}
+	// quota 模式明文为空串是合法状态（api_key_encrypted 存的是加密空串）：
+	// 重加密空串即可。若在此报错，克隆出的配额监控会被 runner 当作
+	// 解密失败而 Unschedule，静默停摆。
+	if strings.TrimSpace(plain) == "" {
+		if monitorCheckModeUsesQuota(defaultCheckMode(source.CheckMode)) {
+			return "", nil
+		}
+		slog.Warn("channel_monitor: decrypted api key for duplicate is empty",
+			"monitor_id", source.ID)
+		return "", ErrChannelMonitorAPIKeyDecryptFailed
+	}
 	return plain, nil
 }
 
@@ -343,10 +367,16 @@ func cloneChannelMonitorJSONMap(source map[string]any) (map[string]any, error) {
 }
 
 // validateCreateParams 把 Create 入参的所有校验聚拢为一个函数，避免 Create 主体超过 30 行。
+// 按 check_mode 分支：probe 沿用 endpoint+api_key 必填；quota 只需关联账号；
+// quota_probe 两者皆需。
 func validateCreateParams(p ChannelMonitorCreateParams) error {
 	if err := validateProvider(p.Provider); err != nil {
 		return err
 	}
+	checkMode := defaultCheckMode(p.CheckMode)
+	if err := validateCheckMode(p.Provider, checkMode); err != nil {
+		return err
+	}
 	if err := validateAPIMode(p.Provider, p.APIMode); err != nil {
 		return err
 	}
@@ -356,18 +386,45 @@ func validateCreateParams(p ChannelMonitorCreateParams) error {
 	if err := validateJitter(p.JitterSeconds, p.IntervalSeconds); err != nil {
 		return err
 	}
-	if err := validateEndpoint(p.Endpoint); err != nil {
-		return err
+	usesQuota := monitorCheckModeUsesQuota(checkMode)
+	// probe 分支（含 quota_probe 的探活部分）仍需 endpoint + api_key；
+	// quota 模式 endpoint/api_key 留空，避免要求用户填无意义的占位值。
+	if checkMode != MonitorCheckModeQuota {
+		if err := validateEndpoint(p.Endpoint); err != nil {
+			return err
+		}
+		if strings.TrimSpace(p.APIKey) == "" {
+			return ErrChannelMonitorMissingAPIKey
+		}
 	}
-	if strings.TrimSpace(p.APIKey) == "" {
-		return ErrChannelMonitorMissingAPIKey
+	if usesQuota && (p.AccountID == nil || *p.AccountID <= 0) {
+		return ErrChannelMonitorAccountRequired
 	}
-	if normalizeMonitorPrimaryModel(p.Provider, p.PrimaryModel) == "" {
+	if normalizeMonitorPrimaryModel(p.Provider, checkMode, p.PrimaryModel) == "" {
 		return ErrChannelMonitorMissingPrimaryModel
 	}
 	return nil
 }
 
+// validateLinkedAccount 校验关联账号存在且平台与监控 provider 一致。
+// fetcher 未注入时 fail-closed（拒绝创建配额监控，而不是创建后静默坏）。
+func (s *ChannelMonitorService) validateLinkedAccount(ctx context.Context, provider string, accountID *int64) error {
+	if accountID == nil || *accountID <= 0 {
+		return nil
+	}
+	if s.quotaFetcher == nil {
+		return ErrChannelMonitorAccountRequired
+	}
+	account, err := s.quotaFetcher.LoadAccount(ctx, *accountID)
+	if err != nil || account == nil {
+		return ErrChannelMonitorAccountRequired
+	}
+	if account.Platform != provider {
+		return ErrChannelMonitorProviderIncompatible
+	}
+	return nil
+}
+
 // Update 更新监控。APIKey 字段：nil 或空字符串 = 不修改；非空 = 加密后覆盖。
 func (s *ChannelMonitorService) Update(ctx context.Context, id int64, p ChannelMonitorUpdateParams) (*ChannelMonitor, error) {
 	existing, err := s.repo.GetByID(ctx, id)
@@ -382,6 +439,14 @@ func (s *ChannelMonitorService) Update(ctx context.Context, id int64, p ChannelM
 	if err != nil {
 		return nil, err
 	}
+	if err := s.validateProbeAPIKey(existing, newPlainAPIKey); err != nil {
+		return nil, err
+	}
+	if p.Provider != nil || p.CheckMode != nil || p.AccountID != nil {
+		if err := s.revalidateLinkedAccount(ctx, existing); err != nil {
+			return nil, err
+		}
+	}
 
 	if err := s.repo.Update(ctx, existing); err != nil {
 		return nil, fmt.Errorf("update channel monitor: %w", err)
@@ -401,6 +466,76 @@ func (s *ChannelMonitorService) Update(ctx context.Context, id int64, p ChannelM
 	return existing, nil
 }
 
+// validateMonitorModeFields 校验 check_mode 与其它字段的组合约束
+// （在 provider/check_mode/account_id/endpoint 全部应用后调用）：
+//   - quota / quota_probe 必须关联账号
+//   - probe / quota_probe 必须持有 endpoint（探活目标）
+func validateMonitorModeFields(m *ChannelMonitor) error {
+	checkMode := defaultCheckMode(m.CheckMode)
+	if monitorCheckModeUsesQuota(checkMode) && m.AccountID == nil {
+		return ErrChannelMonitorAccountRequired
+	}
+	if checkMode != MonitorCheckModeQuota && strings.TrimSpace(m.Endpoint) == "" {
+		return ErrChannelMonitorInvalidEndpoint
+	}
+	return nil
+}
+
+// validateProbeAPIKey 探活模式（probe / quota_probe）必须持有可用明文 key：
+// 存量密文解密为空串（quota 监控切回探活但未重填 key）时拒绝。
+// 密文损坏的情况交给既有 APIKeyDecryptFailed 链路（Get/RunCheck 会显式报错）。
+func (s *ChannelMonitorService) validateProbeAPIKey(m *ChannelMonitor, newPlainKey string) error {
+	if defaultCheckMode(m.CheckMode) == MonitorCheckModeQuota {
+		return nil
+	}
+	if strings.TrimSpace(newPlainKey) != "" {
+		return nil
+	}
+	if strings.TrimSpace(m.APIKey) == "" {
+		return ErrChannelMonitorMissingAPIKey
+	}
+	plain, err := s.encryptor.Decrypt(m.APIKey)
+	if err != nil {
+		return nil
+	}
+	if strings.TrimSpace(plain) == "" {
+		return ErrChannelMonitorMissingAPIKey
+	}
+	return nil
+}
+
+// revalidateLinkedAccount 在 provider/check_mode/account_id 任一变化后复核关联账号：
+//   - 账号已被删除或平台失配：probe 模式自动解绑（静默修复），
+//     quota 模式显式报错（配额监控必须有可用数据源）
+func (s *ChannelMonitorService) revalidateLinkedAccount(ctx context.Context, m *ChannelMonitor) error {
+	usesQuota := monitorCheckModeUsesQuota(defaultCheckMode(m.CheckMode))
+	if m.AccountID == nil {
+		if usesQuota {
+			return ErrChannelMonitorAccountRequired
+		}
+		return nil
+	}
+	if s.quotaFetcher == nil {
+		return ErrChannelMonitorAccountRequired
+	}
+	account, err := s.quotaFetcher.LoadAccount(ctx, *m.AccountID)
+	if err != nil || account == nil {
+		if usesQuota {
+			return ErrChannelMonitorAccountRequired
+		}
+		m.AccountID = nil
+		return nil
+	}
+	if account.Platform != m.Provider {
+		if usesQuota {
+			return ErrChannelMonitorProviderIncompatible
+		}
+		m.AccountID = nil
+		return nil
+	}
+	return nil
+}
+
 // applyAPIKeyUpdate 处理 Update 中的 APIKey 字段：
 //   - 入参 raw 为 nil 或空白：不修改 existing.APIKey（仍为密文），返回 updated=false
 //   - 非空：加密后写入 existing.APIKey；同时把明文返回给调用方，
@@ -454,6 +589,9 @@ func (s *ChannelMonitorService) ListHistory(ctx context.Context, id int64, model
 // 写历史记录并更新 last_checked_at。返回每个模型的检测结果。
 // 仅当 channel_monitor_enabled=true 且 channel_monitor_mode=v1 时真正探测；
 // mode=v2 时返回 ErrChannelMonitorActiveProbesRetired，不产生上游流量。
+//
+// 按 check_mode 分派：probe（默认，现状探活）/ quota（仅查关联账号配额，
+// 零 LLM 成本）/ quota_probe（探活 + 配额快照挂主模型行）。
 func (s *ChannelMonitorService) RunCheck(ctx context.Context, id int64) ([]*CheckResult, error) {
 	rt := s.probeRuntime(ctx)
 	if !rt.Enabled {
@@ -466,14 +604,59 @@ func (s *ChannelMonitorService) RunCheck(ctx context.Context, id int64) ([]*Chec
 	if err != nil {
 		return nil, err
 	}
-	if m.APIKeyDecryptFailed {
+	checkMode := defaultCheckMode(m.CheckMode)
+	if checkMode != MonitorCheckModeQuota && m.APIKeyDecryptFailed {
 		return nil, ErrChannelMonitorAPIKeyDecryptFailed
 	}
-	results := s.runChecksConcurrent(ctx, m)
+
+	var results []*CheckResult
+	switch checkMode {
+	case MonitorCheckModeQuota:
+		results = s.runQuotaOnlyCheck(ctx, m)
+	case MonitorCheckModeQuotaProbe:
+		results = s.runChecksConcurrent(ctx, m)
+		attachQuotaSnapshot(results, s.fetchQuotaSnapshot(ctx, m))
+	default:
+		results = s.runChecksConcurrent(ctx, m)
+	}
 	s.persistCheckResults(ctx, m, results)
 	return results, nil
 }
 
+// runQuotaOnlyCheck quota 模式：一次配额抓取 → 单条 CheckResult
+// （Model=PrimaryModel，默认 "quota"；无 ping/latency，状态由快照推导）。
+func (s *ChannelMonitorService) runQuotaOnlyCheck(ctx context.Context, m *ChannelMonitor) []*CheckResult {
+	snapshot := s.fetchQuotaSnapshot(ctx, m)
+	res := deriveQuotaCheckResult(snapshot, m.PrimaryModel, time.Now())
+	res.Quota = snapshot
+	return []*CheckResult{res}
+}
+
+// fetchQuotaSnapshot 抓取关联账号配额。未关联账号 / fetcher 未注入时返回
+// 显式错误快照（不返回 error，保证检测周期与历史时间线连续）。
+func (s *ChannelMonitorService) fetchQuotaSnapshot(ctx context.Context, m *ChannelMonitor) *domain.MonitorQuotaSnapshot {
+	if m.AccountID == nil {
+		return quotaErrorSnapshot("usage", "linked account not found", time.Now())
+	}
+	if s.quotaFetcher == nil {
+		return quotaErrorSnapshot("usage", "quota fetcher is not configured", time.Now())
+	}
+	return s.quotaFetcher.Fetch(ctx, *m.AccountID)
+}
+
+// attachQuotaSnapshot quota_probe：把配额快照挂到主模型行（results[0]）。
+// 配额失败不改变探活状态，仅在探活 message 为空时附注失败原因。
+func attachQuotaSnapshot(results []*CheckResult, snapshot *domain.MonitorQuotaSnapshot) {
+	if len(results) == 0 || snapshot == nil {
+		return
+	}
+	primary := results[0]
+	primary.Quota = snapshot
+	if !snapshot.Success && strings.TrimSpace(primary.Message) == "" {
+		primary.Message = truncateMessage("quota fetch failed: " + snapshot.Error)
+	}
+}
+
 // persistCheckResults 写入本次检测的历史记录并更新 last_checked_at。
 // 任一写库失败都只记日志，不影响调用方拿到 results（与 MVP 期望一致：宁可漏记历史也要先返回结果）。
 func (s *ChannelMonitorService) persistCheckResults(ctx context.Context, m *ChannelMonitor, results []*CheckResult) {
@@ -487,6 +670,7 @@ func (s *ChannelMonitorService) persistCheckResults(ctx context.Context, m *Chan
 			PingLatencyMs: r.PingLatencyMs,
 			Message:       r.Message,
 			CheckedAt:     r.CheckedAt,
+			Quota:         r.Quota,
 		})
 	}
 	if err := s.repo.InsertHistoryBatch(ctx, rows); err != nil {
@@ -541,6 +725,14 @@ func (s *ChannelMonitorService) SetScheduler(sched MonitorScheduler) {
 	s.scheduler = sched
 }
 
+// SetQuotaFetcher 由 wire 注入配额抓取器（账号侧用量服务聚合）。
+func (s *ChannelMonitorService) SetQuotaFetcher(fetcher *ChannelMonitorQuotaFetcher) {
+	if s == nil {
+		return
+	}
+	s.quotaFetcher = fetcher
+}
+
 // ListEnabledMonitors 返回所有 enabled=true 的监控（解密后），供 runner 启动时建立任务表。
 func (s *ChannelMonitorService) ListEnabledMonitors(ctx context.Context) ([]*ChannelMonitor, error) {
 	all, err := s.repo.ListEnabled(ctx)
@@ -693,14 +885,36 @@ func applyMonitorUpdate(existing *ChannelMonitor, p ChannelMonitorUpdateParams)
 		providerChanged = existing.Provider != *p.Provider
 		existing.Provider = *p.Provider
 	}
-	if p.Endpoint != nil {
-		if err := validateEndpoint(*p.Endpoint); err != nil {
+	if p.CheckMode != nil {
+		mode := defaultCheckMode(*p.CheckMode)
+		if err := validateCheckMode(existing.Provider, mode); err != nil {
 			return err
 		}
+		existing.CheckMode = mode
+	}
+	if p.AccountID != nil {
+		if *p.AccountID > 0 {
+			id := *p.AccountID
+			existing.AccountID = &id
+		} else {
+			existing.AccountID = nil // 0 = 清空关联
+		}
+	}
+	if p.Endpoint != nil {
+		// quota 模式允许清空 endpoint（校验由 validateMonitorModeFields 兜底）。
+		if strings.TrimSpace(*p.Endpoint) != "" {
+			if err := validateEndpoint(*p.Endpoint); err != nil {
+				return err
+			}
+		}
 		existing.Endpoint = normalizeEndpoint(*p.Endpoint)
 	}
+	// 模式与字段的组合校验（provider/check_mode/account_id/endpoint 全部应用后）。
+	if err := validateMonitorModeFields(existing); err != nil {
+		return err
+	}
 	if p.PrimaryModel != nil {
-		primaryModel := normalizeMonitorPrimaryModel(existing.Provider, *p.PrimaryModel)
+		primaryModel := normalizeMonitorPrimaryModel(existing.Provider, defaultCheckMode(existing.CheckMode), *p.PrimaryModel)
 		if primaryModel == "" {
 			return ErrChannelMonitorMissingPrimaryModel
 		}
diff --git a/backend/internal/service/channel_monitor_types.go b/backend/internal/service/channel_monitor_types.go
index 20ab1da935a8..298e037ee118 100644
--- a/backend/internal/service/channel_monitor_types.go
+++ b/backend/internal/service/channel_monitor_types.go
@@ -1,6 +1,10 @@
 package service
 
-import "time"
+import (
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/domain"
+)
 
 // MonitorBodyOverrideMode 自定义请求体处理模式。
 //
@@ -45,6 +49,11 @@ type ChannelMonitor struct {
 	CreatedAt       time.Time
 	UpdatedAt       time.Time
 
+	// 配额模式（check_mode = quota / quota_probe）：
+	// 关联已有账号复用账号侧用量服务，Endpoint/APIKey 可为空（quota 模式）。
+	CheckMode string  // probe（默认）/ quota / quota_probe；空串按 probe 处理
+	AccountID *int64  // 关联账号 ID；账号删除后被 DB 置空（监控保留并报「账号未关联」）
+
 	// 请求自定义快照（来自模板拷贝 or 用户手填，运行时直接读取）
 	TemplateID       *int64            // 仅用于 UI 分组 + 一键应用，运行时不用
 	ExtraHeaders     map[string]string // 与 adapter 默认 headers 合并，用户优先
@@ -89,6 +98,10 @@ type ChannelMonitorCreateParams struct {
 	ExtraHeaders     map[string]string
 	BodyOverrideMode string
 	BodyOverride     map[string]any
+
+	// 配额模式：CheckMode 空串默认 probe；quota/quota_probe 必须关联账号。
+	CheckMode string
+	AccountID *int64
 }
 
 // ChannelMonitorUpdateParams 更新参数（指针字段表示"未提供则不更新"）。
@@ -112,6 +125,11 @@ type ChannelMonitorUpdateParams struct {
 	ExtraHeaders     *map[string]string
 	BodyOverrideMode *string
 	BodyOverride     *map[string]any
+
+	// 配额模式：CheckMode nil = 不更新；AccountID nil = 不更新，
+	// 指向 0 = 清空关联（退回 probe 模式时由 CheckMode 分支兜底）。
+	CheckMode *string
+	AccountID *int64
 }
 
 // CheckResult 单个模型一次检测的结果。
@@ -122,6 +140,8 @@ type CheckResult struct {
 	PingLatencyMs *int
 	Message       string
 	CheckedAt     time.Time
+	// Quota 配额模式附带快照（quota 模式唯一数据；quota_probe 挂在主模型行）。
+	Quota *domain.MonitorQuotaSnapshot
 }
 
 // UserMonitorView 用户只读视图：监控概览（含主模型最近状态 + 7d 可用率 + 附加模型最近状态）。
@@ -137,6 +157,9 @@ type UserMonitorView struct {
 	Availability7d       float64 // 0-100
 	ExtraModels          []ExtraModelStatus
 	Timeline             []UserMonitorTimelinePoint // 主模型最近 N 个历史点（按 checked_at DESC，最新在前）
+	// LatestQuota 主模型最近一次配额快照；channel_monitor_show_quota=false
+	// 时由 handler 服务端剥离。
+	LatestQuota *domain.MonitorQuotaSnapshot
 }
 
 // UserMonitorTimelinePoint 用户视图 timeline 单点数据（去除 message 以减小响应体）。
@@ -183,6 +206,7 @@ type ChannelMonitorHistoryRow struct {
 	PingLatencyMs *int
 	Message       string
 	CheckedAt     time.Time
+	Quota         *domain.MonitorQuotaSnapshot
 }
 
 // ChannelMonitorHistoryEntry 历史记录查询返回行（含 ent 主键 ID）。
@@ -194,6 +218,7 @@ type ChannelMonitorHistoryEntry struct {
 	PingLatencyMs *int
 	Message       string
 	CheckedAt     time.Time
+	Quota         *domain.MonitorQuotaSnapshot
 }
 
 // ChannelMonitorLatest 最近一次检测的简明信息（用于 UserMonitorView 聚合）。
@@ -203,6 +228,7 @@ type ChannelMonitorLatest struct {
 	LatencyMs     *int
 	PingLatencyMs *int
 	CheckedAt     time.Time
+	Quota         *domain.MonitorQuotaSnapshot
 }
 
 // ChannelMonitorAvailability 单个模型在某窗口内的可用率与平均延迟（用于 UserMonitorDetail 聚合）。
@@ -223,4 +249,5 @@ type MonitorStatusSummary struct {
 	PrimaryLatencyMs *int
 	Availability7d   float64 // 0-100，无历史时为 0
 	ExtraModels      []ExtraModelStatus
+	LatestQuota      *domain.MonitorQuotaSnapshot // 主模型最近配额快照（配额模式）
 }
diff --git a/backend/internal/service/channel_monitor_validate.go b/backend/internal/service/channel_monitor_validate.go
index 7740dc83b77d..faf7247fbc8a 100644
--- a/backend/internal/service/channel_monitor_validate.go
+++ b/backend/internal/service/channel_monitor_validate.go
@@ -9,15 +9,81 @@ import (
 // 渠道监控参数校验与归一化辅助函数。
 // 校验失败一律返回 channel_monitor_const.go 中预定义的 Err* 错误，错误信息不含具体 IP/hostname，避免泄露内网拓扑。
 
+// monitorProviders 渠道监控支持的全部 provider（与迁移 226 的 CHECK 约束一致）。
+// 不再以 adapter 表为唯一来源：antigravity 没有探活 adapter，但支持配额模式。
+//
+//nolint:gochecknoglobals // 静态查表，初始化后不变。
+var monitorProviders = map[string]struct{}{
+	MonitorProviderOpenAI:      {},
+	MonitorProviderAnthropic:   {},
+	MonitorProviderGemini:      {},
+	MonitorProviderGrok:        {},
+	MonitorProviderAntigravity: {},
+	MonitorProviderKimi:        {},
+	MonitorProviderZhipu:       {},
+	MonitorProviderDeepseek:    {},
+}
+
+// probeCapableProviders 支持探活（probe / quota_probe）的 provider。
+// antigravity 上游无 Chat/Responses 可打（仅 IDE 代理形态），只允许配额模式。
+//
+//nolint:gochecknoglobals // 静态查表，初始化后不变。
+var probeCapableProviders = map[string]struct{}{
+	MonitorProviderOpenAI:    {},
+	MonitorProviderAnthropic: {},
+	MonitorProviderGemini:    {},
+	MonitorProviderGrok:      {},
+	MonitorProviderKimi:      {},
+	MonitorProviderZhipu:     {},
+	MonitorProviderDeepseek:  {},
+}
+
 // validateProvider 校验 provider 字符串。
-// 唯一来源于 providerAdapters：新增 provider 只需要在 channel_monitor_checker.go 注册 adapter。
 func validateProvider(p string) error {
-	if !isSupportedProvider(p) {
+	if _, ok := monitorProviders[p]; !ok {
 		return ErrChannelMonitorInvalidProvider
 	}
 	return nil
 }
 
+// providerSupportsProbe 该 provider 是否注册了探活 adapter（antigravity 为 false）。
+func providerSupportsProbe(p string) bool {
+	_, ok := probeCapableProviders[p]
+	return ok
+}
+
+// defaultCheckMode 空串归一为 probe，保证存量数据与旧客户端兼容。
+func defaultCheckMode(checkMode string) string {
+	if strings.TrimSpace(checkMode) == "" {
+		return MonitorCheckModeProbe
+	}
+	return strings.TrimSpace(checkMode)
+}
+
+// monitorCheckModeUsesQuota 该模式是否需要关联账号查配额。
+func monitorCheckModeUsesQuota(checkMode string) bool {
+	return checkMode == MonitorCheckModeQuota || checkMode == MonitorCheckModeQuotaProbe
+}
+
+// validateCheckMode 校验 check_mode 与 provider 的组合矩阵：
+//
+//	provider                | probe | quota | quota_probe
+//	------------------------+-------+-------+------------
+//	openai/anthropic/...    |  Y    |  Y    |  Y
+//	antigravity（无 adapter）|  N    |  Y    |  N
+func validateCheckMode(provider, checkMode string) error {
+	checkMode = defaultCheckMode(checkMode)
+	switch checkMode {
+	case MonitorCheckModeProbe, MonitorCheckModeQuota, MonitorCheckModeQuotaProbe:
+	default:
+		return ErrChannelMonitorInvalidCheckMode
+	}
+	if checkMode != MonitorCheckModeQuota && !providerSupportsProbe(provider) {
+		return ErrChannelMonitorInvalidCheckMode
+	}
+	return nil
+}
+
 // validateAPIMode 校验 provider 与 api_mode 的组合。
 // responses 只对 OpenAI 有意义；其它 provider 使用 chat_completions 作为默认占位。
 func validateAPIMode(provider, apiMode string) error {
@@ -124,13 +190,18 @@ func normalizeModels(in []string) []string {
 	return out
 }
 
-// normalizeMonitorPrimaryModel applies the Grok health-check default while
-// preserving the existing required-model behavior for every other provider.
-func normalizeMonitorPrimaryModel(provider, model string) string {
+// normalizeMonitorPrimaryModel applies provider/check_mode defaults while
+// preserving the existing required-model behavior:
+//   - Grok 探活默认轻量测活模型
+//   - quota 模式占位 "quota"（primary_model 列 NotEmpty；历史行/时间线机制无需特判）
+func normalizeMonitorPrimaryModel(provider, checkMode, model string) string {
 	model = strings.TrimSpace(model)
 	if model == "" && provider == MonitorProviderGrok {
 		return MonitorDefaultGrokModel
 	}
+	if model == "" && monitorCheckModeUsesQuota(defaultCheckMode(checkMode)) {
+		return MonitorDefaultQuotaModel
+	}
 	return model
 }
 

From 6a6fd304f68e2a8f5dbe6a96e424c4baffc203c6 Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 04:06:18 +0000
Subject: [PATCH 47/68] feat(settings): channel_monitor_show_quota public
 setting (default off)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 新增公开设置 key（迁移 226 已插入默认 false），控制用户端监控页
  是否展示配额/余额；管理端不受影响
- 解析 fail-closed：仅字面 "true" 视为开启（对齐 available_channels_enabled
  语义，而非 hide_throughput 的 fail-open）
- 全链路贯通：domain key 常量 → setting_public（公开读取 + ChannelMonitorRuntime
  .ShowQuota + 注入 payload）→ settings_view/parse/update → handler DTO
  （admin/user 响应 + admin 更新请求）→ api_contract_test 两个 wantJSON 块
---
 backend/internal/handler/admin/setting_handler.go    |  1 +
 .../internal/handler/admin/setting_handler_update.go |  8 ++++++++
 backend/internal/handler/dto/settings.go             |  2 ++
 backend/internal/handler/setting_handler.go          |  1 +
 backend/internal/server/api_contract_test.go         |  2 ++
 backend/internal/service/domain_constants.go         |  7 +++++++
 backend/internal/service/setting_parse.go            |  4 ++++
 backend/internal/service/setting_public.go           | 12 ++++++++++++
 backend/internal/service/setting_update.go           |  1 +
 backend/internal/service/settings_view.go            |  2 ++
 10 files changed, 40 insertions(+)

diff --git a/backend/internal/handler/admin/setting_handler.go b/backend/internal/handler/admin/setting_handler.go
index a5f3d41d9065..01a066165ba4 100644
--- a/backend/internal/handler/admin/setting_handler.go
+++ b/backend/internal/handler/admin/setting_handler.go
@@ -374,6 +374,7 @@ func (h *SettingHandler) GetSettings(c *gin.Context) {
 		ChannelMonitorMode:                   settings.ChannelMonitorMode,
 		ChannelMonitorDefaultIntervalSeconds: settings.ChannelMonitorDefaultIntervalSeconds,
 		ChannelMonitorHideThroughput:         settings.ChannelMonitorHideThroughput,
+		ChannelMonitorShowQuota:              settings.ChannelMonitorShowQuota,
 
 		GrokDefaultTextModel:           settings.GrokDefaultTextModel,
 		GrokCrossClientModelMapEnabled: settings.GrokCrossClientModelMapEnabled,
diff --git a/backend/internal/handler/admin/setting_handler_update.go b/backend/internal/handler/admin/setting_handler_update.go
index 5884dc392e17..e2399d0891ee 100644
--- a/backend/internal/handler/admin/setting_handler_update.go
+++ b/backend/internal/handler/admin/setting_handler_update.go
@@ -332,6 +332,7 @@ type UpdateSettingsRequest struct {
 	ChannelMonitorMode                   *string `json:"channel_monitor_mode"`
 	ChannelMonitorDefaultIntervalSeconds *int    `json:"channel_monitor_default_interval_seconds"`
 	ChannelMonitorHideThroughput         *bool   `json:"channel_monitor_hide_throughput"`
+	ChannelMonitorShowQuota              *bool   `json:"channel_monitor_show_quota"`
 
 	// Grok model mapping policy
 	GrokDefaultTextModel           *string `json:"grok_default_text_model"`
@@ -1889,6 +1890,12 @@ func (h *SettingHandler) UpdateSettings(c *gin.Context) {
 			}
 			return previousSettings.ChannelMonitorHideThroughput
 		}(),
+		ChannelMonitorShowQuota: func() bool {
+			if req.ChannelMonitorShowQuota != nil {
+				return *req.ChannelMonitorShowQuota
+			}
+			return previousSettings.ChannelMonitorShowQuota
+		}(),
 		GrokDefaultTextModel: func() string {
 			if req.GrokDefaultTextModel != nil {
 				return *req.GrokDefaultTextModel
@@ -2342,6 +2349,7 @@ func (h *SettingHandler) UpdateSettings(c *gin.Context) {
 		ChannelMonitorMode:                   updatedSettings.ChannelMonitorMode,
 		ChannelMonitorDefaultIntervalSeconds: updatedSettings.ChannelMonitorDefaultIntervalSeconds,
 		ChannelMonitorHideThroughput:         updatedSettings.ChannelMonitorHideThroughput,
+		ChannelMonitorShowQuota:              updatedSettings.ChannelMonitorShowQuota,
 
 		GrokDefaultTextModel:           updatedSettings.GrokDefaultTextModel,
 		GrokCrossClientModelMapEnabled: updatedSettings.GrokCrossClientModelMapEnabled,
diff --git a/backend/internal/handler/dto/settings.go b/backend/internal/handler/dto/settings.go
index 5ab5eaa68b6b..f0284eb55ec7 100644
--- a/backend/internal/handler/dto/settings.go
+++ b/backend/internal/handler/dto/settings.go
@@ -305,6 +305,7 @@ type SystemSettings struct {
 	ChannelMonitorMode                   string `json:"channel_monitor_mode"`
 	ChannelMonitorDefaultIntervalSeconds int    `json:"channel_monitor_default_interval_seconds"`
 	ChannelMonitorHideThroughput         bool   `json:"channel_monitor_hide_throughput"`
+	ChannelMonitorShowQuota              bool   `json:"channel_monitor_show_quota"`
 
 	// Grok model mapping policy (admin settings; empty account mapping falls back to these).
 	GrokDefaultTextModel           string `json:"grok_default_text_model"`
@@ -414,6 +415,7 @@ type PublicSettings struct {
 	ChannelMonitorMode                   string `json:"channel_monitor_mode"`
 	ChannelMonitorDefaultIntervalSeconds int    `json:"channel_monitor_default_interval_seconds"`
 	ChannelMonitorHideThroughput         bool   `json:"channel_monitor_hide_throughput"`
+	ChannelMonitorShowQuota              bool   `json:"channel_monitor_show_quota"`
 
 	AvailableChannelsEnabled bool `json:"available_channels_enabled"`
 
diff --git a/backend/internal/handler/setting_handler.go b/backend/internal/handler/setting_handler.go
index a415d67e3484..f0785de9961a 100644
--- a/backend/internal/handler/setting_handler.go
+++ b/backend/internal/handler/setting_handler.go
@@ -107,6 +107,7 @@ func (h *SettingHandler) GetPublicSettings(c *gin.Context) {
 		ChannelMonitorMode:                   settings.ChannelMonitorMode,
 		ChannelMonitorDefaultIntervalSeconds: settings.ChannelMonitorDefaultIntervalSeconds,
 		ChannelMonitorHideThroughput:         settings.ChannelMonitorHideThroughput,
+		ChannelMonitorShowQuota:              settings.ChannelMonitorShowQuota,
 
 		AvailableChannelsEnabled: settings.AvailableChannelsEnabled,
 
diff --git a/backend/internal/server/api_contract_test.go b/backend/internal/server/api_contract_test.go
index b0b1df7bc0f4..7f3b25f76378 100644
--- a/backend/internal/server/api_contract_test.go
+++ b/backend/internal/server/api_contract_test.go
@@ -988,6 +988,7 @@ func TestAPIContracts(t *testing.T) {
 					"channel_monitor_enabled": true,
 					"channel_monitor_mode": "v1",
 					"channel_monitor_hide_throughput": true,
+					"channel_monitor_show_quota": false,
 					"channel_monitor_default_interval_seconds": 60,
 					"available_channels_enabled": false,
 					"model_plaza_enabled": false,
@@ -1298,6 +1299,7 @@ func TestAPIContracts(t *testing.T) {
 					"channel_monitor_enabled": true,
 					"channel_monitor_mode": "v1",
 					"channel_monitor_hide_throughput": true,
+					"channel_monitor_show_quota": false,
 					"channel_monitor_default_interval_seconds": 60,
 					"available_channels_enabled": false,
 					"model_plaza_enabled": false,
diff --git a/backend/internal/service/domain_constants.go b/backend/internal/service/domain_constants.go
index c0efa2b527b9..ba35301be08d 100644
--- a/backend/internal/service/domain_constants.go
+++ b/backend/internal/service/domain_constants.go
@@ -477,6 +477,13 @@ const (
 	// Default false (show rates). Admin endpoints always keep full metrics.
 	SettingKeyChannelMonitorHideThroughput = "channel_monitor_hide_throughput"
 
+	// SettingKeyChannelMonitorShowQuota controls whether quota/balance snapshots
+	// attached to channel monitors (check_mode=quota/quota_probe) are exposed on
+	// the user-facing monitor APIs and UI. Default false (hidden); parsed
+	// fail-closed (only the literal "true" enables it). Admin endpoints always
+	// keep the full snapshots regardless of this flag.
+	SettingKeyChannelMonitorShowQuota = "channel_monitor_show_quota"
+
 	// SettingKeyGrokDefaultTextModel is the fallback Grok text model for empty
 	// request models and built-in Grok aliases (e.g. "grok" → this id). Default grok-4.5.
 	SettingKeyGrokDefaultTextModel = "grok_default_text_model"
diff --git a/backend/internal/service/setting_parse.go b/backend/internal/service/setting_parse.go
index 283c76f8f4c2..9f3dfbddf243 100644
--- a/backend/internal/service/setting_parse.go
+++ b/backend/internal/service/setting_parse.go
@@ -190,6 +190,7 @@ func (s *SettingService) InitializeDefaultSettings(ctx context.Context) error {
 		SettingKeyChannelMonitorMode:                   ChannelMonitorModeV1,
 		SettingKeyChannelMonitorDefaultIntervalSeconds: "60",
 		SettingKeyChannelMonitorHideThroughput:         "true",
+		SettingKeyChannelMonitorShowQuota:              "false",
 
 		// Grok: safe defaults — no cross-vendor model rewrite unless operators enable it.
 		SettingKeyGrokDefaultTextModel:           "grok-4.5",
@@ -798,6 +799,9 @@ func (s *SettingService) parseSettings(settings map[string]string) *SystemSettin
 	// 默认隐藏吞吐（迁移 206 的隐私默认）：未配置时必须与 setting_public.go 的
 	// 公开读取路径给出同一个值，否则管理端看到“未隐藏”而用户端实际已隐藏。
 	result.ChannelMonitorHideThroughput = !isFalseSettingValue(settings[SettingKeyChannelMonitorHideThroughput])
+	// 配额展示默认关闭且 fail-closed：仅字面 "true" 视为开启
+	// （与 setting_public.go 公开读取路径保持一致）。
+	result.ChannelMonitorShowQuota = settings[SettingKeyChannelMonitorShowQuota] == "true"
 
 	// Grok default mapping policy
 	result.GrokDefaultTextModel = strings.TrimSpace(settings[SettingKeyGrokDefaultTextModel])
diff --git a/backend/internal/service/setting_public.go b/backend/internal/service/setting_public.go
index f292e8ac5782..4679d0048467 100644
--- a/backend/internal/service/setting_public.go
+++ b/backend/internal/service/setting_public.go
@@ -231,6 +231,7 @@ func (s *SettingService) GetPublicSettings(ctx context.Context) (*PublicSettings
 		SettingKeyChannelMonitorMode,
 		SettingKeyChannelMonitorDefaultIntervalSeconds,
 		SettingKeyChannelMonitorHideThroughput,
+		SettingKeyChannelMonitorShowQuota,
 		SettingKeyAvailableChannelsEnabled,
 		SettingKeyModelPlazaEnabled,
 		SettingKeyModelPlazaRequireAuth,
@@ -355,6 +356,7 @@ func (s *SettingService) GetPublicSettings(ctx context.Context) (*PublicSettings
 		ChannelMonitorMode:                   normalizeChannelMonitorMode(settings[SettingKeyChannelMonitorMode]),
 		ChannelMonitorDefaultIntervalSeconds: parseChannelMonitorInterval(settings[SettingKeyChannelMonitorDefaultIntervalSeconds]),
 		ChannelMonitorHideThroughput:         !isFalseSettingValue(settings[SettingKeyChannelMonitorHideThroughput]),
+		ChannelMonitorShowQuota:              settings[SettingKeyChannelMonitorShowQuota] == "true",
 
 		AvailableChannelsEnabled: settings[SettingKeyAvailableChannelsEnabled] == "true",
 
@@ -422,6 +424,10 @@ type ChannelMonitorRuntime struct {
 	DefaultIntervalSeconds int
 	// HideThroughput: when true, user-facing V2 APIs omit RPM/TPM scale signals.
 	HideThroughput bool
+	// ShowQuota: when true, user-facing monitor views keep the quota/balance
+	// snapshots; otherwise the user handler strips them server-side.
+	// Parsed fail-closed (only literal "true" enables). Admin always sees them.
+	ShowQuota bool
 }
 
 // ActiveProbesAllowed reports whether V1 active provider probes may run.
@@ -450,6 +456,7 @@ func (s *SettingService) GetChannelMonitorRuntime(ctx context.Context) ChannelMo
 		SettingKeyChannelMonitorMode,
 		SettingKeyChannelMonitorDefaultIntervalSeconds,
 		SettingKeyChannelMonitorHideThroughput,
+		SettingKeyChannelMonitorShowQuota,
 	})
 	if err != nil {
 		return ChannelMonitorRuntime{
@@ -464,6 +471,7 @@ func (s *SettingService) GetChannelMonitorRuntime(ctx context.Context) ChannelMo
 		Mode:                   normalizeChannelMonitorMode(vals[SettingKeyChannelMonitorMode]),
 		DefaultIntervalSeconds: parseChannelMonitorInterval(vals[SettingKeyChannelMonitorDefaultIntervalSeconds]),
 		HideThroughput:         !isFalseSettingValue(vals[SettingKeyChannelMonitorHideThroughput]),
+		ShowQuota:              vals[SettingKeyChannelMonitorShowQuota] == "true",
 	}
 }
 
@@ -606,6 +614,9 @@ type PublicSettingsInjectionPayload struct {
 	// ChannelMonitorHideThroughput is public so the user UI can hide RPM/TPM
 	// without waiting for API redaction alone (defense in depth).
 	ChannelMonitorHideThroughput bool `json:"channel_monitor_hide_throughput"`
+	// ChannelMonitorShowQuota gates the user-facing quota/balance display on
+	// monitors; fail-closed (absent/false = hidden). Admin UI always shows it.
+	ChannelMonitorShowQuota bool `json:"channel_monitor_show_quota"`
 	AvailableChannelsEnabled     bool `json:"available_channels_enabled"`
 	ModelPlazaEnabled            bool `json:"model_plaza_enabled"`
 	ModelPlazaRequireAuth        bool `json:"model_plaza_require_auth"`
@@ -685,6 +696,7 @@ func (s *SettingService) GetPublicSettingsForInjection(ctx context.Context) (any
 		ChannelMonitorMode:                   settings.ChannelMonitorMode,
 		ChannelMonitorDefaultIntervalSeconds: settings.ChannelMonitorDefaultIntervalSeconds,
 		ChannelMonitorHideThroughput:         settings.ChannelMonitorHideThroughput,
+		ChannelMonitorShowQuota:              settings.ChannelMonitorShowQuota,
 		AvailableChannelsEnabled:             settings.AvailableChannelsEnabled,
 		ModelPlazaEnabled:                    settings.ModelPlazaEnabled,
 		ModelPlazaRequireAuth:                settings.ModelPlazaRequireAuth,
diff --git a/backend/internal/service/setting_update.go b/backend/internal/service/setting_update.go
index f77a914fd73a..a4fadded0c06 100644
--- a/backend/internal/service/setting_update.go
+++ b/backend/internal/service/setting_update.go
@@ -417,6 +417,7 @@ func (s *SettingService) buildSystemSettingsUpdates(ctx context.Context, setting
 		updates[SettingKeyChannelMonitorDefaultIntervalSeconds] = strconv.Itoa(v)
 	}
 	updates[SettingKeyChannelMonitorHideThroughput] = strconv.FormatBool(settings.ChannelMonitorHideThroughput)
+	updates[SettingKeyChannelMonitorShowQuota] = strconv.FormatBool(settings.ChannelMonitorShowQuota)
 
 	// Grok model mapping policy
 	if v := strings.TrimSpace(settings.GrokDefaultTextModel); v != "" {
diff --git a/backend/internal/service/settings_view.go b/backend/internal/service/settings_view.go
index 8580f0c8fb45..a939000947bb 100644
--- a/backend/internal/service/settings_view.go
+++ b/backend/internal/service/settings_view.go
@@ -201,6 +201,7 @@ type SystemSettings struct {
 	ChannelMonitorMode                   string `json:"channel_monitor_mode"`
 	ChannelMonitorDefaultIntervalSeconds int    `json:"channel_monitor_default_interval_seconds"`
 	ChannelMonitorHideThroughput         bool   `json:"channel_monitor_hide_throughput"`
+	ChannelMonitorShowQuota              bool   `json:"channel_monitor_show_quota"`
 
 	// Grok model mapping policy (admin settings; empty mapping falls back to these).
 	GrokDefaultTextModel           string `json:"grok_default_text_model"`
@@ -378,6 +379,7 @@ type PublicSettings struct {
 	ChannelMonitorMode                   string `json:"channel_monitor_mode"`
 	ChannelMonitorDefaultIntervalSeconds int    `json:"channel_monitor_default_interval_seconds"`
 	ChannelMonitorHideThroughput         bool   `json:"channel_monitor_hide_throughput"`
+	ChannelMonitorShowQuota              bool   `json:"channel_monitor_show_quota"`
 
 	// Grok model mapping policy (admin settings).
 	GrokDefaultTextModel           string `json:"grok_default_text_model"`

From 41344c20ff54e4a828a09fa0ca77d0167a66b7f4 Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 04:14:15 +0000
Subject: [PATCH 48/68] feat(monitor): wire quota fetcher & expose check_mode
 in handlers
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- handler DTO: create/update 接收 check_mode/account_id，provider oneof 扩至 8 家，
  endpoint/api_key 改为 omitempty（条件必填下沉 service 校验）；
  monitor/checkResult/historyItem 响应透传 check_mode/account_id/quota
- 用户端 latest_quota 由 channel_monitor_show_quota 控制，关闭时服务端剥离
- wire: NewChannelMonitorQuotaFetcher 以具体服务类型收参（窄接口包内保留供
  stub），ProvideChannelMonitorRunner 注入后 SetQuotaFetcher
---
 backend/cmd/server/wire_gen.go                |  3 +-
 .../handler/admin/channel_monitor_handler.go  | 65 ++++++++++++++-----
 .../handler/channel_monitor_user_handler.go   | 24 ++++++-
 .../service/channel_monitor_quota_fetcher.go  | 28 +++++---
 backend/internal/service/wire.go              | 10 ++-
 5 files changed, 97 insertions(+), 33 deletions(-)

diff --git a/backend/cmd/server/wire_gen.go b/backend/cmd/server/wire_gen.go
index 2b0caf982c4a..e68a6bd7d545 100644
--- a/backend/cmd/server/wire_gen.go
+++ b/backend/cmd/server/wire_gen.go
@@ -337,7 +337,8 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	batchImageWorkerRuntime := service.ProvideBatchImageWorkerRuntime(batchImageRepository, accountRepository, batchImageQueue, usageBillingRepository, usageLogRepository, batchImageModelPricingResolver, apiKeyAuthCacheInvalidator, configConfig)
 	scheduledTestRunnerService := service.ProvideScheduledTestRunnerService(scheduledTestPlanRepository, scheduledTestService, accountTestService, rateLimitService, configConfig)
 	paymentOrderExpiryService := service.ProvidePaymentOrderExpiryService(paymentService, leaderLockCache, db)
-	channelMonitorRunner := service.ProvideChannelMonitorRunner(channelMonitorService, settingService)
+	channelMonitorQuotaFetcher := service.NewChannelMonitorQuotaFetcher(accountUsageService, cnProviderQuotaService, cnProviderBalanceService, accountRepository)
+	channelMonitorRunner := service.ProvideChannelMonitorRunner(channelMonitorService, settingService, channelMonitorQuotaFetcher)
 	channelMonitorV2Aggregator := service.ProvideChannelMonitorV2Aggregator(channelMonitorV2Repository, db, settingService)
 	userPlatformQuotaUsageFlusher := service.ProvideUserPlatformQuotaUsageFlusher(configConfig, billingCache, serviceUserPlatformQuotaRepository, timingWheelService)
 	v := provideCleanup(client, redisClient, opsMetricsCollector, opsAggregationService, opsAlertEvaluatorService, opsCleanupService, opsScheduledReportService, opsSystemLogSink, opsService, opsIngressRejectAggregator, apiKeyService, authCacheInvalidationWorker, schedulerSnapshotService, tokenRefreshService, accountExpiryService, cnProviderBalanceCheckService, openAICodexVersionSyncService, proxyExpiryService, subscriptionExpiryService, usageCleanupService, idempotencyCleanupService, batchImageCleanupService, batchImageWorkerRuntime, pricingService, emailQueueService, billingCacheService, usageRecordWorkerPool, subscriptionService, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, grokOAuthService, openAIGatewayService, scheduledTestRunnerService, backupService, paymentOrderExpiryService, channelMonitorRunner, channelMonitorV2Aggregator, userPlatformQuotaUsageFlusher, upstreamBillingProbeService, ollamaCloudUsageService, auditLogService, promptService)
diff --git a/backend/internal/handler/admin/channel_monitor_handler.go b/backend/internal/handler/admin/channel_monitor_handler.go
index ec5735f0ca35..4832be2decd7 100644
--- a/backend/internal/handler/admin/channel_monitor_handler.go
+++ b/backend/internal/handler/admin/channel_monitor_handler.go
@@ -7,6 +7,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/domain"
 	"github.com/Wei-Shaw/sub2api/internal/handler/dto"
 	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/response"
@@ -39,10 +40,10 @@ func NewChannelMonitorHandler(monitorService *service.ChannelMonitorService) *Ch
 
 type channelMonitorCreateRequest struct {
 	Name             string            `json:"name" binding:"required,max=100"`
-	Provider         string            `json:"provider" binding:"required,oneof=openai anthropic gemini grok"`
+	Provider         string            `json:"provider" binding:"required,oneof=openai anthropic gemini grok antigravity kimi zhipu deepseek"`
 	APIMode          string            `json:"api_mode" binding:"omitempty,oneof=chat_completions responses"`
-	Endpoint         string            `json:"endpoint" binding:"required,max=500"`
-	APIKey           string            `json:"api_key" binding:"required,max=2000"`
+	Endpoint         string            `json:"endpoint" binding:"omitempty,max=500"`
+	APIKey           string            `json:"api_key" binding:"omitempty,max=2000"`
 	PrimaryModel     string            `json:"primary_model" binding:"max=200"`
 	ExtraModels      []string          `json:"extra_models"`
 	GroupName        string            `json:"group_name" binding:"max=100"`
@@ -53,11 +54,17 @@ type channelMonitorCreateRequest struct {
 	ExtraHeaders     map[string]string `json:"extra_headers"`
 	BodyOverrideMode string            `json:"body_override_mode" binding:"omitempty,oneof=off merge replace"`
 	BodyOverride     map[string]any    `json:"body_override"`
+
+	// CheckMode: probe（默认）/ quota / quota_probe。quota 模式 endpoint/api_key
+	// 可空（条件必填校验在 service 层按模式分支）。
+	CheckMode string `json:"check_mode" binding:"omitempty,oneof=probe quota quota_probe"`
+	// AccountID: 配额模式关联的账号 ID。
+	AccountID *int64 `json:"account_id"`
 }
 
 type channelMonitorUpdateRequest struct {
 	Name             *string            `json:"name" binding:"omitempty,max=100"`
-	Provider         *string            `json:"provider" binding:"omitempty,oneof=openai anthropic gemini grok"`
+	Provider         *string            `json:"provider" binding:"omitempty,oneof=openai anthropic gemini grok antigravity kimi zhipu deepseek"`
 	APIMode          *string            `json:"api_mode" binding:"omitempty,oneof=chat_completions responses"`
 	Endpoint         *string            `json:"endpoint" binding:"omitempty,max=500"`
 	APIKey           *string            `json:"api_key" binding:"omitempty,max=2000"`
@@ -72,6 +79,10 @@ type channelMonitorUpdateRequest struct {
 	ExtraHeaders     *map[string]string `json:"extra_headers"`
 	BodyOverrideMode *string            `json:"body_override_mode" binding:"omitempty,oneof=off merge replace"`
 	BodyOverride     *map[string]any    `json:"body_override"`
+
+	// CheckMode/AccountID：nil = 不更新；AccountID 指向 0 = 清空关联。
+	CheckMode *string `json:"check_mode" binding:"omitempty,oneof=probe quota quota_probe"`
+	AccountID *int64  `json:"account_id"`
 }
 
 type channelMonitorResponse struct {
@@ -101,25 +112,33 @@ type channelMonitorResponse struct {
 	ExtraHeaders     map[string]string `json:"extra_headers"`
 	BodyOverrideMode string            `json:"body_override_mode"`
 	BodyOverride     map[string]any    `json:"body_override"`
+
+	// 配额模式：check_mode + 关联账号 + 主模型最近配额快照
+	// （LatestQuota 由 List handler 批量聚合后填充；管理端不受 channel_monitor_show_quota 影响）。
+	CheckMode   string                       `json:"check_mode"`
+	AccountID   *int64                       `json:"account_id"`
+	LatestQuota *domain.MonitorQuotaSnapshot `json:"latest_quota,omitempty"`
 }
 
 type channelMonitorCheckResultResponse struct {
-	Model         string `json:"model"`
-	Status        string `json:"status"`
-	LatencyMs     *int   `json:"latency_ms"`
-	PingLatencyMs *int   `json:"ping_latency_ms"`
-	Message       string `json:"message"`
-	CheckedAt     string `json:"checked_at"`
+	Model         string                       `json:"model"`
+	Status        string                       `json:"status"`
+	LatencyMs     *int                         `json:"latency_ms"`
+	PingLatencyMs *int                         `json:"ping_latency_ms"`
+	Message       string                       `json:"message"`
+	CheckedAt     string                       `json:"checked_at"`
+	Quota         *domain.MonitorQuotaSnapshot `json:"quota,omitempty"`
 }
 
 type channelMonitorHistoryItemResponse struct {
-	ID            int64  `json:"id"`
-	Model         string `json:"model"`
-	Status        string `json:"status"`
-	LatencyMs     *int   `json:"latency_ms"`
-	PingLatencyMs *int   `json:"ping_latency_ms"`
-	Message       string `json:"message"`
-	CheckedAt     string `json:"checked_at"`
+	ID            int64                        `json:"id"`
+	Model         string                       `json:"model"`
+	Status        string                       `json:"status"`
+	LatencyMs     *int                         `json:"latency_ms"`
+	PingLatencyMs *int                         `json:"ping_latency_ms"`
+	Message       string                       `json:"message"`
+	CheckedAt     string                       `json:"checked_at"`
+	Quota         *domain.MonitorQuotaSnapshot `json:"quota,omitempty"`
 }
 
 // maskAPIKey 对 API Key 明文做脱敏：前 4 字符 + "***"，长度 ≤ 4 时只显示 "***"。
@@ -163,7 +182,10 @@ func channelMonitorToResponse(m *service.ChannelMonitor) *channelMonitorResponse
 		ExtraHeaders:        headers,
 		BodyOverrideMode:    m.BodyOverrideMode,
 		BodyOverride:        m.BodyOverride,
-		// PrimaryStatus / PrimaryLatencyMs / Availability7d 由 List handler 在批量聚合后填充。
+		CheckMode:           m.CheckMode,
+		AccountID:           m.AccountID,
+		// PrimaryStatus / PrimaryLatencyMs / Availability7d / LatestQuota
+		// 由 List handler 在批量聚合后填充。
 	}
 	if m.LastCheckedAt != nil {
 		s := m.LastCheckedAt.UTC().Format(time.RFC3339)
@@ -180,6 +202,7 @@ func checkResultToResponse(r *service.CheckResult) channelMonitorCheckResultResp
 		PingLatencyMs: r.PingLatencyMs,
 		Message:       r.Message,
 		CheckedAt:     r.CheckedAt.UTC().Format(time.RFC3339),
+		Quota:         r.Quota,
 	}
 }
 
@@ -192,6 +215,7 @@ func historyEntryToResponse(e *service.ChannelMonitorHistoryEntry) channelMonito
 		PingLatencyMs: e.PingLatencyMs,
 		Message:       e.Message,
 		CheckedAt:     e.CheckedAt.UTC().Format(time.RFC3339),
+		Quota:         e.Quota,
 	}
 }
 
@@ -270,6 +294,7 @@ func buildListItemResponse(m *service.ChannelMonitor, summary service.MonitorSta
 	resp.PrimaryStatus = summary.PrimaryStatus
 	resp.PrimaryLatencyMs = summary.PrimaryLatencyMs
 	resp.Availability7d = summary.Availability7d
+	resp.LatestQuota = summary.LatestQuota
 	resp.ExtraModelsStatus = make([]dto.ChannelMonitorExtraModelStatus, 0, len(summary.ExtraModels))
 	for _, e := range summary.ExtraModels {
 		resp.ExtraModelsStatus = append(resp.ExtraModelsStatus, dto.ChannelMonitorExtraModelStatus{
@@ -327,6 +352,8 @@ func (h *ChannelMonitorHandler) Create(c *gin.Context) {
 		ExtraHeaders:     req.ExtraHeaders,
 		BodyOverrideMode: req.BodyOverrideMode,
 		BodyOverride:     req.BodyOverride,
+		CheckMode:        req.CheckMode,
+		AccountID:        req.AccountID,
 	})
 	if err != nil {
 		response.ErrorFrom(c, err)
@@ -421,6 +448,8 @@ func (h *ChannelMonitorHandler) Update(c *gin.Context) {
 		ExtraHeaders:     req.ExtraHeaders,
 		BodyOverrideMode: req.BodyOverrideMode,
 		BodyOverride:     req.BodyOverride,
+		CheckMode:        req.CheckMode,
+		AccountID:        req.AccountID,
 	})
 	if err != nil {
 		response.ErrorFrom(c, err)
diff --git a/backend/internal/handler/channel_monitor_user_handler.go b/backend/internal/handler/channel_monitor_user_handler.go
index 7e42c5a74d27..173ca56dfe01 100644
--- a/backend/internal/handler/channel_monitor_user_handler.go
+++ b/backend/internal/handler/channel_monitor_user_handler.go
@@ -3,6 +3,7 @@ package handler
 import (
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/domain"
 	"github.com/Wei-Shaw/sub2api/internal/handler/admin"
 	"github.com/Wei-Shaw/sub2api/internal/handler/dto"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/response"
@@ -39,6 +40,15 @@ func (h *ChannelMonitorUserHandler) featureEnabled(c *gin.Context) bool {
 	return runtime.Enabled && runtime.Mode == service.ChannelMonitorModeV1
 }
 
+// quotaVisible 返回用户端是否展示配额/余额快照（channel_monitor_show_quota，
+// fail-closed：未配置/非 "true" 一律视为关闭）。settingService 为 nil 时 fail-closed。
+func (h *ChannelMonitorUserHandler) quotaVisible(c *gin.Context) bool {
+	if h.settingService == nil {
+		return false
+	}
+	return h.settingService.GetChannelMonitorRuntime(c.Request.Context()).ShowQuota
+}
+
 // --- Response ---
 
 type channelMonitorUserListItem struct {
@@ -53,6 +63,9 @@ type channelMonitorUserListItem struct {
 	Availability7d       float64                              `json:"availability_7d"`
 	ExtraModels          []dto.ChannelMonitorExtraModelStatus `json:"extra_models"`
 	Timeline             []channelMonitorUserTimelinePoint    `json:"timeline"`
+	// LatestQuota 主模型最近配额快照；channel_monitor_show_quota=false 时
+	// 由 userMonitorViewToItem 的调用方传入 false 剥离（服务端脱敏，非仅前端隐藏）。
+	LatestQuota *domain.MonitorQuotaSnapshot `json:"latest_quota,omitempty"`
 }
 
 // channelMonitorUserTimelinePoint 主模型最近一次检测的 timeline 点。
@@ -82,7 +95,7 @@ type channelMonitorUserModelStat struct {
 	AvgLatency7dMs  *int    `json:"avg_latency_7d_ms"`
 }
 
-func userMonitorViewToItem(v *service.UserMonitorView) channelMonitorUserListItem {
+func userMonitorViewToItem(v *service.UserMonitorView, includeQuota bool) channelMonitorUserListItem {
 	extras := make([]dto.ChannelMonitorExtraModelStatus, 0, len(v.ExtraModels))
 	for _, e := range v.ExtraModels {
 		extras = append(extras, dto.ChannelMonitorExtraModelStatus{
@@ -100,7 +113,7 @@ func userMonitorViewToItem(v *service.UserMonitorView) channelMonitorUserListIte
 			CheckedAt:     p.CheckedAt.UTC().Format(time.RFC3339),
 		})
 	}
-	return channelMonitorUserListItem{
+	item := channelMonitorUserListItem{
 		ID:                   v.ID,
 		Name:                 v.Name,
 		Provider:             v.Provider,
@@ -113,6 +126,10 @@ func userMonitorViewToItem(v *service.UserMonitorView) channelMonitorUserListIte
 		ExtraModels:          extras,
 		Timeline:             timeline,
 	}
+	if includeQuota {
+		item.LatestQuota = v.LatestQuota
+	}
+	return item
 }
 
 func userMonitorDetailToResponse(d *service.UserMonitorDetail) *channelMonitorUserDetailResponse {
@@ -150,9 +167,10 @@ func (h *ChannelMonitorUserHandler) List(c *gin.Context) {
 		response.ErrorFrom(c, err)
 		return
 	}
+	includeQuota := h.quotaVisible(c)
 	items := make([]channelMonitorUserListItem, 0, len(views))
 	for _, v := range views {
-		items = append(items, userMonitorViewToItem(v))
+		items = append(items, userMonitorViewToItem(v, includeQuota))
 	}
 	response.Success(c, gin.H{"items": items})
 }
diff --git a/backend/internal/service/channel_monitor_quota_fetcher.go b/backend/internal/service/channel_monitor_quota_fetcher.go
index 83d8e0f45527..d208dadc1b3b 100644
--- a/backend/internal/service/channel_monitor_quota_fetcher.go
+++ b/backend/internal/service/channel_monitor_quota_fetcher.go
@@ -65,19 +65,27 @@ type monitorQuotaCacheEntry struct {
 }
 
 // NewChannelMonitorQuotaFetcher 构造配额抓取器。
+// 参数取具体服务类型以便 wire 直连；单元测试在同包内用 struct 字面量注入 stub。
 func NewChannelMonitorQuotaFetcher(
-	usage monitorUsageSource,
-	cnQuota monitorCNQuotaSource,
-	cnBalance monitorCNBalanceSource,
-	accounts monitorAccountSource,
+	usage *AccountUsageService,
+	cnQuota *CNProviderQuotaService,
+	cnBalance *CNProviderBalanceService,
+	accounts AccountRepository,
 ) *ChannelMonitorQuotaFetcher {
-	return &ChannelMonitorQuotaFetcher{
-		usage:     usage,
-		cnQuota:   cnQuota,
-		cnBalance: cnBalance,
-		accounts:  accounts,
-		cache:     make(map[int64]monitorQuotaCacheEntry),
+	f := &ChannelMonitorQuotaFetcher{cache: make(map[int64]monitorQuotaCacheEntry)}
+	if usage != nil {
+		f.usage = usage
 	}
+	if cnQuota != nil {
+		f.cnQuota = cnQuota
+	}
+	if cnBalance != nil {
+		f.cnBalance = cnBalance
+	}
+	if accounts != nil {
+		f.accounts = accounts
+	}
+	return f
 }
 
 // LoadAccount 加载账号（不走缓存）。供 Create/Update 时校验
diff --git a/backend/internal/service/wire.go b/backend/internal/service/wire.go
index d6bb56fac423..88c3a1f18523 100644
--- a/backend/internal/service/wire.go
+++ b/backend/internal/service/wire.go
@@ -907,6 +907,7 @@ var ProviderSet = wire.NewSet(
 	ProvideBalanceNotifyService,
 	ProvideChannelMonitorService,
 	ProvideChannelMonitorRunner,
+	NewChannelMonitorQuotaFetcher,
 	ProvideChannelMonitorV2Service,
 	ProvideChannelMonitorV2Aggregator,
 	NewChannelMonitorRequestTemplateService,
@@ -965,13 +966,20 @@ func ProvideChannelMonitorService(
 // 通过 SetScheduler 注入回 service 后再 Start，确保启动时加载所有 enabled monitor，
 // 后续 CRUD 也能即时同步任务表。Runner.Stop 由 cleanup function 调用。
 // settingService 用于 runner 每次 fire 读取功能开关。
-func ProvideChannelMonitorRunner(svc *ChannelMonitorService, settingService *SettingService) *ChannelMonitorRunner {
+// quotaFetcher（账号侧用量聚合）也在此注入：accountUsage/CN 服务在 wire 图中
+// 晚于 channelMonitorService 构造，走 setter 注入避免调整既有构造顺序。
+func ProvideChannelMonitorRunner(
+	svc *ChannelMonitorService,
+	settingService *SettingService,
+	quotaFetcher *ChannelMonitorQuotaFetcher,
+) *ChannelMonitorRunner {
 	r := NewChannelMonitorRunner(svc, settingService)
 	if svc != nil {
 		// Ensure runtime reader is set even if ProvideChannelMonitorService
 		// was constructed without settings (tests / alternate providers).
 		svc.SetRuntimeReader(settingService)
 		svc.SetScheduler(r)
+		svc.SetQuotaFetcher(quotaFetcher)
 	}
 	r.Start()
 	return r

From 7ab6d3db66c4862013a6ee998c014327c843b3ca Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 04:51:02 +0000
Subject: [PATCH 49/68] test(channel-monitor): quota mode
 unit/integration/migration coverage
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- fetcher：海外/CN coding/CN payg 分派、账号缺失、凭据失效标记、
  TTL 命中与失败不缓存、UsageInfo→tiers 全窗口归一、状态推导矩阵
- quota mode：RunCheck 三分派（quota 单行 / quota_probe 仅挂主行 /
  probe 不变）、配额失败不翻探活状态、校验矩阵、关联账号复核
  （quota 报错 probe 自动解绑）、quota→probe 切换强制重填 key、
  Duplicate 空明文重加密
- settings：channel_monitor_show_quota 缺省关闭 + 仅 "true" 开启
- 迁移内容断言（仿 176 grok 迁移测试）
- repo 集成测试：check_mode/account_id 往返、quota JSONB 回读、
  探活旧行 NULL 兼容（testcontainers PG 实跑迁移 226）
- 修复 Fetch 在 nil receiver 上的 panic（缓存查询先于原 nil 守卫）
---
 .../channel_monitor_quota_integration_test.go | 155 ++++++
 .../service/channel_monitor_quota_fetcher.go  |  15 +-
 .../channel_monitor_quota_fetcher_test.go     | 392 +++++++++++++++
 .../channel_monitor_quota_mode_test.go        | 463 ++++++++++++++++++
 .../channel_monitor_service_grok_test.go      |   3 +
 .../service/setting_service_public_test.go    |  23 +
 ...annel_monitor_quota_mode_migration_test.go |  37 ++
 7 files changed, 1083 insertions(+), 5 deletions(-)
 create mode 100644 backend/internal/repository/channel_monitor_quota_integration_test.go
 create mode 100644 backend/internal/service/channel_monitor_quota_fetcher_test.go
 create mode 100644 backend/internal/service/channel_monitor_quota_mode_test.go
 create mode 100644 backend/migrations/channel_monitor_quota_mode_migration_test.go

diff --git a/backend/internal/repository/channel_monitor_quota_integration_test.go b/backend/internal/repository/channel_monitor_quota_integration_test.go
new file mode 100644
index 000000000000..2ffc2ae69280
--- /dev/null
+++ b/backend/internal/repository/channel_monitor_quota_integration_test.go
@@ -0,0 +1,155 @@
+//go:build integration
+
+package repository
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/domain"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/stretchr/testify/require"
+)
+
+// 配额模式 repo 层集成测试：
+//   - Create/GetByID/Update 的 check_mode/account_id 往返
+//   - InsertHistoryBatch → ListHistory / ListLatestForMonitorIDs 的 quota JSONB 回读
+//     （裸 SQL 列 + scanMonitorQuota），以及探活模式旧行 quota=NULL 的兼容
+//
+// 注意 channelMonitorRepository 的 GetByID/裸 SQL 走全局 client（不识别 tx ctx），
+// 因此本文件用 integrationEntClient 直连 + t.Cleanup 显式清理，不走 testEntTx 回滚。
+
+func TestChannelMonitorQuotaModeRoundTrip(t *testing.T) {
+	ctx := context.Background()
+	repo := NewChannelMonitorRepository(integrationEntClient, integrationDB)
+
+	account := mustCreateAccount(t, integrationEntClient, &service.Account{
+		Name: "quota-linked-kimi", Platform: domain.PlatformKimi, Type: service.AccountTypeAPIKey,
+		Credentials: map[string]any{"api_key": "sk-kimi", "account_mode": service.AccountModeCoding},
+	})
+	t.Cleanup(func() {
+		_ = integrationEntClient.Account.DeleteOneID(account.ID).Exec(ctx)
+	})
+
+	created := &service.ChannelMonitor{
+		Name:             "kimi-quota-roundtrip",
+		Provider:         service.MonitorProviderKimi,
+		APIMode:          service.MonitorAPIModeChatCompletions,
+		Endpoint:         "",
+		APIKey:           "encrypted-empty",
+		PrimaryModel:     "quota",
+		Enabled:          true,
+		IntervalSeconds:  60,
+		CheckMode:        service.MonitorCheckModeQuota,
+		AccountID:        &account.ID,
+		BodyOverrideMode: service.MonitorBodyOverrideModeOff,
+	}
+	require.NoError(t, repo.Create(ctx, created))
+	t.Cleanup(func() {
+		_ = repo.Delete(ctx, created.ID)
+	})
+
+	loaded, err := repo.GetByID(ctx, created.ID)
+	require.NoError(t, err)
+	require.Equal(t, service.MonitorCheckModeQuota, loaded.CheckMode)
+	require.NotNil(t, loaded.AccountID)
+	require.Equal(t, account.ID, *loaded.AccountID)
+
+	// Update：切换模式并清空关联账号（probe 化）。
+	loaded.CheckMode = service.MonitorCheckModeProbe
+	loaded.AccountID = nil
+	loaded.Endpoint = "https://api.moonshot.cn"
+	require.NoError(t, repo.Update(ctx, loaded))
+
+	reloaded, err := repo.GetByID(ctx, created.ID)
+	require.NoError(t, err)
+	require.Equal(t, service.MonitorCheckModeProbe, reloaded.CheckMode)
+	require.Nil(t, reloaded.AccountID)
+
+	// 重新绑定账号。
+	reloaded.CheckMode = service.MonitorCheckModeQuotaProbe
+	reloaded.AccountID = &account.ID
+	require.NoError(t, repo.Update(ctx, reloaded))
+	final, err := repo.GetByID(ctx, created.ID)
+	require.NoError(t, err)
+	require.Equal(t, service.MonitorCheckModeQuotaProbe, final.CheckMode)
+	require.NotNil(t, final.AccountID)
+}
+
+func TestChannelMonitorHistoryQuotaRoundTrip(t *testing.T) {
+	ctx := context.Background()
+	repo := NewChannelMonitorRepository(integrationEntClient, integrationDB)
+
+	monitor := &service.ChannelMonitor{
+		Name:             "quota-history-roundtrip",
+		Provider:         service.MonitorProviderOpenAI,
+		APIMode:          service.MonitorAPIModeChatCompletions,
+		Endpoint:         "https://api.openai.com",
+		APIKey:           "encrypted",
+		PrimaryModel:     "gpt-test",
+		ExtraModels:      []string{"gpt-extra"},
+		Enabled:          true,
+		IntervalSeconds:  60,
+		BodyOverrideMode: service.MonitorBodyOverrideModeOff,
+	}
+	require.NoError(t, repo.Create(ctx, monitor))
+	t.Cleanup(func() {
+		_ = repo.Delete(ctx, monitor.ID) // histories 级联删除
+	})
+
+	now := time.Now().UTC()
+	rows := []*service.ChannelMonitorHistoryRow{
+		{
+			MonitorID: monitor.ID, Model: "gpt-test", Status: service.MonitorStatusOperational,
+			Message: "ok", CheckedAt: now,
+			Quota: &domain.MonitorQuotaSnapshot{
+				Source:    "usage",
+				Success:   true,
+				PlanLevel: "PRO",
+				Tiers: []domain.MonitorQuotaTier{
+					{Window: "5h", UsedPercent: 42.5, Used: 17, Limit: 40, ResetAt: now.Add(time.Hour).Format(time.RFC3339)},
+				},
+				FetchedAt: now,
+			},
+		},
+		{
+			// 探活模式旧行：无 quota（NULL 兼容）。
+			MonitorID: monitor.ID, Model: "gpt-extra", Status: service.MonitorStatusOperational,
+			Message: "ok", CheckedAt: now,
+		},
+	}
+	require.NoError(t, repo.InsertHistoryBatch(ctx, rows))
+
+	history, err := repo.ListHistory(ctx, monitor.ID, "", 10)
+	require.NoError(t, err)
+	require.Len(t, history, 2)
+
+	byModel := map[string]*service.ChannelMonitorHistoryEntry{}
+	for _, entry := range history {
+		byModel[entry.Model] = entry
+	}
+	withQuota := byModel["gpt-test"]
+	require.NotNil(t, withQuota.Quota)
+	require.True(t, withQuota.Quota.Success)
+	require.Equal(t, "usage", withQuota.Quota.Source)
+	require.Equal(t, "PRO", withQuota.Quota.PlanLevel)
+	require.Len(t, withQuota.Quota.Tiers, 1)
+	require.Equal(t, "5h", withQuota.Quota.Tiers[0].Window)
+	require.InDelta(t, 42.5, withQuota.Quota.Tiers[0].UsedPercent, 0.001)
+	require.Nil(t, byModel["gpt-extra"].Quota, "probe rows must read back as NULL quota")
+
+	// 用户视图聚合：主模型最近一行带快照。
+	latest, err := repo.ListLatestForMonitorIDs(ctx, []int64{monitor.ID})
+	require.NoError(t, err)
+	primaryRows := latest[monitor.ID]
+	require.NotEmpty(t, primaryRows)
+	var primaryQuota *domain.MonitorQuotaSnapshot
+	for _, row := range primaryRows {
+		if row.Model == "gpt-test" {
+			primaryQuota = row.Quota
+		}
+	}
+	require.NotNil(t, primaryQuota, "ListLatestForMonitorIDs must surface quota for the primary model")
+	require.True(t, primaryQuota.Success)
+}
diff --git a/backend/internal/service/channel_monitor_quota_fetcher.go b/backend/internal/service/channel_monitor_quota_fetcher.go
index d208dadc1b3b..9fa990337f0a 100644
--- a/backend/internal/service/channel_monitor_quota_fetcher.go
+++ b/backend/internal/service/channel_monitor_quota_fetcher.go
@@ -100,6 +100,11 @@ func (f *ChannelMonitorQuotaFetcher) LoadAccount(ctx context.Context, id int64)
 // Fetch 抓取账号的最新配额快照。永不返回 error：失败降级为
 // Success=false 快照（Error 带摘要），保证检测历史的时间线连续。
 func (f *ChannelMonitorQuotaFetcher) Fetch(ctx context.Context, accountID int64) *domain.MonitorQuotaSnapshot {
+	if f == nil {
+		// fail-closed：fetcher 未注入（存量测试构造）时不 panic，降级为错误快照。
+		return quotaErrorSnapshot("usage", "quota fetcher is not configured", time.Now())
+	}
+
 	now := time.Now()
 
 	if cached, ok := f.cachedSnapshot(accountID, now); ok {
@@ -171,11 +176,11 @@ func (f *ChannelMonitorQuotaFetcher) fetchUsage(ctx context.Context, accountID i
 		}
 	}
 	snapshot := &domain.MonitorQuotaSnapshot{
-		Source:          "usage",
-		Success:         true,
-		PlanLevel:       usage.SubscriptionTier,
-		Tiers:           usageQuotaTiers(usage),
-		FetchedAt:       now,
+		Source:    "usage",
+		Success:   true,
+		PlanLevel: usage.SubscriptionTier,
+		Tiers:     usageQuotaTiers(usage),
+		FetchedAt: now,
 	}
 	if snapshot.PlanLevel == "" {
 		snapshot.PlanLevel = usage.SubscriptionTierRaw
diff --git a/backend/internal/service/channel_monitor_quota_fetcher_test.go b/backend/internal/service/channel_monitor_quota_fetcher_test.go
new file mode 100644
index 000000000000..5807058d9053
--- /dev/null
+++ b/backend/internal/service/channel_monitor_quota_fetcher_test.go
@@ -0,0 +1,392 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/domain"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/xai"
+	"github.com/stretchr/testify/require"
+)
+
+// --- fetcher 依赖 stub ---
+
+type stubMonitorUsageSource struct {
+	usage   *UsageInfo
+	err     error
+	calls   int
+	lastCtx context.Context
+}
+
+func (s *stubMonitorUsageSource) GetUsage(ctx context.Context, accountID int64, force ...bool) (*UsageInfo, error) {
+	s.calls++
+	s.lastCtx = ctx
+	return s.usage, s.err
+}
+
+type stubMonitorCNQuotaSource struct {
+	result *CNProviderQuotaProbeResult
+	err    error
+	calls  int
+}
+
+func (s *stubMonitorCNQuotaSource) QueryUsage(ctx context.Context, accountID int64) (*CNProviderQuotaProbeResult, error) {
+	s.calls++
+	return s.result, s.err
+}
+
+type stubMonitorCNBalanceSource struct {
+	result *CNProviderBalanceResult
+	err    error
+	calls  int
+}
+
+func (s *stubMonitorCNBalanceSource) QueryBalance(ctx context.Context, accountID int64) (*CNProviderBalanceResult, error) {
+	s.calls++
+	return s.result, s.err
+}
+
+type stubMonitorAccountSource struct {
+	accounts map[int64]*Account
+	err      error
+	calls    int
+}
+
+func (s *stubMonitorAccountSource) GetByID(ctx context.Context, id int64) (*Account, error) {
+	s.calls++
+	if s.err != nil {
+		return nil, s.err
+	}
+	return s.accounts[id], nil
+}
+
+func newQuotaFetcherTestSetup(t *testing.T) (*ChannelMonitorQuotaFetcher, *stubMonitorUsageSource, *stubMonitorCNQuotaSource, *stubMonitorCNBalanceSource, *stubMonitorAccountSource) {
+	t.Helper()
+	usage := &stubMonitorUsageSource{}
+	cnQuota := &stubMonitorCNQuotaSource{}
+	cnBalance := &stubMonitorCNBalanceSource{}
+	accounts := &stubMonitorAccountSource{accounts: make(map[int64]*Account)}
+	fetcher := &ChannelMonitorQuotaFetcher{
+		usage:     usage,
+		cnQuota:   cnQuota,
+		cnBalance: cnBalance,
+		accounts:  accounts,
+		cache:     make(map[int64]monitorQuotaCacheEntry),
+	}
+	return fetcher, usage, cnQuota, cnBalance, accounts
+}
+
+// --- 分派 ---
+
+func TestQuotaFetcher_OverseasAccountUsesUsageService(t *testing.T) {
+	fetcher, usage, _, cnQuota, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[7] = &Account{ID: 7, Platform: domain.PlatformAnthropic}
+	resets := time.Now().Add(2 * time.Hour).UTC()
+	usage.usage = &UsageInfo{
+		FiveHour:         &UsageProgress{Utilization: 42.5, UsedRequests: 17, LimitRequests: 40, ResetsAt: &resets},
+		SevenDay:         &UsageProgress{Utilization: 10},
+		SubscriptionTier: "PRO",
+	}
+
+	snapshot := fetcher.Fetch(context.Background(), 7)
+
+	require.True(t, snapshot.Success)
+	require.Equal(t, "usage", snapshot.Source)
+	require.Equal(t, "PRO", snapshot.PlanLevel)
+	require.False(t, snapshot.CredentialInvalid)
+	require.Empty(t, snapshot.Error)
+	require.Len(t, snapshot.Tiers, 2)
+
+	fiveHour := snapshot.Tiers[0]
+	require.Equal(t, "5h", fiveHour.Window)
+	require.Empty(t, fiveHour.Label)
+	require.InDelta(t, 42.5, fiveHour.UsedPercent, 0.001)
+	require.Equal(t, float64(17), fiveHour.Used)
+	require.Equal(t, float64(40), fiveHour.Limit)
+	require.NotEmpty(t, fiveHour.ResetAt)
+
+	require.Equal(t, "7d", snapshot.Tiers[1].Window)
+	require.Equal(t, 1, usage.calls)
+	require.Equal(t, 0, cnQuota.calls)
+}
+
+func TestQuotaFetcher_CodingPlanAccountUsesCNQuota(t *testing.T) {
+	fetcher, _, cnQuota, cnBalance, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[9] = &Account{
+		ID:          9,
+		Platform:    domain.PlatformKimi,
+		Credentials: map[string]any{"account_mode": AccountModeCoding},
+	}
+	cnQuota.result = &CNProviderQuotaProbeResult{
+		Success:         true,
+		CredentialValid: true,
+		PlanLevel:       "",
+		Tiers: []CNQuotaTier{
+			{Window: "5h", UsedPercent: 33.3, ResetAt: "2026-08-18T06:00:00Z"},
+			{Window: "weekly", UsedPercent: 12},
+		},
+	}
+
+	snapshot := fetcher.Fetch(context.Background(), 9)
+
+	require.True(t, snapshot.Success)
+	require.Equal(t, "cn_quota", snapshot.Source)
+	require.Len(t, snapshot.Tiers, 2)
+	require.Equal(t, "5h", snapshot.Tiers[0].Window)
+	require.InDelta(t, 33.3, snapshot.Tiers[0].UsedPercent, 0.001)
+	require.Equal(t, "weekly", snapshot.Tiers[1].Window)
+	require.Equal(t, 1, cnQuota.calls)
+	require.Equal(t, 0, cnBalance.calls)
+}
+
+func TestQuotaFetcher_PayGAccountUsesCNBalance(t *testing.T) {
+	fetcher, _, _, cnBalance, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[11] = &Account{
+		ID:          11,
+		Platform:    domain.PlatformDeepseek,
+		Credentials: map[string]any{"account_mode": AccountModePayG},
+	}
+	cnBalance.result = &CNProviderBalanceResult{
+		Success:  true,
+		Balance:  12.34,
+		Currency: "CNY",
+		Balances: []CNProviderBalanceEntry{
+			{Currency: "CNY", Balance: 12.34},
+			{Currency: "USD", Balance: 1.5},
+		},
+	}
+
+	snapshot := fetcher.Fetch(context.Background(), 11)
+
+	require.True(t, snapshot.Success)
+	require.Equal(t, "cn_balance", snapshot.Source)
+	require.NotNil(t, snapshot.Balance)
+	require.InDelta(t, 12.34, *snapshot.Balance, 0.001)
+	require.Equal(t, "CNY", snapshot.Currency)
+	require.Len(t, snapshot.Balances, 2)
+	require.Equal(t, "USD", snapshot.Balances[1].Currency)
+	require.Empty(t, snapshot.Error)
+}
+
+// --- 失败路径（Fetch 永不返回 error） ---
+
+func TestQuotaFetcher_AccountMissingYieldsLinkedAccountSnapshot(t *testing.T) {
+	fetcher, usage, _, _, accounts := newQuotaFetcherTestSetup(t)
+	accounts.err = errors.New("not found")
+
+	snapshot := fetcher.Fetch(context.Background(), 404)
+
+	require.False(t, snapshot.Success)
+	require.Equal(t, "linked account not found", snapshot.Error)
+	require.Equal(t, 0, usage.calls) // 未走到数据源
+}
+
+func TestQuotaFetcher_UsageAuthErrorMarksCredentialInvalid(t *testing.T) {
+	fetcher, usage, _, _, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[3] = &Account{ID: 3, Platform: domain.PlatformOpenAI}
+	usage.err = errors.New("API returned 401: unauthorized")
+
+	snapshot := fetcher.Fetch(context.Background(), 3)
+
+	require.False(t, snapshot.Success)
+	require.True(t, snapshot.CredentialInvalid)
+	require.Contains(t, snapshot.Error, "401")
+}
+
+func TestQuotaFetcher_CNQuotaCredentialInvalidFlagPropagates(t *testing.T) {
+	fetcher, _, cnQuota, _, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[5] = &Account{
+		ID:          5,
+		Platform:    domain.PlatformZhipu,
+		Credentials: map[string]any{"account_mode": AccountModeCoding},
+	}
+	cnQuota.result = &CNProviderQuotaProbeResult{Success: false, CredentialValid: false, Error: "api key expired"}
+
+	snapshot := fetcher.Fetch(context.Background(), 5)
+
+	require.False(t, snapshot.Success)
+	require.True(t, snapshot.CredentialInvalid)
+	require.Equal(t, "api key expired", snapshot.Error)
+}
+
+func TestQuotaFetcher_CNBalanceHTTP403MarksCredentialInvalid(t *testing.T) {
+	fetcher, _, _, cnBalance, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[6] = &Account{ID: 6, Platform: domain.PlatformKimi}
+	cnBalance.result = &CNProviderBalanceResult{Success: false, StatusCode: 403, Error: "forbidden"}
+
+	snapshot := fetcher.Fetch(context.Background(), 6)
+
+	require.False(t, snapshot.Success)
+	require.True(t, snapshot.CredentialInvalid)
+}
+
+func TestQuotaFetcher_NilDependenciesProduceErrorSnapshots(t *testing.T) {
+	// fetcher 本体为 nil：直接降级为错误快照，不 panic。
+	var nilFetcher *ChannelMonitorQuotaFetcher
+	snapshot := nilFetcher.Fetch(context.Background(), 1)
+	require.False(t, snapshot.Success)
+	require.Equal(t, "quota fetcher is not configured", snapshot.Error)
+
+	// 数据源缺失：账号能加载，但对应服务未注入。
+	fetcher, _, _, _, accounts := newQuotaFetcherTestSetup(t)
+	fetcher.usage = nil
+	accounts.accounts[2] = &Account{ID: 2, Platform: domain.PlatformOpenAI}
+	snapshot = fetcher.Fetch(context.Background(), 2)
+	require.False(t, snapshot.Success)
+	require.Contains(t, snapshot.Error, "not configured")
+}
+
+// --- TTL 缓存 ---
+
+func TestQuotaFetcher_CachesSuccessSnapshotPerAccount(t *testing.T) {
+	fetcher, usage, _, _, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[8] = &Account{ID: 8, Platform: domain.PlatformOpenAI}
+	usage.usage = &UsageInfo{FiveHour: &UsageProgress{Utilization: 10}}
+
+	for i := 0; i < 3; i++ {
+		snapshot := fetcher.Fetch(context.Background(), 8)
+		require.True(t, snapshot.Success)
+	}
+	require.Equal(t, 1, usage.calls, "success snapshots should be served from cache")
+
+	// 缓存过期后重新拉取。
+	fetcher.mu.Lock()
+	entry := fetcher.cache[8]
+	entry.expiry = time.Now().Add(-time.Second)
+	fetcher.cache[8] = entry
+	fetcher.mu.Unlock()
+
+	_ = fetcher.Fetch(context.Background(), 8)
+	require.Equal(t, 2, usage.calls)
+}
+
+func TestQuotaFetcher_DoesNotCacheFailures(t *testing.T) {
+	fetcher, usage, _, _, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[4] = &Account{ID: 4, Platform: domain.PlatformOpenAI}
+	usage.err = errors.New("boom")
+
+	_ = fetcher.Fetch(context.Background(), 4)
+	_ = fetcher.Fetch(context.Background(), 4)
+
+	require.Equal(t, 2, usage.calls, "failed snapshots must not be cached")
+}
+
+// --- UsageInfo → tiers 归一 ---
+
+func TestUsageQuotaTiers_MapsAllWindowKinds(t *testing.T) {
+	limit := int64(1000)
+	remaining := int64(400)
+	resetUnix := int64(1777283883)
+	usage := &UsageInfo{
+		FiveHour:           &UsageProgress{Utilization: 50},
+		SevenDay:           &UsageProgress{Utilization: 60},
+		SevenDaySonnet:     &UsageProgress{Utilization: 70},
+		SevenDayFable:      &UsageProgress{Utilization: 80},
+		ThirtyDay:          &UsageProgress{Utilization: 20},
+		GeminiSharedDaily:  &UsageProgress{Utilization: 11},
+		GeminiProDaily:     &UsageProgress{Utilization: 22},
+		GeminiFlashDaily:   &UsageProgress{Utilization: 33},
+		GrokRequestQuota:   &xai.QuotaWindow{Limit: &limit, Remaining: &remaining, ResetUnix: &resetUnix},
+		GrokTokenQuota:     &xai.QuotaWindow{Limit: &limit, Remaining: &remaining, ResetAt: "2026-08-19T00:00:00Z"},
+		AntigravityQuota: map[string]*AntigravityModelQuota{
+			"gemini-3-pro": {Utilization: 45},
+			"gemini-3-flash": {Utilization: 55},
+		},
+	}
+
+	tiers := usageQuotaTiers(usage)
+
+	// 5h/7d/7d-sonnet/7d-fable/30d + gemini×3 + grok×2 + antigravity×2
+	require.Len(t, tiers, 12)
+
+	byKey := make(map[string]domain.MonitorQuotaTier, len(tiers))
+	for _, tier := range tiers {
+		key := tier.Window
+		if tier.Label != "" {
+			key = tier.Window + "/" + tier.Label
+		}
+		byKey[key] = tier
+	}
+
+	require.Contains(t, byKey, "5h")
+	require.Contains(t, byKey, "7d")
+	require.Contains(t, byKey, "7d-sonnet")
+	require.Contains(t, byKey, "7d-fable")
+	require.Contains(t, byKey, "30d")
+	require.Contains(t, byKey, "daily/shared")
+	require.Contains(t, byKey, "daily/pro")
+	require.Contains(t, byKey, "daily/flash")
+	require.Contains(t, byKey, "daily/requests")
+	require.Contains(t, byKey, "daily/tokens")
+	require.Contains(t, byKey, "total/gemini-3-pro")
+	require.Contains(t, byKey, "total/gemini-3-flash")
+
+	// grok requests 窗口：used = limit - remaining，百分比 60%。
+	requests := byKey["daily/requests"]
+	require.Equal(t, float64(600), requests.Used)
+	require.Equal(t, float64(1000), requests.Limit)
+	require.InDelta(t, 60.0, requests.UsedPercent, 0.001)
+	require.NotEmpty(t, requests.ResetAt, "ResetUnix should fall back to RFC3339")
+
+	tokens := byKey["daily/tokens"]
+	require.Equal(t, "2026-08-19T00:00:00Z", tokens.ResetAt)
+}
+
+func TestUsageQuotaTiers_NilAndEmptyInputs(t *testing.T) {
+	require.Nil(t, usageQuotaTiers(nil))
+	require.Nil(t, usageQuotaTiers(&UsageInfo{}))
+
+	// Grok 窗口 limit<=0 时跳过，避免除零。
+	var zero int64
+	tiers := usageQuotaTiers(&UsageInfo{
+		GrokRequestQuota: &xai.QuotaWindow{Limit: &zero, Remaining: &zero},
+	})
+	require.Nil(t, tiers)
+}
+
+// --- 状态推导 ---
+
+func TestDeriveQuotaCheckResult_StatusMatrix(t *testing.T) {
+	now := time.Now()
+
+	healthy := &domain.MonitorQuotaSnapshot{Success: true, Tiers: []domain.MonitorQuotaTier{{Window: "5h", UsedPercent: 40}}}
+	res := deriveQuotaCheckResult(healthy, "quota", now)
+	require.Equal(t, MonitorStatusOperational, res.Status)
+	require.Equal(t, "quota", res.Model)
+	require.Empty(t, res.Message)
+
+	highUsage := &domain.MonitorQuotaSnapshot{Success: true, Tiers: []domain.MonitorQuotaTier{
+		{Window: "5h", UsedPercent: 30},
+		{Window: "daily", Label: "pro", UsedPercent: 95},
+	}}
+	res = deriveQuotaCheckResult(highUsage, "quota", now)
+	require.Equal(t, MonitorStatusDegraded, res.Status)
+	require.Contains(t, res.Message, "pro/daily")
+	require.Contains(t, res.Message, "95.0%")
+
+	balance := -0.5
+	depleted := &domain.MonitorQuotaSnapshot{Success: true, Balance: &balance, Currency: "CNY"}
+	res = deriveQuotaCheckResult(depleted, "quota", now)
+	require.Equal(t, MonitorStatusDegraded, res.Status)
+	require.Contains(t, res.Message, "balance depleted")
+
+	invalid := &domain.MonitorQuotaSnapshot{Success: false, CredentialInvalid: true, Error: "401 unauthorized"}
+	res = deriveQuotaCheckResult(invalid, "quota", now)
+	require.Equal(t, MonitorStatusFailed, res.Status)
+
+	unlinked := &domain.MonitorQuotaSnapshot{Success: false, Error: "linked account not found"}
+	res = deriveQuotaCheckResult(unlinked, "quota", now)
+	require.Equal(t, MonitorStatusDegraded, res.Status)
+
+	other := &domain.MonitorQuotaSnapshot{Success: false, Error: "connection refused"}
+	res = deriveQuotaCheckResult(other, "quota", now)
+	require.Equal(t, MonitorStatusError, res.Status)
+	require.Equal(t, "connection refused", res.Message)
+
+	res = deriveQuotaCheckResult(nil, "quota", now)
+	require.Equal(t, MonitorStatusError, res.Status)
+}
diff --git a/backend/internal/service/channel_monitor_quota_mode_test.go b/backend/internal/service/channel_monitor_quota_mode_test.go
new file mode 100644
index 000000000000..0e4b3ea7da98
--- /dev/null
+++ b/backend/internal/service/channel_monitor_quota_mode_test.go
@@ -0,0 +1,463 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/domain"
+	"github.com/stretchr/testify/require"
+)
+
+// --- repo / fetcher 装配 ---
+
+// quotaModeRepoStub 记录 RunCheck 落库行为（历史行 + MarkChecked）。
+type quotaModeRepoStub struct {
+	ChannelMonitorRepository
+	monitor   *ChannelMonitor
+	history   []*ChannelMonitorHistoryRow
+	markedIDs []int64
+	updated   []*ChannelMonitor
+}
+
+func (r *quotaModeRepoStub) GetByID(_ context.Context, id int64) (*ChannelMonitor, error) {
+	if r.monitor == nil || r.monitor.ID != id {
+		return nil, ErrChannelMonitorNotFound
+	}
+	clone := *r.monitor
+	return &clone, nil
+}
+
+func (r *quotaModeRepoStub) InsertHistoryBatch(_ context.Context, rows []*ChannelMonitorHistoryRow) error {
+	r.history = append(r.history, rows...)
+	return nil
+}
+
+func (r *quotaModeRepoStub) MarkChecked(_ context.Context, id int64, _ time.Time) error {
+	r.markedIDs = append(r.markedIDs, id)
+	return nil
+}
+
+func (r *quotaModeRepoStub) Update(_ context.Context, m *ChannelMonitor) error {
+	clone := *m
+	r.updated = append(r.updated, &clone)
+	return nil
+}
+
+// newQuotaModeService 构造启用 V1 探活的 service（复用 retirement/duplicate 测试的 stub）。
+func newQuotaModeService(repo *quotaModeRepoStub) *ChannelMonitorService {
+	svc := NewChannelMonitorService(repo, &duplicateChannelMonitorEncryptor{})
+	svc.SetRuntimeReader(channelMonitorRuntimeStub{rt: ChannelMonitorRuntime{
+		Enabled: true,
+		Mode:    ChannelMonitorModeV1,
+	}})
+	return svc
+}
+
+func newQuotaModeFetcher(accounts map[int64]*Account, usage *stubMonitorUsageSource) *ChannelMonitorQuotaFetcher {
+	if accounts == nil {
+		accounts = make(map[int64]*Account)
+	}
+	if usage == nil {
+		usage = &stubMonitorUsageSource{}
+	}
+	return &ChannelMonitorQuotaFetcher{
+		usage:    usage,
+		accounts: &stubMonitorAccountSource{accounts: accounts},
+		cache:    make(map[int64]monitorQuotaCacheEntry),
+	}
+}
+
+// --- RunCheck 分派 ---
+
+func TestRunCheck_QuotaModeProducesSingleQuotaResult(t *testing.T) {
+	repo := &quotaModeRepoStub{monitor: &ChannelMonitor{
+		ID:              1,
+		Name:            "kimi-quota",
+		Provider:        MonitorProviderKimi,
+		APIMode:         MonitorAPIModeChatCompletions,
+		PrimaryModel:    "quota",
+		Enabled:         true,
+		IntervalSeconds: 60,
+		CheckMode:       MonitorCheckModeQuota,
+		AccountID:       int64Ptr(9),
+	}}
+	svc := newQuotaModeService(repo)
+	fetcher := newQuotaModeFetcher(map[int64]*Account{
+		9: {ID: 9, Platform: domain.PlatformKimi, Credentials: map[string]any{"account_mode": AccountModeCoding}},
+	}, nil)
+	fetcher.cnQuota = &stubMonitorCNQuotaSource{result: &CNProviderQuotaProbeResult{
+		Success:         true,
+		CredentialValid: true,
+		Tiers:           []CNQuotaTier{{Window: "5h", UsedPercent: 30}},
+	}}
+	svc.SetQuotaFetcher(fetcher)
+
+	results, err := svc.RunCheck(context.Background(), 1)
+	require.NoError(t, err)
+	require.Len(t, results, 1)
+
+	res := results[0]
+	require.Equal(t, "quota", res.Model)
+	require.Equal(t, MonitorStatusOperational, res.Status)
+	require.Nil(t, res.LatencyMs)
+	require.Nil(t, res.PingLatencyMs)
+	require.NotNil(t, res.Quota)
+	require.True(t, res.Quota.Success)
+	require.Equal(t, "cn_quota", res.Quota.Source)
+
+	// 历史行携带配额快照，并推进 last_checked_at。
+	require.Len(t, repo.history, 1)
+	require.Equal(t, "quota", repo.history[0].Model)
+	require.NotNil(t, repo.history[0].Quota)
+	require.Equal(t, []int64{1}, repo.markedIDs)
+}
+
+func TestRunCheck_QuotaModeUnlinkedAccountDegrades(t *testing.T) {
+	repo := &quotaModeRepoStub{monitor: &ChannelMonitor{
+		ID:              2,
+		Provider:        MonitorProviderDeepseek,
+		APIMode:         MonitorAPIModeChatCompletions,
+		Endpoint:        "",
+		PrimaryModel:    "quota",
+		Enabled:         true,
+		IntervalSeconds: 60,
+		CheckMode:       MonitorCheckModeQuota,
+		AccountID:       nil, // FK ON DELETE SET NULL 后的形态
+	}}
+	svc := newQuotaModeService(repo)
+	svc.SetQuotaFetcher(newQuotaModeFetcher(nil, nil))
+
+	results, err := svc.RunCheck(context.Background(), 2)
+	require.NoError(t, err)
+	require.Len(t, results, 1)
+	require.Equal(t, MonitorStatusDegraded, results[0].Status)
+	require.Contains(t, results[0].Message, "linked account not found")
+	require.False(t, results[0].Quota.Success)
+}
+
+func TestRunCheck_QuotaModeNilFetcherFailsClosed(t *testing.T) {
+	repo := &quotaModeRepoStub{monitor: &ChannelMonitor{
+		ID:              3,
+		Provider:        MonitorProviderZhipu,
+		APIMode:         MonitorAPIModeChatCompletions,
+		PrimaryModel:    "quota",
+		Enabled:         true,
+		IntervalSeconds: 60,
+		CheckMode:       MonitorCheckModeQuota,
+		AccountID:       int64Ptr(5),
+	}}
+	svc := newQuotaModeService(repo) // 不注入 fetcher
+
+	results, err := svc.RunCheck(context.Background(), 3)
+	require.NoError(t, err)
+	require.Len(t, results, 1)
+	require.Equal(t, MonitorStatusError, results[0].Status)
+	require.Contains(t, results[0].Message, "not configured")
+}
+
+func TestRunCheck_QuotaProbeAttachesSnapshotToPrimaryRowOnly(t *testing.T) {
+	h := &openAICaptureHandler{}
+	endpoint := setupFakeOpenAI(t, h)
+	repo := &quotaModeRepoStub{monitor: &ChannelMonitor{
+		ID:              4,
+		Provider:        MonitorProviderOpenAI,
+		APIMode:         MonitorAPIModeChatCompletions,
+		Endpoint:        endpoint,
+		APIKey:          "OLD:sk-openai",
+		PrimaryModel:    "gpt-test",
+		ExtraModels:     []string{"gpt-extra"},
+		Enabled:         true,
+		IntervalSeconds: 60,
+		CheckMode:       MonitorCheckModeQuotaProbe,
+		AccountID:       int64Ptr(12),
+	}}
+	svc := newQuotaModeService(repo)
+	usage := &stubMonitorUsageSource{usage: &UsageInfo{
+		FiveHour: &UsageProgress{Utilization: 20},
+	}}
+	svc.SetQuotaFetcher(newQuotaModeFetcher(map[int64]*Account{
+		12: {ID: 12, Platform: domain.PlatformOpenAI},
+	}, usage))
+
+	results, err := svc.RunCheck(context.Background(), 4)
+	require.NoError(t, err)
+	require.Len(t, results, 2)
+
+	// 探活状态为准，配额只挂主模型行。
+	require.Equal(t, MonitorStatusOperational, results[0].Status)
+	require.NotNil(t, results[0].Quota)
+	require.True(t, results[0].Quota.Success)
+	require.Equal(t, "usage", results[0].Quota.Source)
+	require.Nil(t, results[1].Quota, "extra model rows must not carry quota")
+
+	// 历史落库时同样只有主模型行带快照。
+	require.Len(t, repo.history, 2)
+	require.NotNil(t, repo.history[0].Quota)
+	require.Equal(t, "gpt-test", repo.history[0].Model)
+	require.Nil(t, repo.history[1].Quota)
+}
+
+func TestRunCheck_QuotaProbeQuotaFailureKeepsProbeStatus(t *testing.T) {
+	h := &openAICaptureHandler{}
+	endpoint := setupFakeOpenAI(t, h)
+	repo := &quotaModeRepoStub{monitor: &ChannelMonitor{
+		ID:              5,
+		Provider:        MonitorProviderOpenAI,
+		APIMode:         MonitorAPIModeChatCompletions,
+		Endpoint:        endpoint,
+		APIKey:          "OLD:sk-openai",
+		PrimaryModel:    "gpt-test",
+		Enabled:         true,
+		IntervalSeconds: 60,
+		CheckMode:       MonitorCheckModeQuotaProbe,
+		AccountID:       nil, // 配额侧失效
+	}}
+	svc := newQuotaModeService(repo)
+	svc.SetQuotaFetcher(newQuotaModeFetcher(nil, nil))
+
+	results, err := svc.RunCheck(context.Background(), 5)
+	require.NoError(t, err)
+	require.Len(t, results, 1)
+	require.Equal(t, MonitorStatusOperational, results[0].Status, "quota failure must not flip probe status")
+	require.False(t, results[0].Quota.Success)
+}
+
+// --- attachQuotaSnapshot 细节 ---
+
+func TestAttachQuotaSnapshot_NoteOnlyWhenProbeMessageEmpty(t *testing.T) {
+	results := []*CheckResult{
+		{Model: "primary", Status: MonitorStatusOperational, Message: "challenge passed"},
+		{Model: "extra"},
+	}
+	failed := &domain.MonitorQuotaSnapshot{Success: false, Error: "boom"}
+
+	attachQuotaSnapshot(results, failed)
+
+	require.Equal(t, "challenge passed", results[0].Message, "existing probe message wins")
+	require.Equal(t, failed, results[0].Quota)
+	require.Nil(t, results[1].Quota)
+
+	quiet := []*CheckResult{{Model: "primary", Status: MonitorStatusOperational}}
+	attachQuotaSnapshot(quiet, failed)
+	require.Contains(t, quiet[0].Message, "quota fetch failed: boom")
+
+	attachQuotaSnapshot(nil, failed)  // 空结果不 panic
+	attachQuotaSnapshot(results, nil) // 空快照不动结果
+}
+
+// --- 校验矩阵 ---
+
+func TestValidateCreateParams_CheckModeMatrix(t *testing.T) {
+	accountID := int64(9)
+
+	cases := []struct {
+		name    string
+		params  ChannelMonitorCreateParams
+		wantErr error
+	}{
+		{
+			name: "probe requires endpoint",
+			params: ChannelMonitorCreateParams{
+				Provider: MonitorProviderOpenAI, CheckMode: MonitorCheckModeProbe,
+				APIKey: "sk", IntervalSeconds: 60, PrimaryModel: "gpt-5",
+			},
+			wantErr: ErrChannelMonitorInvalidEndpoint,
+		},
+		{
+			name: "probe requires api key",
+			params: ChannelMonitorCreateParams{
+				Provider: MonitorProviderOpenAI, CheckMode: MonitorCheckModeProbe,
+				Endpoint: "https://api.openai.com", IntervalSeconds: 60, PrimaryModel: "gpt-5",
+			},
+			wantErr: ErrChannelMonitorMissingAPIKey,
+		},
+		{
+			name: "quota drops endpoint and api key requirements",
+			params: ChannelMonitorCreateParams{
+				Provider: MonitorProviderAntigravity, CheckMode: MonitorCheckModeQuota,
+				IntervalSeconds: 60, AccountID: &accountID,
+			},
+			wantErr: nil, // primary_model 默认 "quota"
+		},
+		{
+			name: "quota requires account",
+			params: ChannelMonitorCreateParams{
+				Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeQuota,
+				IntervalSeconds: 60, PrimaryModel: "quota",
+			},
+			wantErr: ErrChannelMonitorAccountRequired,
+		},
+		{
+			name: "quota_probe requires endpoint and api key too",
+			params: ChannelMonitorCreateParams{
+				Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeQuotaProbe,
+				IntervalSeconds: 60, AccountID: &accountID, PrimaryModel: "kimi-k2",
+			},
+			wantErr: ErrChannelMonitorInvalidEndpoint,
+		},
+		{
+			name: "antigravity probe unsupported",
+			params: ChannelMonitorCreateParams{
+				Provider: MonitorProviderAntigravity, CheckMode: MonitorCheckModeProbe,
+				Endpoint: "https://example.com", APIKey: "k",
+				IntervalSeconds: 60, AccountID: &accountID, PrimaryModel: "gemini-3-pro",
+			},
+			wantErr: ErrChannelMonitorInvalidCheckMode,
+		},
+		{
+			name: "antigravity quota_probe unsupported",
+			params: ChannelMonitorCreateParams{
+				Provider: MonitorProviderAntigravity, CheckMode: MonitorCheckModeQuotaProbe,
+				Endpoint: "https://example.com", APIKey: "k",
+				IntervalSeconds: 60, AccountID: &accountID, PrimaryModel: "gemini-3-pro",
+			},
+			wantErr: ErrChannelMonitorInvalidCheckMode,
+		},
+		{
+			name: "unknown mode rejected",
+			params: ChannelMonitorCreateParams{
+				Provider: MonitorProviderOpenAI, CheckMode: "auto",
+				Endpoint: "https://api.openai.com", APIKey: "sk",
+				IntervalSeconds: 60, PrimaryModel: "gpt-5",
+			},
+			wantErr: ErrChannelMonitorInvalidCheckMode,
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			err := validateCreateParams(tc.params)
+			if tc.wantErr == nil {
+				require.NoError(t, err)
+			} else {
+				require.ErrorIs(t, err, tc.wantErr)
+			}
+		})
+	}
+}
+
+func TestNormalizeMonitorPrimaryModel_QuotaDefault(t *testing.T) {
+	require.Equal(t, "quota", normalizeMonitorPrimaryModel(MonitorProviderKimi, MonitorCheckModeQuota, ""))
+	require.Equal(t, "quota", normalizeMonitorPrimaryModel(MonitorProviderAntigravity, MonitorCheckModeQuota, "  "))
+	// 探活模式沿用原语义：grok 默认模型，其余必填（空串报错在 validateCreateParams）。
+	require.Equal(t, MonitorDefaultGrokModel, normalizeMonitorPrimaryModel(MonitorProviderGrok, MonitorCheckModeProbe, ""))
+	require.Equal(t, "kimi-k2", normalizeMonitorPrimaryModel(MonitorProviderKimi, MonitorCheckModeQuotaProbe, "kimi-k2"))
+}
+
+func TestProviderProbeCapabilityMatrix(t *testing.T) {
+	require.False(t, providerSupportsProbe(MonitorProviderAntigravity))
+	for _, p := range []string{
+		MonitorProviderOpenAI, MonitorProviderAnthropic, MonitorProviderGemini,
+		MonitorProviderGrok, MonitorProviderKimi, MonitorProviderZhipu, MonitorProviderDeepseek,
+	} {
+		require.True(t, providerSupportsProbe(p), p)
+	}
+	for _, p := range []string{
+		MonitorProviderOpenAI, MonitorProviderAnthropic, MonitorProviderGemini,
+		MonitorProviderGrok, MonitorProviderAntigravity,
+		MonitorProviderKimi, MonitorProviderZhipu, MonitorProviderDeepseek,
+	} {
+		require.NoError(t, validateProvider(p), p)
+	}
+}
+
+// --- 关联账号校验 ---
+
+func TestValidateLinkedAccount_Matrix(t *testing.T) {
+	svc := NewChannelMonitorService(nil, nil)
+	fetcher := newQuotaModeFetcher(map[int64]*Account{
+		1: {ID: 1, Platform: domain.PlatformKimi},
+	}, nil)
+	svc.SetQuotaFetcher(fetcher)
+
+	require.NoError(t, svc.validateLinkedAccount(context.Background(), MonitorProviderKimi, nil))
+	require.NoError(t, svc.validateLinkedAccount(context.Background(), MonitorProviderKimi, int64Ptr(0)))
+	require.NoError(t, svc.validateLinkedAccount(context.Background(), MonitorProviderKimi, int64Ptr(1)))
+	require.ErrorIs(t, svc.validateLinkedAccount(context.Background(), MonitorProviderZhipu, int64Ptr(1)), ErrChannelMonitorProviderIncompatible)
+	require.ErrorIs(t, svc.validateLinkedAccount(context.Background(), MonitorProviderKimi, int64Ptr(404)), ErrChannelMonitorAccountRequired)
+
+	noFetcher := NewChannelMonitorService(nil, nil)
+	require.ErrorIs(t, noFetcher.validateLinkedAccount(context.Background(), MonitorProviderKimi, int64Ptr(1)), ErrChannelMonitorAccountRequired)
+}
+
+func TestRevalidateLinkedAccount_QuotaErrorsProbeUnbinds(t *testing.T) {
+	fetcher := newQuotaModeFetcher(nil, nil) // 账号一律加载失败
+	svc := NewChannelMonitorService(nil, nil)
+	svc.SetQuotaFetcher(fetcher)
+
+	quota := &ChannelMonitor{Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeQuota, AccountID: int64Ptr(9)}
+	require.ErrorIs(t, svc.revalidateLinkedAccount(context.Background(), quota), ErrChannelMonitorAccountRequired)
+	require.NotNil(t, quota.AccountID)
+
+	probe := &ChannelMonitor{Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeProbe, AccountID: int64Ptr(9)}
+	require.NoError(t, svc.revalidateLinkedAccount(context.Background(), probe))
+	require.Nil(t, probe.AccountID, "probe mode should silently unbind stale account")
+
+	quotaNoAccount := &ChannelMonitor{Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeQuota}
+	require.ErrorIs(t, svc.revalidateLinkedAccount(context.Background(), quotaNoAccount), ErrChannelMonitorAccountRequired)
+}
+
+func TestRevalidateLinkedAccount_PlatformMismatch(t *testing.T) {
+	svc := NewChannelMonitorService(nil, nil)
+	svc.SetQuotaFetcher(newQuotaModeFetcher(map[int64]*Account{
+		2: {ID: 2, Platform: domain.PlatformDeepseek},
+	}, nil))
+
+	quota := &ChannelMonitor{Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeQuota, AccountID: int64Ptr(2)}
+	require.ErrorIs(t, svc.revalidateLinkedAccount(context.Background(), quota), ErrChannelMonitorProviderIncompatible)
+
+	probe := &ChannelMonitor{Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeProbe, AccountID: int64Ptr(2)}
+	require.NoError(t, svc.revalidateLinkedAccount(context.Background(), probe))
+	require.Nil(t, probe.AccountID)
+}
+
+// --- quota → probe 切换的 key 管控（validateProbeAPIKey） ---
+
+func TestValidateProbeAPIKey_QuotaToProbeRequiresFreshKey(t *testing.T) {
+	svc := NewChannelMonitorService(nil, &duplicateChannelMonitorEncryptor{})
+
+	quota := &ChannelMonitor{Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeQuota, APIKey: "NEW:"}
+	require.NoError(t, svc.validateProbeAPIKey(quota, "")) // quota 模式不管 key
+
+	quota.CheckMode = MonitorCheckModeProbe
+	// 存量密文解出空明文（quota 监控存的加密空串）→ 必须重填 key。
+	require.ErrorIs(t, svc.validateProbeAPIKey(quota, ""), ErrChannelMonitorMissingAPIKey)
+	// 提供新明文 key → 放行。
+	require.NoError(t, svc.validateProbeAPIKey(quota, "sk-fresh"))
+	// 密文解出非空明文 → 放行。
+	require.NoError(t, svc.validateProbeAPIKey(
+		&ChannelMonitor{Provider: MonitorProviderKimi, CheckMode: MonitorCheckModeProbe, APIKey: "OLD:sk-live"}, ""))
+}
+
+// --- Duplicate：quota 模式空明文重加密 ---
+
+func TestDuplicateChannelMonitorQuotaModeReencryptsEmptyKey(t *testing.T) {
+	accountID := int64(9)
+	source := &ChannelMonitor{
+		ID:              42,
+		Name:            "kimi-quota",
+		Provider:        MonitorProviderKimi,
+		APIMode:         MonitorAPIModeChatCompletions,
+		Endpoint:        "",
+		APIKey:          "OLD:", // 解密为空串（quota 监控的加密空 key）
+		PrimaryModel:    "quota",
+		Enabled:         true,
+		IntervalSeconds: 60,
+		CheckMode:       MonitorCheckModeQuota,
+		AccountID:       &accountID,
+	}
+	repo := &duplicateChannelMonitorRepoStub{source: source}
+	service := NewChannelMonitorService(repo, &duplicateChannelMonitorEncryptor{})
+
+	dup, err := service.Duplicate(context.Background(), 42, 7, "admin:7", "op-1")
+	require.NoError(t, err)
+	require.Equal(t, MonitorCheckModeQuota, dup.CheckMode)
+	require.NotNil(t, dup.AccountID)
+	require.Equal(t, accountID, *dup.AccountID)
+	require.Empty(t, dup.APIKey, "plaintext stays empty for quota monitors")
+	require.Len(t, repo.created, 1)
+	require.Equal(t, "NEW:", repo.created[0].APIKey, "empty key must be re-encrypted, not dropped")
+}
diff --git a/backend/internal/service/channel_monitor_service_grok_test.go b/backend/internal/service/channel_monitor_service_grok_test.go
index 20c9db266663..e3a89075f485 100644
--- a/backend/internal/service/channel_monitor_service_grok_test.go
+++ b/backend/internal/service/channel_monitor_service_grok_test.go
@@ -9,6 +9,7 @@ func TestApplyMonitorUpdate_ProviderOnlySwitchToGrokUsesDefaultModel(t *testing.
 	existing := &ChannelMonitor{
 		Provider:        MonitorProviderOpenAI,
 		APIMode:         MonitorAPIModeResponses,
+		Endpoint:        "https://api.openai.com/v1",
 		PrimaryModel:    "gpt-5",
 		IntervalSeconds: 60,
 	}
@@ -31,6 +32,7 @@ func TestApplyMonitorUpdate_SwitchToGrokPreservesExplicitModel(t *testing.T) {
 	existing := &ChannelMonitor{
 		Provider:        MonitorProviderOpenAI,
 		APIMode:         MonitorAPIModeChatCompletions,
+		Endpoint:        "https://api.openai.com/v1",
 		PrimaryModel:    "gpt-5",
 		IntervalSeconds: 60,
 	}
@@ -52,6 +54,7 @@ func TestApplyMonitorUpdate_SameGrokProviderDoesNotResetExistingModel(t *testing
 	existing := &ChannelMonitor{
 		Provider:        MonitorProviderGrok,
 		APIMode:         MonitorAPIModeChatCompletions,
+		Endpoint:        "https://api.x.ai",
 		PrimaryModel:    "grok-4.3",
 		IntervalSeconds: 60,
 	}
diff --git a/backend/internal/service/setting_service_public_test.go b/backend/internal/service/setting_service_public_test.go
index 67bab243ca48..3fa0d40426b6 100644
--- a/backend/internal/service/setting_service_public_test.go
+++ b/backend/internal/service/setting_service_public_test.go
@@ -116,6 +116,29 @@ func TestSettingService_ChannelMonitorHideThroughputDefaultsToPrivate(t *testing
 	}
 }
 
+func TestSettingService_ChannelMonitorShowQuotaFailsClosed(t *testing.T) {
+	// 缺省（迁移插入 'false' / 老库无行）一律不展示。
+	missingRuntime := NewSettingService(&settingPublicRepoStub{values: map[string]string{}}, &config.Config{}).GetChannelMonitorRuntime(context.Background())
+	require.False(t, missingRuntime.ShowQuota)
+	missingPublic, err := NewSettingService(&settingPublicRepoStub{values: map[string]string{}}, &config.Config{}).
+		GetPublicSettings(context.Background())
+	require.NoError(t, err)
+	require.False(t, missingPublic.ChannelMonitorShowQuota)
+
+	// 仅字面 "true" 视为开启；其余值（含异常值）fail-closed。
+	runtime := NewSettingService(&settingPublicRepoStub{values: map[string]string{
+		SettingKeyChannelMonitorShowQuota: "true",
+	}}, &config.Config{}).GetChannelMonitorRuntime(context.Background())
+	require.True(t, runtime.ShowQuota)
+
+	for _, value := range []string{"false", "TRUE", "1", "yes", "on", "garbage"} {
+		rt := NewSettingService(&settingPublicRepoStub{values: map[string]string{
+			SettingKeyChannelMonitorShowQuota: value,
+		}}, &config.Config{}).GetChannelMonitorRuntime(context.Background())
+		require.False(t, rt.ShowQuota, "value=%q", value)
+	}
+}
+
 func TestSettingService_GetPublicSettings_ExposesForceEmailOnThirdPartySignup(t *testing.T) {
 	repo := &settingPublicRepoStub{
 		values: map[string]string{
diff --git a/backend/migrations/channel_monitor_quota_mode_migration_test.go b/backend/migrations/channel_monitor_quota_mode_migration_test.go
new file mode 100644
index 000000000000..69d2dacaef22
--- /dev/null
+++ b/backend/migrations/channel_monitor_quota_mode_migration_test.go
@@ -0,0 +1,37 @@
+package migrations
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestChannelMonitorQuotaModeMigration(t *testing.T) {
+	content, err := FS.ReadFile("226_channel_monitor_quota_mode.sql")
+	require.NoError(t, err)
+
+	sql := strings.Join(strings.Fields(string(content)), " ")
+
+	// provider CHECK 两张表扩到 8 平台，且带幂等守卫（仿 176 grok 迁移）。
+	require.Contains(t, sql, "channel_monitors_provider_check")
+	require.Contains(t, sql, "channel_monitor_request_templates_provider_check")
+	require.Contains(t, sql, "CHECK (provider IN ('openai', 'anthropic', 'gemini', 'grok', 'antigravity', 'kimi', 'zhipu', 'deepseek'))")
+	require.Contains(t, sql, "position('kimi' IN monitor_constraint_def) = 0")
+	require.Contains(t, sql, "position('kimi' IN template_constraint_def) = 0")
+
+	// check_mode 三态，默认 probe。
+	require.Contains(t, sql, "ADD COLUMN IF NOT EXISTS check_mode VARCHAR(32) NOT NULL DEFAULT 'probe'")
+	require.Contains(t, sql, "CHECK (check_mode IN ('probe', 'quota', 'quota_probe'))")
+
+	// account_id 关联账号，账号删除置空（监控保留，运行时报「账号未关联」）。
+	require.Contains(t, sql, "ADD COLUMN IF NOT EXISTS account_id BIGINT REFERENCES accounts(id) ON DELETE SET NULL")
+	require.Contains(t, sql, "CREATE INDEX IF NOT EXISTS idx_channel_monitors_account_id ON channel_monitors(account_id)")
+
+	// 历史表配额快照列。
+	require.Contains(t, sql, "ADD COLUMN IF NOT EXISTS quota JSONB")
+
+	// 公开设置默认关闭。
+	require.Contains(t, sql, "VALUES ('channel_monitor_show_quota', 'false')")
+	require.Contains(t, sql, "ON CONFLICT (key) DO NOTHING")
+}

From c51fd7d0b351da9df3459da280778d30b182a24e Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 04:51:06 +0000
Subject: [PATCH 50/68] test(channel-monitor): adapt checker body test to 3-arg
 normalizeMonitorPrimaryModel

---
 backend/internal/service/channel_monitor_checker_body_test.go | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/backend/internal/service/channel_monitor_checker_body_test.go b/backend/internal/service/channel_monitor_checker_body_test.go
index 3375ee6ac998..235764633705 100644
--- a/backend/internal/service/channel_monitor_checker_body_test.go
+++ b/backend/internal/service/channel_monitor_checker_body_test.go
@@ -199,7 +199,7 @@ func TestGrokMonitorConfiguration(t *testing.T) {
 	if err := validateProvider(MonitorProviderGrok); err != nil {
 		t.Fatalf("grok provider should be supported: %v", err)
 	}
-	if got := normalizeMonitorPrimaryModel(MonitorProviderGrok, ""); got != MonitorDefaultGrokModel {
+	if got := normalizeMonitorPrimaryModel(MonitorProviderGrok, MonitorCheckModeProbe, ""); got != MonitorDefaultGrokModel {
 		t.Fatalf("expected default Grok model %q, got %q", MonitorDefaultGrokModel, got)
 	}
 	if err := validateAPIMode(MonitorProviderGrok, MonitorAPIModeChatCompletions); err != nil {

From bb6c3b4f6a0543788dac7cafb8025846ae293709 Mon Sep 17 00:00:00 2001
From: spongehah <2635879218@qq.com>
Date: Tue, 18 Aug 2026 13:03:20 +0800
Subject: [PATCH 51/68] fix: unify Codex OAuth outbound identity onto the
 inference resolver

Token exchange, PAT whoami, models, probes, and pre-writes now follow
the same UA/version chain as Codex inference instead of hardcoded
codex-cli/0.91.0 or compile-time constants.
---
 .../repository/openai_oauth_service.go        |  4 +--
 .../repository/openai_oauth_service_test.go   | 11 ++++++++
 .../internal/service/account_test_service.go  | 13 +++++++---
 .../internal/service/account_usage_service.go |  9 ++++---
 .../internal/service/openai_alpha_search.go   | 19 +++++++-------
 .../internal/service/openai_codex_identity.go | 25 ++++++++++++++++++-
 .../service/openai_codex_identity_test.go     | 23 +++++++++++++++++
 .../service/openai_codex_models_service.go    | 14 +++++++----
 .../openai_codex_models_service_test.go       |  8 +++---
 .../service/openai_codex_pat_service.go       |  4 +--
 .../service/openai_codex_pat_service_test.go  |  2 +-
 .../service/openai_gateway_forward.go         |  6 ++---
 .../service/openai_gateway_passthrough.go     |  7 +++---
 .../service/openai_ws_forwarder_payload.go    |  2 +-
 14 files changed, 106 insertions(+), 41 deletions(-)

diff --git a/backend/internal/repository/openai_oauth_service.go b/backend/internal/repository/openai_oauth_service.go
index acb270a3b33f..32b1652f6554 100644
--- a/backend/internal/repository/openai_oauth_service.go
+++ b/backend/internal/repository/openai_oauth_service.go
@@ -48,7 +48,7 @@ func (s *openaiOAuthService) ExchangeCode(ctx context.Context, code, codeVerifie
 
 	resp, err := client.R().
 		SetContext(ctx).
-		SetHeader("User-Agent", "codex-cli/0.91.0").
+		SetHeader("User-Agent", service.CodexCanonicalUserAgent()).
 		SetFormDataFromValues(formData).
 		SetSuccessResult(&tokenResp).
 		Post(s.tokenURL)
@@ -96,7 +96,7 @@ func (s *openaiOAuthService) refreshTokenWithClientID(ctx context.Context, refre
 
 	resp, err := client.R().
 		SetContext(ctx).
-		SetHeader("User-Agent", "codex-cli/0.91.0").
+		SetHeader("User-Agent", service.CodexCanonicalUserAgent()).
 		SetFormDataFromValues(formData).
 		SetSuccessResult(&tokenResp).
 		Post(s.tokenURL)
diff --git a/backend/internal/repository/openai_oauth_service_test.go b/backend/internal/repository/openai_oauth_service_test.go
index b43e2b52fc76..b3feda7483d4 100644
--- a/backend/internal/repository/openai_oauth_service_test.go
+++ b/backend/internal/repository/openai_oauth_service_test.go
@@ -10,6 +10,7 @@ import (
 
 	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
+	"github.com/Wei-Shaw/sub2api/internal/service"
 	"github.com/stretchr/testify/require"
 	"github.com/stretchr/testify/suite"
 )
@@ -77,6 +78,11 @@ func (s *OpenAIOAuthServiceSuite) TestExchangeCode_DefaultRedirectURI() {
 			w.WriteHeader(http.StatusBadRequest)
 			return
 		}
+		if got := r.Header.Get("User-Agent"); got != service.CodexCanonicalUserAgent() {
+			errCh <- "user-agent mismatch"
+			w.WriteHeader(http.StatusBadRequest)
+			return
+		}
 
 		w.Header().Set("Content-Type", "application/json")
 		_, _ = io.WriteString(w, `{"access_token":"at","refresh_token":"rt","token_type":"bearer","expires_in":3600}`)
@@ -121,6 +127,11 @@ func (s *OpenAIOAuthServiceSuite) TestRefreshToken_FormFields() {
 			w.WriteHeader(http.StatusBadRequest)
 			return
 		}
+		if got := r.Header.Get("User-Agent"); got != service.CodexCanonicalUserAgent() {
+			errCh <- "user-agent mismatch"
+			w.WriteHeader(http.StatusBadRequest)
+			return
+		}
 
 		w.Header().Set("Content-Type", "application/json")
 		_, _ = io.WriteString(w, `{"access_token":"at2","refresh_token":"rt2","token_type":"bearer","expires_in":3600}`)
diff --git a/backend/internal/service/account_test_service.go b/backend/internal/service/account_test_service.go
index 496ef57aed85..8e21ba149f58 100644
--- a/backend/internal/service/account_test_service.go
+++ b/backend/internal/service/account_test_service.go
@@ -731,11 +731,12 @@ func (s *AccountTestService) testOpenAIAccountConnection(c *gin.Context, account
 		req.Host = "chatgpt.com"
 		req.Header.Set("accept", "text/event-stream")
 		req.Header.Set("OpenAI-Beta", "responses=experimental")
-		req.Header.Set("Originator", openai.CodexDefaultOriginator)
+		canonical := resolveCodexOutboundIdentity("")
+		req.Header.Set("Originator", canonical.originator)
 		if customUA := strings.TrimSpace(credentialAccount.GetOpenAIUserAgent()); customUA != "" {
 			req.Header.Set("User-Agent", customUA)
 		} else {
-			req.Header.Set("User-Agent", codexCLIUserAgent)
+			req.Header.Set("User-Agent", canonical.userAgent)
 		}
 		setOpenAIChatGPTAccountHeaders(req.Header, credentialAccount)
 		// 与真实转发一致：账号级自定义 UA 同样作为管理员显式配置传入，否则测试用的身份
@@ -2063,6 +2064,9 @@ func (s *AccountTestService) testOpenAICompactConnection(c *gin.Context, account
 		req.Header.Set("Authorization", "Bearer "+authToken)
 	}
 	applyOpenAICodexProbeHeaders(req.Header)
+	if isOAuth {
+		enforceCodexIdentityHeadersWithUA(req.Header, credentialAccount.GetOpenAIUserAgent())
+	}
 	probeSessionID := compactProbeSessionID(account.ID)
 	req.Header.Set("Session_ID", probeSessionID)
 	req.Header.Set("Conversation_ID", probeSessionID)
@@ -2968,11 +2972,12 @@ func (s *AccountTestService) testOpenAIImageOAuth(c *gin.Context, ctx context.Co
 	req.Header.Set("Content-Type", "application/json")
 	req.Header.Set("Accept", "text/event-stream")
 	req.Header.Set("OpenAI-Beta", "responses=experimental")
-	req.Header.Set("originator", openai.CodexDefaultOriginator)
+	canonical := resolveCodexOutboundIdentity("")
+	req.Header.Set("originator", canonical.originator)
 	if customUA := strings.TrimSpace(credentialAccount.GetOpenAIUserAgent()); customUA != "" {
 		req.Header.Set("User-Agent", customUA)
 	} else {
-		req.Header.Set("User-Agent", codexCLIUserAgent)
+		req.Header.Set("User-Agent", canonical.userAgent)
 	}
 	setOpenAIChatGPTAccountHeaders(req.Header, credentialAccount)
 	// 与真实转发一致：账号级自定义 UA 同样作为管理员显式配置传入，否则测试用的身份
diff --git a/backend/internal/service/account_usage_service.go b/backend/internal/service/account_usage_service.go
index 084d153ff2c0..3cdc71337752 100644
--- a/backend/internal/service/account_usage_service.go
+++ b/backend/internal/service/account_usage_service.go
@@ -113,7 +113,7 @@ const (
 	openAIProbeCacheTTL     = 10 * time.Minute
 	grokProbeRetryTTL       = 1 * time.Minute
 	grokFreeQuotaWindow     = 24 * time.Hour
-	openAICodexProbeVersion = codexCLIVersion // 与网关出站身份同源，避免两处硬编码版本各自漂移
+	openAICodexProbeVersion = codexCLIVersion // 编译期兜底；运行时探针 version 走 CodexCanonicalClientVersion
 )
 
 // UsageCache 封装账户使用量相关的缓存
@@ -860,9 +860,10 @@ func (s *AccountUsageService) probeOpenAICodexSnapshot(ctx context.Context, acco
 	}
 	req.Header.Set("Accept", "text/event-stream")
 	req.Header.Set("OpenAI-Beta", "responses=experimental")
-	req.Header.Set("Originator", openaipkg.CodexDefaultOriginator)
-	req.Header.Set("Version", openAICodexProbeVersion)
-	req.Header.Set("User-Agent", codexCLIUserAgent)
+	canonical := resolveCodexOutboundIdentity("")
+	req.Header.Set("Originator", canonical.originator)
+	req.Header.Set("Version", canonical.version)
+	req.Header.Set("User-Agent", canonical.userAgent)
 	if s.identityCache != nil {
 		if fp, fpErr := s.identityCache.GetFingerprint(reqCtx, account.ID); fpErr == nil && fp != nil && strings.TrimSpace(fp.UserAgent) != "" {
 			req.Header.Set("User-Agent", strings.TrimSpace(fp.UserAgent))
diff --git a/backend/internal/service/openai_alpha_search.go b/backend/internal/service/openai_alpha_search.go
index 0ddc8f525551..7e7a13f3b5c0 100644
--- a/backend/internal/service/openai_alpha_search.go
+++ b/backend/internal/service/openai_alpha_search.go
@@ -11,7 +11,6 @@ import (
 	"strings"
 	"time"
 
-	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	"github.com/gin-gonic/gin"
 	"github.com/tidwall/gjson"
 )
@@ -239,25 +238,26 @@ func (s *OpenAIGatewayService) buildOpenAIAlphaSearchResponsesWebSearchRequest(c
 	if turnMetadata := openAIAlphaSearchInboundHeader(c, "X-Codex-Turn-Metadata"); turnMetadata != "" {
 		req.Header.Set("X-Codex-Turn-Metadata", turnMetadata)
 	}
+	canonical := resolveCodexOutboundIdentity("")
 	if version := openAIAlphaSearchInboundHeader(c, "Version"); version != "" {
 		req.Header.Set("Version", version)
 	} else {
-		req.Header.Set("Version", codexCLIVersion)
+		req.Header.Set("Version", canonical.version)
 	}
 	if originator := openAIAlphaSearchInboundHeader(c, "Originator"); originator != "" {
 		req.Header.Set("Originator", originator)
 	} else {
-		req.Header.Set("Originator", openai.CodexDefaultOriginator)
+		req.Header.Set("Originator", canonical.originator)
 	}
 	if customUA := account.GetOpenAIUserAgent(); customUA != "" {
 		req.Header.Set("User-Agent", customUA)
 	} else if userAgent := openAIAlphaSearchInboundHeader(c, "User-Agent"); userAgent != "" {
 		req.Header.Set("User-Agent", userAgent)
 	} else {
-		req.Header.Set("User-Agent", codexCLIUserAgent)
+		req.Header.Set("User-Agent", canonical.userAgent)
 	}
 	if s.cfg != nil && s.cfg.Gateway.ForceCodexCLI {
-		req.Header.Set("User-Agent", codexCLIUserAgent)
+		req.Header.Set("User-Agent", canonical.userAgent)
 	}
 	apiKeyID := getAPIKeyIDFromContext(c)
 	if sessionID := strings.TrimSpace(gjson.GetBytes(alphaBody, "id").String()); sessionID != "" {
@@ -382,25 +382,26 @@ func (s *OpenAIGatewayService) buildOpenAIAlphaSearchRequest(ctx context.Context
 		if turnMetadata := openAIAlphaSearchInboundHeader(c, "X-Codex-Turn-Metadata"); turnMetadata != "" {
 			req.Header.Set("X-Codex-Turn-Metadata", turnMetadata)
 		}
+		canonical := resolveCodexOutboundIdentity("")
 		if version := openAIAlphaSearchInboundHeader(c, "Version"); version != "" {
 			req.Header.Set("Version", version)
 		} else {
-			req.Header.Set("Version", codexCLIVersion)
+			req.Header.Set("Version", canonical.version)
 		}
 		if originator := openAIAlphaSearchInboundHeader(c, "Originator"); originator != "" {
 			req.Header.Set("Originator", originator)
 		} else {
-			req.Header.Set("Originator", openai.CodexDefaultOriginator)
+			req.Header.Set("Originator", canonical.originator)
 		}
 		if customUA := account.GetOpenAIUserAgent(); customUA != "" {
 			req.Header.Set("User-Agent", customUA)
 		} else if userAgent := openAIAlphaSearchInboundHeader(c, "User-Agent"); userAgent != "" {
 			req.Header.Set("User-Agent", userAgent)
 		} else {
-			req.Header.Set("User-Agent", codexCLIUserAgent)
+			req.Header.Set("User-Agent", canonical.userAgent)
 		}
 		if s.cfg != nil && s.cfg.Gateway.ForceCodexCLI {
-			req.Header.Set("User-Agent", codexCLIUserAgent)
+			req.Header.Set("User-Agent", canonical.userAgent)
 		}
 		enforceCodexIdentityHeadersWithUA(req.Header, s.codexIdentityOverrideUA(account))
 	}
diff --git a/backend/internal/service/openai_codex_identity.go b/backend/internal/service/openai_codex_identity.go
index 510ee07cf29a..7d9edc4e28a6 100644
--- a/backend/internal/service/openai_codex_identity.go
+++ b/backend/internal/service/openai_codex_identity.go
@@ -76,6 +76,29 @@ func SetCodexCanonicalUserAgentResolver(resolver func() string) {
 	codexCanonicalUAResolver = resolver
 }
 
+// CodexCanonicalUserAgent 返回当前生效的规范 Codex User-Agent。
+// 取值走与推理相同的解析链：面板 UA 指纹 + 面板/自动同步版本号 + 编译期兜底。
+// 供无账号句柄的出站路径（OAuth 换 Token / 刷新）使用。
+func CodexCanonicalUserAgent() string {
+	return resolveCodexOutboundIdentity("").userAgent
+}
+
+// ApplyCodexCanonicalIdentity 为无账号句柄的出站请求写入与推理同源的身份三元组。
+func ApplyCodexCanonicalIdentity(h http.Header) {
+	if h == nil {
+		return
+	}
+	identity := resolveCodexOutboundIdentity("")
+	h.Set("user-agent", identity.userAgent)
+	h.Set("originator", identity.originator)
+	h.Set("version", identity.version)
+}
+
+// CodexCanonicalClientVersion 返回当前生效的 Codex 客户端版本号。
+func CodexCanonicalClientVersion() string {
+	return resolveCodexOutboundIdentity("").version
+}
+
 // codexCanonicalUserAgent 返回出站规范 User-Agent。
 func codexCanonicalUserAgent() string {
 	codexCanonicalUAMu.RLock()
@@ -210,6 +233,6 @@ func pairCodexIdentityHeaders(h http.Header) {
 	h.Set("user-agent", pairedUA)
 	h.Set("originator", originator)
 	if v := strings.TrimSpace(h.Get("version")); v != "" && CompareVersions(v, codexUpstreamMinVersion) < 0 {
-		h.Set("version", codexCLIVersion)
+		h.Set("version", resolveCodexOutboundIdentity("").version)
 	}
 }
diff --git a/backend/internal/service/openai_codex_identity_test.go b/backend/internal/service/openai_codex_identity_test.go
index 9b295241ce11..782064643159 100644
--- a/backend/internal/service/openai_codex_identity_test.go
+++ b/backend/internal/service/openai_codex_identity_test.go
@@ -322,3 +322,26 @@ func TestBuildCodexCLIUserAgent(t *testing.T) {
 	require.Equal(t, codexCLIUserAgent, buildCodexCLIUserAgent("bogus version"))
 	require.Equal(t, codexCLIUserAgent, buildCodexCLIUserAgent(""))
 }
+
+func TestCodexCanonicalUserAgentFollowsResolver(t *testing.T) {
+	SetCodexCanonicalUserAgentResolver(func() string {
+		return "codex_cli_rs/0.200.1" + codexCLIUserAgentSuffix
+	})
+	t.Cleanup(func() { SetCodexCanonicalUserAgentResolver(nil) })
+
+	require.Equal(t, "codex_cli_rs/0.200.1"+codexCLIUserAgentSuffix, CodexCanonicalUserAgent())
+	require.Equal(t, "0.200.1", CodexCanonicalClientVersion())
+
+	h := make(http.Header)
+	ApplyCodexCanonicalIdentity(h)
+	require.Equal(t, "codex_cli_rs", h.Get("originator"))
+	require.Equal(t, "codex_cli_rs/0.200.1"+codexCLIUserAgentSuffix, h.Get("user-agent"))
+	require.Equal(t, "0.200.1", h.Get("version"))
+}
+
+func TestCodexCanonicalUserAgentFallsBackWithoutResolver(t *testing.T) {
+	SetCodexCanonicalUserAgentResolver(nil)
+
+	require.Equal(t, codexCLIUserAgent, CodexCanonicalUserAgent())
+	require.Equal(t, codexCLIVersion, CodexCanonicalClientVersion())
+}
diff --git a/backend/internal/service/openai_codex_models_service.go b/backend/internal/service/openai_codex_models_service.go
index c36c46eb134e..8b5bcda91552 100644
--- a/backend/internal/service/openai_codex_models_service.go
+++ b/backend/internal/service/openai_codex_models_service.go
@@ -18,7 +18,6 @@ import (
 
 	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/httpclient"
-	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	"golang.org/x/net/http2"
 	"golang.org/x/sync/singleflight"
 )
@@ -243,7 +242,7 @@ func (s *OpenAIGatewayService) FetchCodexModelsManifest(ctx context.Context, acc
 
 	clientVersion = strings.TrimSpace(clientVersion)
 	if clientVersion == "" {
-		clientVersion = openAICodexProbeVersion
+		clientVersion = CodexCanonicalClientVersion()
 	}
 
 	requestEndpoint := chatgptCodexModelsURL
@@ -305,9 +304,14 @@ func (s *OpenAIGatewayService) FetchCodexModelsManifest(ctx context.Context, acc
 		setOpenAIChatGPTAccountHeaders(headers, credAccount)
 	}
 	headers.Set("Accept", "application/json")
-	headers.Set("Originator", openai.CodexDefaultOriginator)
-	headers.Set("Version", clientVersion)
-	headers.Set("User-Agent", codexCLIUserAgent)
+	overrideUA := ""
+	if !useAPIKeyUpstream {
+		overrideUA = credAccount.GetOpenAIUserAgent()
+	}
+	identity := resolveCodexOutboundIdentity(overrideUA)
+	headers.Set("Originator", identity.originator)
+	headers.Set("User-Agent", identity.userAgent)
+	headers.Set("Version", identity.version)
 
 	proxyURL := ""
 	if account.ProxyID != nil && account.Proxy != nil {
diff --git a/backend/internal/service/openai_codex_models_service_test.go b/backend/internal/service/openai_codex_models_service_test.go
index 585ad95d5e48..c7eb7e29ec52 100644
--- a/backend/internal/service/openai_codex_models_service_test.go
+++ b/backend/internal/service/openai_codex_models_service_test.go
@@ -348,8 +348,8 @@ func TestFetchCodexModelsManifestDefaultClientVersion(t *testing.T) {
 	if _, err := s.FetchCodexModelsManifest(context.Background(), newCodexModelsTestAccount(), "", ""); err != nil {
 		t.Fatalf("FetchCodexModelsManifest returned error: %v", err)
 	}
-	if gotClientVersion != openAICodexProbeVersion {
-		t.Errorf("default client_version: got %q, want %q", gotClientVersion, openAICodexProbeVersion)
+	if gotClientVersion != CodexCanonicalClientVersion() {
+		t.Errorf("default client_version: got %q, want %q", gotClientVersion, CodexCanonicalClientVersion())
 	}
 }
 
@@ -451,10 +451,10 @@ func TestFetchCodexModelsManifestAPIKeyCustomUpstream(t *testing.T) {
 	if gotRequest.Header.Get("Originator") != openai.CodexDefaultOriginator {
 		t.Errorf("originator header: got %q", gotRequest.Header.Get("Originator"))
 	}
-	if gotRequest.Header.Get("Version") != "0.144.0" {
+	if gotRequest.Header.Get("Version") != CodexCanonicalClientVersion() {
 		t.Errorf("version header: got %q", gotRequest.Header.Get("Version"))
 	}
-	if gotRequest.Header.Get("User-Agent") != codexCLIUserAgent {
+	if gotRequest.Header.Get("User-Agent") != CodexCanonicalUserAgent() {
 		t.Errorf("user-agent header: got %q", gotRequest.Header.Get("User-Agent"))
 	}
 	if gotRequest.Header.Get("chatgpt-account-id") != "" {
diff --git a/backend/internal/service/openai_codex_pat_service.go b/backend/internal/service/openai_codex_pat_service.go
index dce6785920f3..f4c4cc9f1138 100644
--- a/backend/internal/service/openai_codex_pat_service.go
+++ b/backend/internal/service/openai_codex_pat_service.go
@@ -10,7 +10,6 @@ import (
 
 	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/httpclient"
-	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 )
 
 const openAICodexPATWhoamiURLDefault = "https://auth.openai.com/api/accounts/v1/user-auth-credential/whoami"
@@ -59,8 +58,7 @@ func (s *OpenAIOAuthService) ValidateCodexPersonalAccessToken(ctx context.Contex
 	}
 	req.Header.Set("authorization", "Bearer "+accessToken)
 	req.Header.Set("accept", "application/json")
-	req.Header.Set("originator", openai.CodexDefaultOriginator)
-	req.Header.Set("user-agent", codexCLIUserAgent)
+	ApplyCodexCanonicalIdentity(req.Header)
 
 	resp, err := client.Do(req)
 	if err != nil {
diff --git a/backend/internal/service/openai_codex_pat_service_test.go b/backend/internal/service/openai_codex_pat_service_test.go
index e7656868ba8e..3210ed3062ec 100644
--- a/backend/internal/service/openai_codex_pat_service_test.go
+++ b/backend/internal/service/openai_codex_pat_service_test.go
@@ -40,7 +40,7 @@ func TestOpenAIOAuthService_ValidateCodexPersonalAccessToken(t *testing.T) {
 	require.NoError(t, err)
 	require.Equal(t, "Bearer at-test-token", gotAuthorization)
 	require.Equal(t, openai.CodexDefaultOriginator, gotOriginator)
-	require.Equal(t, codexCLIUserAgent, gotUserAgent)
+	require.Equal(t, CodexCanonicalUserAgent(), gotUserAgent)
 	require.Equal(t, OpenAIAuthModePersonalAccessToken, info.AuthMode)
 	require.Equal(t, "user@example.com", info.Email)
 	require.Equal(t, "user-123", info.ChatGPTUserID)
diff --git a/backend/internal/service/openai_gateway_forward.go b/backend/internal/service/openai_gateway_forward.go
index 6dab7fb702c3..9d8218a11485 100644
--- a/backend/internal/service/openai_gateway_forward.go
+++ b/backend/internal/service/openai_gateway_forward.go
@@ -1128,7 +1128,7 @@ func (s *OpenAIGatewayService) buildUpstreamRequest(ctx context.Context, c *gin.
 		if isOpenAIResponsesCompactPath(c) {
 			req.Header.Set("accept", "application/json")
 			if req.Header.Get("version") == "" {
-				req.Header.Set("version", codexCLIVersion)
+				req.Header.Set("version", CodexCanonicalClientVersion())
 			}
 			compactSession := resolveOpenAICompactSessionID(c)
 			req.Header.Set("session_id", isolateOpenAISessionID(apiKeyID, compactSession))
@@ -1154,10 +1154,10 @@ func (s *OpenAIGatewayService) buildUpstreamRequest(ctx context.Context, c *gin.
 		req.Header.Set("user-agent", customUA)
 	}
 
-	// 若开启 ForceCodexCLI，则强制将上游 User-Agent 伪装为 Codex CLI。
+	// 若开启 ForceCodexCLI，则强制将上游 User-Agent 伪装为规范 Codex 身份。
 	// 用于网关未透传/改写 User-Agent 时，仍能命中 Codex 侧识别逻辑。
 	if s.cfg != nil && s.cfg.Gateway.ForceCodexCLI {
-		req.Header.Set("user-agent", codexCLIUserAgent)
+		req.Header.Set("user-agent", CodexCanonicalUserAgent())
 	}
 
 	// 指纹收敛：使用 Forward() 中预计算的收敛 ID 改写出站头，与请求体使用同一份 IDs。
diff --git a/backend/internal/service/openai_gateway_passthrough.go b/backend/internal/service/openai_gateway_passthrough.go
index 4bfd2fa26c8b..7440a403510c 100644
--- a/backend/internal/service/openai_gateway_passthrough.go
+++ b/backend/internal/service/openai_gateway_passthrough.go
@@ -17,7 +17,6 @@ import (
 	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
-	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
 	"github.com/gin-gonic/gin"
 	"github.com/tidwall/gjson"
@@ -445,7 +444,7 @@ func (s *OpenAIGatewayService) buildUpstreamRequestOpenAIPassthrough(
 		if isOpenAIResponsesCompactPath(c) {
 			req.Header.Set("accept", "application/json")
 			if req.Header.Get("version") == "" {
-				req.Header.Set("version", codexCLIVersion)
+				req.Header.Set("version", CodexCanonicalClientVersion())
 			}
 			if clientSessionID == "" {
 				clientSessionID = resolveOpenAICompactSessionID(c)
@@ -454,7 +453,7 @@ func (s *OpenAIGatewayService) buildUpstreamRequestOpenAIPassthrough(
 			req.Header.Set("accept", "text/event-stream")
 		}
 		if req.Header.Get("originator") == "" {
-			req.Header.Set("originator", openai.CodexDefaultOriginator)
+			req.Header.Set("originator", resolveCodexOutboundIdentity("").originator)
 		}
 		// 用隔离后的 session 标识符覆盖客户端透传值，防止跨用户会话碰撞。
 		if clientSessionID == "" {
@@ -482,7 +481,7 @@ func (s *OpenAIGatewayService) buildUpstreamRequestOpenAIPassthrough(
 		req.Header.Set("user-agent", customUA)
 	}
 	if s.cfg != nil && s.cfg.Gateway.ForceCodexCLI {
-		req.Header.Set("user-agent", codexCLIUserAgent)
+		req.Header.Set("user-agent", CodexCanonicalUserAgent())
 	}
 	// 指纹收敛：使用 forwardOpenAIPassthrough 中预计算的收敛 ID 改写出站头，
 	// 与请求体 client_metadata 共享同一份 IDs（与非透传路径相同的相对位置：
diff --git a/backend/internal/service/openai_ws_forwarder_payload.go b/backend/internal/service/openai_ws_forwarder_payload.go
index 797361a3a646..7eabe7d31f1a 100644
--- a/backend/internal/service/openai_ws_forwarder_payload.go
+++ b/backend/internal/service/openai_ws_forwarder_payload.go
@@ -161,7 +161,7 @@ func (s *OpenAIGatewayService) buildOpenAIWSHeaders(
 		}
 	}
 	if s != nil && s.cfg != nil && s.cfg.Gateway.ForceCodexCLI {
-		headers.Set("user-agent", codexCLIUserAgent)
+		headers.Set("user-agent", CodexCanonicalUserAgent())
 	}
 	// 终态收口：WS 握手与 HTTP 出站共用同一套身份语义，账号级自定义 UA 同样作为
 	// 管理员显式配置传入（上面写进 headers 的值只在强制统一被关闭时才参与配对）。

From 16e4f7ecc33ad895587a3581b8f8d1c6f3dbf44e Mon Sep 17 00:00:00 2001
From: o2e <24642961+o2e@users.noreply.github.com>
Date: Tue, 18 Aug 2026 13:02:29 +0800
Subject: [PATCH 52/68] =?UTF-8?q?=E4=BF=AE=E5=A4=8D=20Codex=20=E9=A2=9D?=
 =?UTF-8?q?=E5=BA=A6=E6=8E=A2=E9=92=88=E6=A8=A1=E5=9E=8B=E5=85=BC=E5=AE=B9?=
 =?UTF-8?q?=E6=80=A7?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 backend/internal/pkg/openai/constants.go          | 3 +++
 backend/internal/service/account_usage_service.go | 2 +-
 2 files changed, 4 insertions(+), 1 deletion(-)

diff --git a/backend/internal/pkg/openai/constants.go b/backend/internal/pkg/openai/constants.go
index a863a19f8192..772143ac9223 100644
--- a/backend/internal/pkg/openai/constants.go
+++ b/backend/internal/pkg/openai/constants.go
@@ -45,6 +45,9 @@ func DefaultModelIDs() []string {
 // DefaultTestModel default model for testing OpenAI accounts
 const DefaultTestModel = "gpt-5.4"
 
+// CodexUsageProbeModel is the model used for OAuth Codex usage probes.
+const CodexUsageProbeModel = "codex-auto-review"
+
 // DefaultInstructions default instructions for non-Codex CLI requests.
 // 内容为真实 Codex CLI 的 GPT-5-Codex base prompt（codex 系模型默认）。
 //
diff --git a/backend/internal/service/account_usage_service.go b/backend/internal/service/account_usage_service.go
index 084d153ff2c0..5dc1b1b071db 100644
--- a/backend/internal/service/account_usage_service.go
+++ b/backend/internal/service/account_usage_service.go
@@ -830,7 +830,7 @@ func (s *AccountUsageService) probeOpenAICodexSnapshot(ctx context.Context, acco
 	if accessToken == "" && !account.IsOpenAIAgentIdentity() {
 		return nil, fmt.Errorf("no access token available")
 	}
-	modelID := openaipkg.DefaultTestModel
+	modelID := openaipkg.CodexUsageProbeModel
 	payload := createOpenAITestPayload(modelID, true)
 	payloadBytes, err := json.Marshal(payload)
 	if err != nil {

From a20e1c00cc1736378d7510c48ab52cac66489739 Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 05:09:58 +0000
Subject: [PATCH 53/68] =?UTF-8?q?feat(monitor-ui):=20=E9=85=8D=E9=A2=9D?=
 =?UTF-8?q?=E6=A8=A1=E5=BC=8F=E8=A1=A8=E5=8D=95=E3=80=81=E7=94=A8=E9=87=8F?=
 =?UTF-8?q?=E5=BF=AB=E7=85=A7=E8=A7=86=E5=9B=BE=E4=B8=8E=208=20=E5=B9=B3?=
 =?UTF-8?q?=E5=8F=B0=E6=94=AF=E6=8C=81?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

管理端：
- MonitorFormDialog：新增检查方式三态分段控件（探活/配额/探活+配额），
  provider 网格扩展至 8 平台（+antigravity/kimi/zhipu/deepseek），
  国产平台默认 endpoint 预填、antigravity 强制 quota-only；配额模式
  显示关联账号选择器（复用 adminAPI.accounts.list 按平台过滤），
  隐藏 endpoint/api_key/模型等探活专属字段
- 新建共用 MonitorQuotaView：用量窗口进度条（阈值/样式对齐账号页
  CNProviderQuotaCell）、多币种余额、plan_level 徽章与错误态；
  window/label 机器 token 走 i18n，未知 token 原样透出（前向兼容）
- 列表主模型单元格内联 latest_quota；运行结果弹窗逐行展示配额；
  筛选栏/模板管理/格式化 composable 同步 8 平台（配色对齐
  utils/platformColors.ts）；ProviderIcon 搬运 4 家官方 SVG

用户端：
- MonitorCard 支持 latest_quota，渲染受 channel_monitor_show_quota
  系统开关（isChannelMonitorQuotaVisible）控制，服务端剥离为第一道防线

设置与 i18n：
- 管理设置页新增「向用户展示渠道用量/余额」开关（默认关闭）；
  zh/en 全量补充 providers/checkMode/quota 文案
---
 frontend/src/api/admin/channelMonitor.ts      |  65 +++++-
 frontend/src/api/admin/settings.ts            |   2 +
 frontend/src/api/channelMonitor.ts            |   7 +-
 .../admin/monitor/MonitorFiltersBar.vue       |   8 +
 .../admin/monitor/MonitorFormDialog.vue       | 219 ++++++++++++++++--
 .../admin/monitor/MonitorPrimaryModelCell.vue |  13 +-
 .../admin/monitor/MonitorRunResultDialog.vue  |   2 +
 .../monitor/MonitorTemplateManagerDialog.vue  |  16 +-
 .../components/common/MonitorQuotaView.vue    | 156 +++++++++++++
 .../components/user/monitor/MonitorCard.vue   |  14 ++
 .../components/user/monitor/ProviderIcon.vue  |  22 ++
 .../composables/useChannelMonitorFormat.ts    |  56 ++++-
 frontend/src/constants/channelMonitor.ts      |  28 ++-
 .../src/i18n/locales/en/admin/channels.ts     |  12 +
 .../src/i18n/locales/en/admin/settings.ts     |   3 +
 frontend/src/i18n/locales/en/dashboard.ts     |  34 ++-
 .../src/i18n/locales/zh/admin/channels.ts     |  12 +
 .../src/i18n/locales/zh/admin/settings.ts     |   3 +
 frontend/src/i18n/locales/zh/dashboard.ts     |  34 ++-
 frontend/src/types/index.ts                   |   2 +
 frontend/src/utils/featureFlags.ts            |  11 +
 frontend/src/views/admin/SettingsView.vue     |  18 ++
 22 files changed, 700 insertions(+), 37 deletions(-)
 create mode 100644 frontend/src/components/common/MonitorQuotaView.vue

diff --git a/frontend/src/api/admin/channelMonitor.ts b/frontend/src/api/admin/channelMonitor.ts
index 5eb78b798fb0..f3353ebd45c8 100644
--- a/frontend/src/api/admin/channelMonitor.ts
+++ b/frontend/src/api/admin/channelMonitor.ts
@@ -5,10 +5,57 @@
 
 import { apiClient } from '../client'
 
-export type Provider = 'openai' | 'anthropic' | 'gemini' | 'grok'
+export type Provider =
+  | 'openai'
+  | 'anthropic'
+  | 'gemini'
+  | 'grok'
+  | 'antigravity'
+  | 'kimi'
+  | 'zhipu'
+  | 'deepseek'
 export type MonitorStatus = 'operational' | 'degraded' | 'failed' | 'error'
 export type BodyOverrideMode = 'off' | 'merge' | 'replace'
 export type APIMode = 'chat_completions' | 'responses'
+/**
+ * probe = LLM 探活（默认）；quota = 仅查关联账号用量（零 LLM 成本）；
+ * quota_probe = 探活 + 配额快照挂主模型行。
+ */
+export type CheckMode = 'probe' | 'quota' | 'quota_probe'
+
+/** 配额快照中的单个用量窗口（与后端 domain.MonitorQuotaTier 一致）。 */
+export interface MonitorQuotaTier {
+  /** 5h | 7d | 7d-sonnet | 7d-fable | 30d | daily | weekly | total */
+  window: string
+  /** 同窗口多档时的机器标识（gemini shared/pro/flash、grok requests/tokens、antigravity 模型名） */
+  label?: string
+  used_percent: number
+  used?: number
+  limit?: number
+  /** RFC3339；空表示无重置时间 */
+  reset_at?: string
+}
+
+export interface MonitorBalance {
+  currency: string
+  balance: number
+}
+
+/** 归一化配额快照（与后端 domain.MonitorQuotaSnapshot 一致）。 */
+export interface MonitorQuotaSnapshot {
+  /** usage | cn_quota | cn_balance */
+  source: string
+  success: boolean
+  tiers?: MonitorQuotaTier[]
+  balance?: number | null
+  balances?: MonitorBalance[]
+  currency?: string
+  plan_level?: string
+  /** 401/403 鉴权失败标记（推导为 failed 状态） */
+  credential_invalid?: boolean
+  error?: string
+  fetched_at: string
+}
 
 export interface ChannelMonitor {
   id: number
@@ -47,6 +94,12 @@ export interface ChannelMonitor {
   extra_headers: Record<string, string>
   body_override_mode: BodyOverrideMode
   body_override: Record<string, unknown> | null
+  /** 检测模式：probe（默认）/ quota / quota_probe */
+  check_mode: CheckMode
+  /** 配额模式关联的账号 ID；探活模式为 null */
+  account_id: number | null
+  /** 主模型最近一次配额快照（配额模式；无历史时为 null） */
+  latest_quota?: MonitorQuotaSnapshot | null
 }
 
 export interface ExtraModelStatus {
@@ -75,8 +128,14 @@ export interface CreateParams {
   name: string
   provider: Provider
   api_mode?: APIMode
+  /** 探活模式必填（base origin）；quota 模式可留空 */
   endpoint: string
+  /** 探活模式必填；quota 模式可留空 */
   api_key: string
+  /** 缺省 probe；antigravity 仅支持 quota */
+  check_mode?: CheckMode
+  /** 配额模式必填：数据源账号（provider 需与账号平台一致） */
+  account_id?: number | null
   primary_model: string
   extra_models?: string[]
   group_name?: string
@@ -101,6 +160,8 @@ export interface CheckResult {
   ping_latency_ms: number | null
   message: string
   checked_at: string
+  /** 配额模式（quota / quota_probe 主模型行）附带的配额快照 */
+  quota?: MonitorQuotaSnapshot | null
 }
 
 export interface RunNowResponse {
@@ -115,6 +176,8 @@ export interface HistoryItem {
   ping_latency_ms: number | null
   message: string
   checked_at: string
+  /** 配额快照（配额模式行；探活行为空） */
+  quota?: MonitorQuotaSnapshot | null
 }
 
 export interface HistoryParams {
diff --git a/frontend/src/api/admin/settings.ts b/frontend/src/api/admin/settings.ts
index c8dbe8fb78f0..f5f19918a864 100644
--- a/frontend/src/api/admin/settings.ts
+++ b/frontend/src/api/admin/settings.ts
@@ -718,6 +718,7 @@ export interface SystemSettings {
   channel_monitor_mode?: 'v1' | 'v2';
   channel_monitor_default_interval_seconds: number;
   channel_monitor_hide_throughput?: boolean;
+  channel_monitor_show_quota?: boolean;
 
   // Available Channels feature switch
   available_channels_enabled: boolean;
@@ -1016,6 +1017,7 @@ export interface UpdateSettingsRequest {
   channel_monitor_mode?: 'v1' | 'v2';
   channel_monitor_default_interval_seconds?: number;
   channel_monitor_hide_throughput?: boolean;
+  channel_monitor_show_quota?: boolean;
 
   // Available Channels feature switch
   available_channels_enabled?: boolean;
diff --git a/frontend/src/api/channelMonitor.ts b/frontend/src/api/channelMonitor.ts
index 38dd0c99a8ba..68076db77de0 100644
--- a/frontend/src/api/channelMonitor.ts
+++ b/frontend/src/api/channelMonitor.ts
@@ -4,7 +4,7 @@
  */
 
 import { apiClient } from './client'
-import type { Provider, MonitorStatus } from './admin/channelMonitor'
+import type { MonitorQuotaSnapshot, Provider, MonitorStatus } from './admin/channelMonitor'
 
 export type { Provider, MonitorStatus } from './admin/channelMonitor'
 
@@ -33,6 +33,11 @@ export interface UserMonitorView {
   availability_7d: number
   extra_models: UserMonitorExtraModel[]
   timeline: MonitorTimelinePoint[]
+  /**
+   * 主模型最近配额快照。仅当系统开启 channel_monitor_show_quota 时
+   * 服务端才会下发（关闭时服务端已剥离，前端 flag 仅作纵深防御）。
+   */
+  latest_quota?: MonitorQuotaSnapshot | null
 }
 
 export interface UserMonitorListResponse {
diff --git a/frontend/src/components/admin/monitor/MonitorFiltersBar.vue b/frontend/src/components/admin/monitor/MonitorFiltersBar.vue
index 544238f49ce2..d79e374dcea5 100644
--- a/frontend/src/components/admin/monitor/MonitorFiltersBar.vue
+++ b/frontend/src/components/admin/monitor/MonitorFiltersBar.vue
@@ -71,6 +71,10 @@ import {
   PROVIDER_ANTHROPIC,
   PROVIDER_GEMINI,
   PROVIDER_GROK,
+  PROVIDER_ANTIGRAVITY,
+  PROVIDER_KIMI,
+  PROVIDER_ZHIPU,
+  PROVIDER_DEEPSEEK,
 } from '@/constants/channelMonitor'
 
 defineProps<{
@@ -96,6 +100,10 @@ const providerFilterOptions = computed(() => [
   { value: PROVIDER_ANTHROPIC, label: t('monitorCommon.providers.anthropic') },
   { value: PROVIDER_GEMINI, label: t('monitorCommon.providers.gemini') },
   { value: PROVIDER_GROK, label: t('monitorCommon.providers.grok') },
+  { value: PROVIDER_ANTIGRAVITY, label: t('monitorCommon.providers.antigravity') },
+  { value: PROVIDER_KIMI, label: t('monitorCommon.providers.kimi') },
+  { value: PROVIDER_ZHIPU, label: t('monitorCommon.providers.zhipu') },
+  { value: PROVIDER_DEEPSEEK, label: t('monitorCommon.providers.deepseek') },
 ])
 
 const enabledFilterOptions = computed(() => [
diff --git a/frontend/src/components/admin/monitor/MonitorFormDialog.vue b/frontend/src/components/admin/monitor/MonitorFormDialog.vue
index 14a9e2dd151f..b67bcb775a63 100644
--- a/frontend/src/components/admin/monitor/MonitorFormDialog.vue
+++ b/frontend/src/components/admin/monitor/MonitorFormDialog.vue
@@ -11,6 +11,27 @@
         <input v-model="form.name" type="text" required class="input" :placeholder="t('admin.channelMonitor.form.namePlaceholder')" />
       </div>
 
+      <!-- 检测模式：probe（探活）/ quota（仅配额）/ quota_probe（探活+配额） -->
+      <div>
+        <label class="input-label">{{ t('admin.channelMonitor.form.checkMode') }}</label>
+        <div class="grid gap-3 sm:grid-cols-3" data-testid="monitor-check-mode">
+          <button
+            v-for="opt in checkModeOptions"
+            :key="opt.value"
+            type="button"
+            :data-testid="`monitor-check-mode-${opt.value}`"
+            :aria-pressed="form.check_mode === opt.value"
+            :disabled="opt.disabled"
+            class="rounded-lg border-2 px-3 py-2 text-left transition-colors disabled:cursor-not-allowed disabled:opacity-50"
+            :class="checkModeButtonClass(opt.value)"
+            @click="selectCheckMode(opt.value)"
+          >
+            <span class="block text-sm font-semibold">{{ opt.label }}</span>
+            <span class="mt-0.5 block text-xs opacity-80">{{ opt.hint }}</span>
+          </button>
+        </div>
+      </div>
+
       <div>
         <label class="input-label">{{ t('admin.channelMonitor.form.provider') }} <span class="text-red-500">*</span></label>
         <div class="grid grid-cols-2 gap-3 sm:grid-cols-4">
@@ -30,7 +51,23 @@
         </div>
       </div>
 
-      <div v-if="form.provider === PROVIDER_OPENAI" class="rounded-lg border border-blue-100 bg-blue-50/50 p-3 dark:border-blue-500/20 dark:bg-blue-500/10">
+      <!-- 配额模式数据源：关联账号（复用账号侧用量/余额服务） -->
+      <div v-if="usesQuotaMode">
+        <label class="input-label">
+          {{ t('admin.channelMonitor.form.linkedAccount') }} <span class="text-red-500">*</span>
+        </label>
+        <Select
+          v-model="accountSelectValue"
+          :options="accountOptions"
+          :placeholder="t('admin.channelMonitor.form.linkedAccountPlaceholder')"
+        />
+        <p class="mt-1 text-xs text-gray-400">{{ t('admin.channelMonitor.form.linkedAccountHint') }}</p>
+        <p v-if="accountOptions.length === 0 && !accountsLoading" class="mt-1 text-xs text-amber-600 dark:text-amber-400">
+          {{ t('admin.channelMonitor.form.linkedAccountEmpty') }}
+        </p>
+      </div>
+
+      <div v-if="form.provider === PROVIDER_OPENAI && usesProbePart" class="rounded-lg border border-blue-100 bg-blue-50/50 p-3 dark:border-blue-500/20 dark:bg-blue-500/10">
         <label class="input-label">{{ t('admin.channelMonitor.form.apiMode') }}</label>
         <div class="grid gap-3 sm:grid-cols-2">
           <button
@@ -48,7 +85,7 @@
         </div>
       </div>
 
-      <div>
+      <div v-if="usesProbePart">
         <label class="input-label">{{ t('admin.channelMonitor.form.endpoint') }} <span class="text-red-500">*</span></label>
         <div class="flex gap-2">
           <input v-model="form.endpoint" data-testid="monitor-endpoint" type="text" required class="input flex-1" :placeholder="t('admin.channelMonitor.form.endpointPlaceholder')" />
@@ -58,7 +95,7 @@
         </div>
       </div>
 
-      <div>
+      <div v-if="usesProbePart">
         <label class="input-label">
           {{ t('admin.channelMonitor.form.apiKey') }}<span v-if="!editing" class="text-red-500"> *</span>
         </label>
@@ -77,7 +114,7 @@
         <p v-if="editing && editing.api_key_masked" class="mt-1 text-xs text-gray-400">{{ editing.api_key_masked }}</p>
       </div>
 
-      <div>
+      <div v-if="usesProbePart">
         <label class="input-label">{{ t('admin.channelMonitor.form.primaryModel') }} <span class="text-red-500">*</span></label>
         <input
           v-model="form.primary_model"
@@ -90,7 +127,7 @@
         />
       </div>
 
-      <div>
+      <div v-if="usesProbePart">
         <label class="input-label">{{ t('admin.channelMonitor.form.extraModels') }}</label>
         <ModelTagInput
           :models="form.extra_models"
@@ -122,8 +159,8 @@
         <Toggle v-model="form.enabled" />
       </div>
 
-      <!-- 高级设置区：请求模板 + 自定义 headers/body -->
-      <details class="rounded-lg border border-gray-200 bg-gray-50/50 p-3 dark:border-dark-700 dark:bg-dark-900/30">
+      <!-- 高级设置区：请求模板 + 自定义 headers/body（仅探活模式有意义） -->
+      <details v-if="usesProbePart" class="rounded-lg border border-gray-200 bg-gray-50/50 p-3 dark:border-dark-700 dark:bg-dark-900/30">
         <summary class="cursor-pointer text-sm font-medium text-gray-700 dark:text-gray-300">
           {{ t('admin.channelMonitor.advanced.section') }}
         </summary>
@@ -197,6 +234,7 @@ import type {
   ChannelMonitor,
   CreateParams,
   APIMode,
+  CheckMode,
   Provider,
   UpdateParams,
 } from '@/api/admin/channelMonitor'
@@ -216,10 +254,20 @@ import {
   PROVIDER_ANTHROPIC,
   PROVIDER_GEMINI,
   PROVIDER_GROK,
+  PROVIDER_ANTIGRAVITY,
+  PROVIDER_KIMI,
+  PROVIDER_ZHIPU,
+  PROVIDER_DEEPSEEK,
   API_MODE_CHAT_COMPLETIONS,
   API_MODE_RESPONSES,
+  CHECK_MODE_PROBE,
+  CHECK_MODE_QUOTA,
+  CHECK_MODE_QUOTA_PROBE,
   DEFAULT_GROK_ENDPOINT,
   DEFAULT_GROK_MODEL,
+  DEFAULT_KIMI_ENDPOINT,
+  DEFAULT_ZHIPU_ENDPOINT,
+  DEFAULT_DEEPSEEK_ENDPOINT,
   DEFAULT_INTERVAL_SECONDS,
 } from '@/constants/channelMonitor'
 
@@ -259,6 +307,8 @@ interface MonitorForm {
   name: string
   provider: Provider
   api_mode: APIMode
+  check_mode: CheckMode
+  account_id: number | null
   endpoint: string
   api_key: string
   primary_model: string
@@ -278,6 +328,8 @@ const form = reactive<MonitorForm>({
   name: '',
   provider: PROVIDER_ANTHROPIC,
   api_mode: API_MODE_CHAT_COMPLETIONS,
+  check_mode: CHECK_MODE_PROBE,
+  account_id: null,
   endpoint: '',
   api_key: '',
   primary_model: '',
@@ -292,6 +344,10 @@ const form = reactive<MonitorForm>({
   body_override: null,
 })
 
+// quota / quota_probe 需要关联账号；probe / quota_probe 需要探活字段。
+const usesQuotaMode = computed(() => form.check_mode !== CHECK_MODE_PROBE)
+const usesProbePart = computed(() => form.check_mode !== CHECK_MODE_QUOTA)
+
 // jitter 上限与后端校验一致：interval - jitter 不得低于最小检测间隔 15 秒。
 const maxJitterSeconds = computed<number>(() => Math.max(0, (form.interval_seconds || 0) - 15))
 
@@ -402,8 +458,122 @@ const providerOptions = computed<ProviderOption[]>(() => [
   { value: PROVIDER_OPENAI, label: t('monitorCommon.providers.openai') },
   { value: PROVIDER_GEMINI, label: t('monitorCommon.providers.gemini') },
   { value: PROVIDER_GROK, label: t('monitorCommon.providers.grok') },
+  { value: PROVIDER_ANTIGRAVITY, label: t('monitorCommon.providers.antigravity') },
+  { value: PROVIDER_KIMI, label: t('monitorCommon.providers.kimi') },
+  { value: PROVIDER_ZHIPU, label: t('monitorCommon.providers.zhipu') },
+  { value: PROVIDER_DEEPSEEK, label: t('monitorCommon.providers.deepseek') },
+])
+
+// 国产 provider 预填的官方 endpoint（仅探活侧；配额模式 endpoint 可留空）。
+const PROVIDER_DEFAULT_ENDPOINTS: Partial<Record<Provider, string>> = {
+  [PROVIDER_KIMI]: DEFAULT_KIMI_ENDPOINT,
+  [PROVIDER_ZHIPU]: DEFAULT_ZHIPU_ENDPOINT,
+  [PROVIDER_DEEPSEEK]: DEFAULT_DEEPSEEK_ENDPOINT,
+}
+
+interface CheckModeOption {
+  value: CheckMode
+  label: string
+  hint: string
+  disabled: boolean
+}
+
+const checkModeOptions = computed<CheckModeOption[]>(() => [
+  {
+    value: CHECK_MODE_PROBE,
+    label: t('admin.channelMonitor.form.checkModeProbe'),
+    hint: t('admin.channelMonitor.form.checkModeProbeHint'),
+    // antigravity 无探活 adapter，仅配额模式。
+    disabled: form.provider === PROVIDER_ANTIGRAVITY,
+  },
+  {
+    value: CHECK_MODE_QUOTA,
+    label: t('admin.channelMonitor.form.checkModeQuota'),
+    hint: t('admin.channelMonitor.form.checkModeQuotaHint'),
+    disabled: false,
+  },
+  {
+    value: CHECK_MODE_QUOTA_PROBE,
+    label: t('admin.channelMonitor.form.checkModeQuotaProbe'),
+    hint: t('admin.channelMonitor.form.checkModeQuotaProbeHint'),
+    // antigravity 无探活 adapter，只支持配额模式。
+    disabled: form.provider === PROVIDER_ANTIGRAVITY,
+  },
 ])
 
+function checkModeButtonClass(mode: CheckMode): string {
+  const active = form.check_mode === mode
+  if (active) {
+    return 'border-primary-500 bg-white text-primary-700 shadow-sm dark:border-primary-400 dark:bg-primary-500/15 dark:text-primary-300'
+  }
+  return 'border-blue-100 bg-white/70 text-gray-600 hover:border-primary-300 dark:border-dark-700 dark:bg-dark-800 dark:text-gray-400'
+}
+
+function selectCheckMode(mode: CheckMode) {
+  if (checkModeOptions.value.find((opt) => opt.value === mode)?.disabled) return
+  form.check_mode = mode
+  if (!usesQuotaMode.value) form.account_id = null
+}
+
+// --- 关联账号选择器 ---
+
+interface LinkedAccount {
+  id: number
+  name: string
+}
+
+const linkedAccounts = ref<LinkedAccount[]>([])
+const accountsLoading = ref(false)
+
+const accountOptions = computed(() =>
+  linkedAccounts.value.map((a) => ({
+    value: String(a.id),
+    label: `${a.name} (#${a.id})`,
+  })),
+)
+
+// Select 组件绑定 string，与 number | null 互转。
+const accountSelectValue = computed<string>({
+  get: () => (form.account_id == null ? '' : String(form.account_id)),
+  set: (raw: string) => {
+    if (raw === '') {
+      form.account_id = null
+      return
+    }
+    const id = Number(raw)
+    if (Number.isFinite(id)) form.account_id = id
+  },
+})
+
+// 拉取当前 provider 平台的账号（大分页一次取齐，客户端不再过滤）。
+// 失败不阻塞表单：下拉为空 + 空态提示。
+async function loadLinkedAccounts() {
+  if (!usesQuotaMode.value || !props.show) return
+  accountsLoading.value = true
+  try {
+    const res = await adminAPI.accounts.list(1, 200, { platform: form.provider })
+    linkedAccounts.value = (res.items || []).map((a) => ({ id: a.id, name: a.name }))
+    // 关联账号必须与 provider 同平台：平台切换后原选择失效，自动清掉。
+    if (form.account_id != null && !linkedAccounts.value.some((a) => a.id === form.account_id)) {
+      form.account_id = null
+    }
+  } catch (err: unknown) {
+    console.warn('load linked accounts failed', err)
+    linkedAccounts.value = []
+  } finally {
+    accountsLoading.value = false
+  }
+}
+
+watch(
+  () => [props.show, form.provider, form.check_mode] as const,
+  ([show]) => {
+    if (!show) return
+    void loadLinkedAccounts()
+  },
+  { immediate: true },
+)
+
 function selectProvider(provider: Provider) {
   if (form.provider === provider) return
   const previousProvider = form.provider
@@ -411,14 +581,23 @@ function selectProvider(provider: Provider) {
     previousProvider === PROVIDER_GROK && form.endpoint === DEFAULT_GROK_ENDPOINT
   const clearGrokModel =
     previousProvider === PROVIDER_GROK && form.primary_model === DEFAULT_GROK_MODEL
+  const clearPrevDefaultEndpoint =
+    !!PROVIDER_DEFAULT_ENDPOINTS[previousProvider] && form.endpoint === PROVIDER_DEFAULT_ENDPOINTS[previousProvider]
   form.provider = provider
+  form.account_id = null
+  // antigravity 仅配额模式：切到它时强制 quota（checkModeOptions 同步禁用其余项）。
+  if (provider === PROVIDER_ANTIGRAVITY && form.check_mode !== CHECK_MODE_QUOTA) {
+    form.check_mode = CHECK_MODE_QUOTA
+  }
   if (provider === PROVIDER_GROK) {
     if (!form.endpoint.trim()) form.endpoint = DEFAULT_GROK_ENDPOINT
     if (!form.primary_model.trim()) form.primary_model = DEFAULT_GROK_MODEL
     return
   }
-  if (clearGrokEndpoint) form.endpoint = ''
+  if (clearGrokEndpoint || clearPrevDefaultEndpoint) form.endpoint = ''
   if (clearGrokModel) form.primary_model = ''
+  const defaultEndpoint = PROVIDER_DEFAULT_ENDPOINTS[provider]
+  if (defaultEndpoint && !form.endpoint.trim()) form.endpoint = defaultEndpoint
 }
 
 // Clear api_key whenever provider changes to avoid cross-provider key mismatch.
@@ -447,6 +626,8 @@ function resetForm() {
   form.name = ''
   form.provider = PROVIDER_ANTHROPIC
   form.api_mode = API_MODE_CHAT_COMPLETIONS
+  form.check_mode = CHECK_MODE_PROBE
+  form.account_id = null
   form.endpoint = ''
   form.api_key = ''
   form.primary_model = ''
@@ -467,6 +648,8 @@ function loadFromMonitor(m: ChannelMonitor) {
   form.name = m.name
   form.provider = m.provider
   form.api_mode = normalizeAPIMode(m.api_mode)
+  form.check_mode = m.check_mode || CHECK_MODE_PROBE
+  form.account_id = m.account_id ?? null
   form.endpoint = m.endpoint
   form.api_key = ''
   form.primary_model = m.primary_model
@@ -533,15 +716,17 @@ function buildPayload(): CreateParams {
     name: form.name.trim(),
     provider: form.provider,
     api_mode: form.provider === PROVIDER_OPENAI ? form.api_mode : API_MODE_CHAT_COMPLETIONS,
-    endpoint: form.endpoint.trim(),
-    api_key: form.api_key.trim(),
-    primary_model: form.primary_model.trim(),
-    extra_models: form.extra_models,
+    check_mode: form.check_mode,
+    account_id: usesQuotaMode.value ? form.account_id : null,
+    endpoint: usesProbePart.value ? form.endpoint.trim() : '',
+    api_key: usesProbePart.value ? form.api_key.trim() : '',
+    primary_model: usesProbePart.value ? form.primary_model.trim() : 'quota',
+    extra_models: usesProbePart.value ? form.extra_models : [],
     group_name: form.group_name.trim(),
     enabled: form.enabled,
     interval_seconds: form.interval_seconds,
     jitter_seconds: form.jitter_seconds || 0,
-    template_id: form.template_id,
+    template_id: usesProbePart.value ? form.template_id : null,
     extra_headers: form.extra_headers,
     body_override_mode: form.body_override_mode,
     body_override: form.body_override,
@@ -554,7 +739,11 @@ async function handleSubmit() {
     appStore.showError(t('admin.channelMonitor.nameRequired'))
     return
   }
-  if (!form.primary_model.trim()) {
+  if (usesQuotaMode.value && form.account_id == null) {
+    appStore.showError(t('admin.channelMonitor.linkedAccountRequired'))
+    return
+  }
+  if (usesProbePart.value && !form.primary_model.trim()) {
     appStore.showError(t('admin.channelMonitor.primaryModelRequired'))
     return
   }
@@ -568,7 +757,7 @@ async function handleSubmit() {
       // Only send api_key if user typed a new value
       if (api_key) req.api_key = api_key
       // template_id=null 用 clear_template=true 明确告诉后端清空（pointer 语义）
-      if (form.template_id == null) {
+      if (usesProbePart.value && form.template_id == null) {
         req.clear_template = true
         delete req.template_id
       }
diff --git a/frontend/src/components/admin/monitor/MonitorPrimaryModelCell.vue b/frontend/src/components/admin/monitor/MonitorPrimaryModelCell.vue
index eccec8284873..67797904c7c7 100644
--- a/frontend/src/components/admin/monitor/MonitorPrimaryModelCell.vue
+++ b/frontend/src/components/admin/monitor/MonitorPrimaryModelCell.vue
@@ -1,7 +1,8 @@
 <template>
-  <div class="flex items-center gap-2">
-    <span class="text-sm text-gray-900 dark:text-gray-100">{{ row.primary_model }}</span>
-    <HelpTooltip>
+  <div class="flex flex-col gap-0.5">
+    <div class="flex items-center gap-2">
+      <span class="text-sm text-gray-900 dark:text-gray-100">{{ row.primary_model }}</span>
+      <HelpTooltip>
       <template #trigger>
         <span
           class="inline-flex items-center rounded-full px-2 py-0.5 text-[11px] font-medium"
@@ -52,7 +53,10 @@
           </table>
         </div>
       </div>
-    </HelpTooltip>
+      </HelpTooltip>
+    </div>
+    <!-- 配额模式监控：主模型行内联展示最新用量/余额快照（管理端不受用户端开关限制） -->
+    <MonitorQuotaView :snapshot="row.latest_quota" />
   </div>
 </template>
 
@@ -60,6 +64,7 @@
 import { useI18n } from 'vue-i18n'
 import type { ChannelMonitor } from '@/api/admin/channelMonitor'
 import HelpTooltip from '@/components/common/HelpTooltip.vue'
+import MonitorQuotaView from '@/components/common/MonitorQuotaView.vue'
 import { useChannelMonitorFormat } from '@/composables/useChannelMonitorFormat'
 
 defineProps<{
diff --git a/frontend/src/components/admin/monitor/MonitorRunResultDialog.vue b/frontend/src/components/admin/monitor/MonitorRunResultDialog.vue
index 02fa6e8d0526..e8f90f8b610b 100644
--- a/frontend/src/components/admin/monitor/MonitorRunResultDialog.vue
+++ b/frontend/src/components/admin/monitor/MonitorRunResultDialog.vue
@@ -14,6 +14,7 @@
         <div class="flex flex-col">
           <span class="font-medium text-gray-900 dark:text-white">{{ r.model }}</span>
           <span v-if="r.message" class="text-xs text-gray-500 dark:text-gray-400">{{ r.message }}</span>
+          <MonitorQuotaView :snapshot="r.quota" class="mt-1" />
         </div>
         <div class="flex items-center gap-2">
           <span
@@ -40,6 +41,7 @@
 import { useI18n } from 'vue-i18n'
 import type { CheckResult } from '@/api/admin/channelMonitor'
 import BaseDialog from '@/components/common/BaseDialog.vue'
+import MonitorQuotaView from '@/components/common/MonitorQuotaView.vue'
 import { useChannelMonitorFormat } from '@/composables/useChannelMonitorFormat'
 
 defineProps<{
diff --git a/frontend/src/components/admin/monitor/MonitorTemplateManagerDialog.vue b/frontend/src/components/admin/monitor/MonitorTemplateManagerDialog.vue
index 63b87db75aca..e33b1102fdd0 100644
--- a/frontend/src/components/admin/monitor/MonitorTemplateManagerDialog.vue
+++ b/frontend/src/components/admin/monitor/MonitorTemplateManagerDialog.vue
@@ -249,6 +249,11 @@ import {
   PROVIDER_OPENAI,
   PROVIDER_GEMINI,
   PROVIDER_GROK,
+  PROVIDER_ANTIGRAVITY,
+  PROVIDER_KIMI,
+  PROVIDER_ZHIPU,
+  PROVIDER_DEEPSEEK,
+  PROVIDERS,
   API_MODE_CHAT_COMPLETIONS,
   API_MODE_RESPONSES,
 } from '@/constants/channelMonitor'
@@ -269,6 +274,10 @@ const providerTabs = computed<{ value: Provider; label: string }[]>(() => [
   { value: PROVIDER_OPENAI, label: t('monitorCommon.providers.openai') },
   { value: PROVIDER_GEMINI, label: t('monitorCommon.providers.gemini') },
   { value: PROVIDER_GROK, label: t('monitorCommon.providers.grok') },
+  { value: PROVIDER_ANTIGRAVITY, label: t('monitorCommon.providers.antigravity') },
+  { value: PROVIDER_KIMI, label: t('monitorCommon.providers.kimi') },
+  { value: PROVIDER_ZHIPU, label: t('monitorCommon.providers.zhipu') },
+  { value: PROVIDER_DEEPSEEK, label: t('monitorCommon.providers.deepseek') },
 ])
 
 const activeProvider = ref<Provider>(PROVIDER_ANTHROPIC)
@@ -280,12 +289,7 @@ const templatesForActiveProvider = computed(() =>
 )
 
 const countByProvider = computed<Record<Provider, number>>(() => {
-  const out: Record<Provider, number> = {
-    anthropic: 0,
-    openai: 0,
-    gemini: 0,
-    grok: 0,
-  }
+  const out = Object.fromEntries(PROVIDERS.map((p) => [p, 0])) as Record<Provider, number>
   for (const t of templates.value) out[t.provider]++
   return out
 })
diff --git a/frontend/src/components/common/MonitorQuotaView.vue b/frontend/src/components/common/MonitorQuotaView.vue
new file mode 100644
index 000000000000..060e890636e5
--- /dev/null
+++ b/frontend/src/components/common/MonitorQuotaView.vue
@@ -0,0 +1,156 @@
+<template>
+  <div v-if="snapshot" class="space-y-1" data-testid="monitor-quota-view">
+    <!-- 套餐等级徽章（如智谱 plan level / Claude 订阅档） -->
+    <div v-if="snapshot.plan_level" class="flex flex-wrap items-center gap-1.5">
+      <span class="rounded bg-gray-100 px-1.5 py-0.5 text-[10px] font-medium text-gray-600 dark:bg-dark-600 dark:text-gray-300">
+        {{ snapshot.plan_level }}
+      </span>
+    </div>
+
+    <!-- 用量窗口条形图（样式/阈值对齐账号页 CNProviderQuotaCell） -->
+    <div v-if="snapshot.success && tierRows.length" class="space-y-1">
+      <div v-for="row in tierRows" :key="row.key" class="flex items-center gap-1.5 text-[10px]">
+        <span class="w-14 shrink-0 truncate text-gray-500 dark:text-gray-400" :title="row.title">
+          {{ row.label }}
+        </span>
+        <div class="h-1.5 w-16 shrink-0 overflow-hidden rounded-full bg-gray-200 dark:bg-dark-600">
+          <div
+            class="h-full rounded-full transition-all"
+            :class="utilizationColor(row.tier.used_percent)"
+            :style="{ width: `${Math.min(100, Math.max(0, row.tier.used_percent))}%` }"
+          />
+        </div>
+        <span :class="['shrink-0 font-medium', utilizationTextColor(row.tier.used_percent)]">
+          {{ Math.round(row.tier.used_percent) }}%
+        </span>
+        <span v-if="row.tier.reset_at" class="truncate text-gray-400 dark:text-gray-500" :title="row.tier.reset_at">
+          · {{ formatReset(row.tier.reset_at) }}
+        </span>
+      </div>
+    </div>
+
+    <!-- 余额（国产 payg；支持多币种） -->
+    <div v-if="snapshot.success && balanceRows.length" class="flex flex-wrap items-center gap-x-2 gap-y-0.5 text-[10px]">
+      <span
+        v-for="b in balanceRows"
+        :key="b.currency"
+        :class="['font-medium', b.balance <= 0 ? 'text-red-600 dark:text-red-400' : 'text-gray-600 dark:text-gray-300']"
+      >
+        {{ b.balance.toFixed(2) }} {{ b.currency }}
+      </span>
+    </div>
+
+    <div v-if="!snapshot.success" class="truncate text-[10px] text-red-600 dark:text-red-400" :title="snapshot.error" data-testid="monitor-quota-error">
+      {{ truncatedError }}
+    </div>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { computed } from 'vue'
+import { useI18n } from 'vue-i18n'
+import type { MonitorQuotaSnapshot, MonitorQuotaTier } from '@/api/admin/channelMonitor'
+
+/**
+ * 配额快照渲染（管理端监控列表/运行结果 + 用户端监控卡片共用）。
+ * 展示形态对齐账号管理侧的用量视图（CNProviderQuotaCell：同阈值配色、
+ * 同倒计时格式）；tier 的 Window/Label 是后端约定的机器 token，
+ * 已知 token 走 i18n，未知 token 原样展示（前向兼容）。
+ */
+const props = defineProps<{
+  snapshot?: MonitorQuotaSnapshot | null
+}>()
+
+const { t, te } = useI18n()
+
+interface QuotaTierRow {
+  key: string
+  label: string
+  title: string
+  tier: MonitorQuotaTier
+}
+
+// 已知的 window/label 机器 token → i18n key（monitorCommon.quota.*）。
+const windowI18nKeys: Record<string, string> = {
+  '5h': 'monitorCommon.quota.windows.5h',
+  '7d': 'monitorCommon.quota.windows.7d',
+  '7d-sonnet': 'monitorCommon.quota.windows.7dSonnet',
+  '7d-fable': 'monitorCommon.quota.windows.7dFable',
+  weekly: 'monitorCommon.quota.windows.weekly',
+  daily: 'monitorCommon.quota.windows.daily',
+  '30d': 'monitorCommon.quota.windows.30d',
+  total: 'monitorCommon.quota.windows.total',
+}
+
+const labelI18nKeys: Record<string, string> = {
+  requests: 'monitorCommon.quota.labels.requests',
+  tokens: 'monitorCommon.quota.labels.tokens',
+  shared: 'monitorCommon.quota.labels.shared',
+  pro: 'monitorCommon.quota.labels.pro',
+  flash: 'monitorCommon.quota.labels.flash',
+}
+
+function windowLabel(window: string): string {
+  const key = windowI18nKeys[window]
+  return key && te(key) ? t(key) : window
+}
+
+function tierLabel(tier: MonitorQuotaTier): string {
+  const window = windowLabel(tier.window)
+  if (!tier.label) return window
+  const labelKey = labelI18nKeys[tier.label]
+  const label = labelKey && te(labelKey) ? t(labelKey) : tier.label
+  return `${label}/${window}`
+}
+
+const tierRows = computed<QuotaTierRow[]>(() =>
+  (props.snapshot?.tiers || []).map((tier, idx) => ({
+    key: `${tier.window}-${tier.label || ''}-${idx}`,
+    label: tierLabel(tier),
+    title: tierLabel(tier),
+    tier,
+  })),
+)
+
+const balanceRows = computed(() => {
+  const snapshot = props.snapshot
+  if (!snapshot) return []
+  if (snapshot.balances?.length) return snapshot.balances
+  if (snapshot.balance != null) {
+    return [{ currency: snapshot.currency || '?', balance: snapshot.balance }]
+  }
+  return []
+})
+
+const truncatedError = computed(() => {
+  const error = props.snapshot?.error || t('monitorCommon.quota.unavailable')
+  return error.length > 48 ? `${error.slice(0, 48)}…` : error
+})
+
+const utilizationColor = (pct: number) => {
+  if (pct >= 90) return 'bg-red-500'
+  if (pct >= 75) return 'bg-amber-500'
+  return 'bg-emerald-500'
+}
+
+const utilizationTextColor = (pct: number) => {
+  if (pct >= 90) return 'text-red-600 dark:text-red-400'
+  if (pct >= 75) return 'text-amber-600 dark:text-amber-400'
+  return 'text-emerald-600 dark:text-emerald-400'
+}
+
+// 重置时间相对/绝对简短显示（与账号页一致）。
+const formatReset = (iso: string) => {
+  const d = new Date(iso)
+  if (isNaN(d.getTime())) return iso
+  const now = Date.now()
+  const diffMs = d.getTime() - now
+  if (diffMs <= 0) return t('monitorCommon.quota.resetSoon')
+  if (diffMs < 3_600_000) return `${Math.max(1, Math.round(diffMs / 60_000))}m`
+  const hours = Math.round(diffMs / 3_600_000)
+  if (hours < 48) return `${hours}h`
+  const mm = String(d.getMonth() + 1).padStart(2, '0')
+  const dd = String(d.getDate()).padStart(2, '0')
+  return `${mm}-${dd}`
+}
+</script>
diff --git a/frontend/src/components/user/monitor/MonitorCard.vue b/frontend/src/components/user/monitor/MonitorCard.vue
index 1b66b1caedcd..c8835a15dddf 100644
--- a/frontend/src/components/user/monitor/MonitorCard.vue
+++ b/frontend/src/components/user/monitor/MonitorCard.vue
@@ -54,6 +54,9 @@
       secondary-unit="ms"
     />
 
+    <!-- 配额模式：最新用量/余额快照（服务端已按系统开关剥离，此处 flag 为纵深防御） -->
+    <MonitorQuotaView v-if="quotaVisible" :snapshot="item.latest_quota" class="mt-2" />
+
     <!-- Divider -->
     <div class="mt-4 border-t border-gray-100 dark:border-dark-700/60"></div>
 
@@ -80,16 +83,23 @@ import {
   useChannelMonitorFormat,
   providerGradient,
 } from '@/composables/useChannelMonitorFormat'
+import { isChannelMonitorQuotaVisible } from '@/utils/featureFlags'
 import ProviderIcon from './ProviderIcon.vue'
 import MonitorMetricPair from './MonitorMetricPair.vue'
 import MonitorAvailabilityRow from './MonitorAvailabilityRow.vue'
 import MonitorTimeline from './MonitorTimeline.vue'
+import MonitorQuotaView from '@/components/common/MonitorQuotaView.vue'
 
+// 图标配色与 utils/platformColors.ts 的平台色对齐（新 4 家）。
 const PROVIDER_TINT: Record<string, string> = {
   openai: 'text-emerald-600 dark:text-emerald-300',
   anthropic: 'text-orange-600 dark:text-orange-300',
   gemini: 'text-sky-600 dark:text-sky-300',
   grok: 'text-zinc-700 dark:text-zinc-200',
+  antigravity: 'text-purple-600 dark:text-purple-300',
+  kimi: 'text-pink-600 dark:text-pink-300',
+  zhipu: 'text-indigo-600 dark:text-indigo-300',
+  deepseek: 'text-teal-600 dark:text-teal-300',
 }
 
 const props = defineProps<{
@@ -116,6 +126,10 @@ const providerTintClass = computed(() =>
   PROVIDER_TINT[props.item.provider] ?? 'text-gray-500 dark:text-gray-300'
 )
 
+const quotaVisible = computed(
+  () => isChannelMonitorQuotaVisible() && !!props.item.latest_quota
+)
+
 const availabilityLabel = computed(() => {
   const win = t(`channelStatus.windowTab.${props.window}`)
   return `${t('monitorCommon.availabilityPrefix')} · ${win}`
diff --git a/frontend/src/components/user/monitor/ProviderIcon.vue b/frontend/src/components/user/monitor/ProviderIcon.vue
index 17e4e76bc3e0..f0a08f555f06 100644
--- a/frontend/src/components/user/monitor/ProviderIcon.vue
+++ b/frontend/src/components/user/monitor/ProviderIcon.vue
@@ -56,6 +56,28 @@ const PROVIDER_ICONS: Record<Provider, IconData> = {
       'M9.27 15.29l7.978-5.897c.391-.29.95-.177 1.137.272.98 2.369.542 5.215-1.41 7.169-1.951 1.954-4.667 2.382-7.149 1.406l-2.711 1.257c3.889 2.661 8.611 2.003 11.562-.953 2.341-2.344 3.066-5.539 2.388-8.42l.006.007c-.983-4.232.242-5.924 2.75-9.383.06-.082.12-.164.179-.248l-3.301 3.305v-.01L9.267 15.292M7.623 16.723c-2.792-2.67-2.31-6.801.071-9.184 1.761-1.763 4.647-2.483 7.166-1.425l2.705-1.25a7.808 7.808 0 00-1.829-1A8.975 8.975 0 005.984 5.83c-2.533 2.536-3.33 6.436-1.962 9.764 1.022 2.487-.653 4.246-2.34 6.022-.599.63-1.199 1.259-1.682 1.925l7.62-6.815',
     ],
   },
+  // antigravity / kimi / zhipu / deepseek 的官方 logo mark 搬自
+  // src/components/common/PlatformIcon.vue，保持同步。
+  antigravity: {
+    paths: [
+      'M19.35 10.04C18.67 6.59 15.64 4 12 4 9.11 4 6.6 5.64 5.35 8.04 2.34 8.36 0 10.91 0 14c0 3.31 2.69 6 6 6h13c2.76 0 5-2.24 5-5 0-2.64-2.05-4.78-4.65-4.96z',
+    ],
+  },
+  kimi: {
+    paths: [
+      'M21.765.351C22.998.351 24 1.353 24 2.586S22.998 4.82 21.765 4.82h-1.974c-.15 0-.26-.12-.26-.26V2.586A2.237 2.237 0 0 1 21.765.35M9.41 13.388l8.447-8.377c.16-.16.07-.471-.14-.471h-4.55s-.1.02-.14.06l-9.099 9.029c-.14.14-.35.02-.35-.21V4.81c0-.15-.1-.27-.221-.27H.22c-.12 0-.22.12-.22.27v18.57c0 .15.1.27.22.27h3.137c.12 0 .22-.12.22-.27v-3.79c0-.08.03-.16.08-.21l2.826-2.796c.07-.07.16-.08.241-.03l7.546 5.551a8.9 8.9 0 0 0 4.018 1.493c.12.01.23-.11.23-.27V19.76c0-.14-.08-.25-.19-.26a5.8 5.8 0 0 1-2.355-.942l-6.533-4.73c-.14-.09-.15-.32-.03-.441',
+    ],
+  },
+  zhipu: {
+    paths: [
+      'M11.991 23.503a.24.24 0 0 0-.244.248a.24.24 0 0 0 .244.249a.24.24 0 0 0 .245-.249a.24.24 0 0 0-.22-.247zM9.671 5.365a1.697 1.697 0 0 1 1.099 2.132l-.071.172l-.016.04l-.018.054c-.07.16-.104.32-.104.498c-.035.71.47 1.279 1.186 1.314h.366c1.309.053 2.338 1.173 2.286 2.523c-.052 1.332-1.152 2.38-2.478 2.327h-.174c-.715.018-1.274.64-1.239 1.368c0 .124.018.23.053.337c.209.373.54.658.96.8c.75.23 1.517-.125 1.9-.782l.018-.035c.402-.64 1.17-.96 1.92-.711c.854.284 1.378 1.226 1.099 2.167a1.66 1.66 0 0 1-2.077 1.102a1.7 1.7 0 0 1-.907-.711l-.017-.035c-.2-.323-.463-.58-.851-.711l-.056-.018a1.646 1.646 0 0 0-1.954.746a1.66 1.66 0 0 1-1.065.764a1.677 1.677 0 0 1-1.989-1.279c-.209-.906.332-1.83 1.257-2.043a1.5 1.5 0 0 1 .296-.035h.018c.68-.071 1.151-.622 1.116-1.333a1.3 1.3 0 0 0-.227-.693a2.5 2.5 0 0 1-.366-1.403a2.4 2.4 0 0 1 .366-1.208c.14-.195.21-.444.227-.693c.018-.71-.506-1.261-1.186-1.332l-.07-.018a1.4 1.4 0 0 1-.299-.07l-.05-.019a1.7 1.7 0 0 1-1.047-2.114a1.68 1.68 0 0 1 2.094-1.101m-5.575 10.11c.26-.264.639-.367.994-.27s.633.379.728.74c.095.362-.007.748-.267 1.013c-.402.41-1.053.41-1.455 0a1.06 1.06 0 0 1 0-1.482zm14.845-.294c.359-.09.738.024.992.297c.254.274.344.665.237 1.025s-.396.634-.756.718c-.551.128-1.1-.22-1.23-.781a1.05 1.05 0 0 1 .757-1.26zm-.064-4.39c.314.32.49.753.49 1.206s-.176.886-.49 1.206c-.315.32-.74.5-1.185.5c-.444 0-.87-.18-1.184-.5a1.727 1.727 0 0 1 0-2.412a1.654 1.654 0 0 1 2.369 0m-11.243.163c.364.484.447 1.128.218 1.691a1.665 1.665 0 0 1-2.188.923c-.855-.36-1.26-1.358-.907-2.228a1.68 1.68 0 0 1 1.33-1.038a1.66 1.66 0 0 1 1.547.652m11.545-4.221c.368 0 .708.2.892.524s.184.724 0 1.048a1.03 1.03 0 0 1-.892.524a1.04 1.04 0 0 1-1.03-1.048a1.04 1.04 0 0 1 1.03-1.048m-14.358 0c.368 0 .707.2.891.524s.184.724 0 1.048a1.03 1.03 0 0 1-.891.524a1.04 1.04 0 0 1-1.03-1.048c0-.579.461-1.048 1.03-1.048m10.031-1.475c.925 0 1.675.764 1.675 1.706s-.75 1.705-1.675 1.705s-1.674-.763-1.674-1.705s.75-1.706 1.674-1.706m-2.626-.684c.362-.082.653-.356.761-.718a1.06 1.06 0 0 0-.238-1.028a1.02 1.02 0 0 0-.996-.294c-.547.14-.881.7-.752 1.257c.13.558.675.907 1.225.783m0 16.876c.359-.087.644-.36.75-.72a1.06 1.06 0 0 0-.237-1.019a1.02 1.02 0 0 0-.985-.301a1.04 1.04 0 0 0-.762.717c-.108.361-.017.754.239 1.028c.245.263.606.377.953.305zM17.19 3.5a.63.63 0 0 0 .628-.64a.63.63 0 0 0-.628-.64a.63.63 0 0 0-.628.64c0 .355.28.64.628.64m-10.38 0a.63.63 0 0 0 .628-.64c0-.355-.28-.64-.628-.64a.63.63 0 0 0-.628.64c0 .355.279.64.628.64m-5.182 7.852a.63.63 0 0 0-.628.64c0 .354.28.639.628.639a.63.63 0 0 0 .627-.606l.001-.034a.62.62 0 0 0-.628-.64zm5.182 9.13a.63.63 0 0 0-.628.64c0 .355.279.64.628.64a.63.63 0 0 0 .628-.64c0-.355-.28-.64-.628-.64m10.38.018a.63.63 0 0 0-.628.64c0 .355.28.64.628.64a.63.63 0 0 0 .628-.64a.63.63 0 0 0-.628-.64m5.182-9.148a.63.63 0 0 0-.628.64c0 .354.279.639.628.639a.63.63 0 0 0 .628-.64c0-.355-.28-.64-.628-.64zm-.384-4.992a.24.24 0 0 0 .244-.249a.24.24 0 0 0-.244-.249a.24.24 0 0 0-.244.249c0 .142.122.249.244.249M11.991.497a.24.24 0 0 0 .245-.248A.24.24 0 0 0 11.99 0a.24.24 0 0 0-.244.249c0 .133.108.236.223.247zM2.011 6.36a.24.24 0 0 0 .245-.249a.24.24 0 0 0-.244-.249a.24.24 0 0 0-.244.249a.24.24 0 0 0 .244.249zm0 11.263a.24.24 0 0 0-.243.248a.24.24 0 0 0 .244.249a.24.24 0 0 0 .244-.249a.25.25 0 0 0-.244-.248zm19.995-.018a.24.24 0 0 0-.245.248a.24.24 0 0 0 .245.25a.24.24 0 0 0 .244-.25a.25.25 0 0 0-.244-.248z',
+    ],
+  },
+  deepseek: {
+    paths: [
+      'M23.748 4.651c-.254-.124-.364.113-.512.233-.051.04-.094.09-.137.137-.372.397-.806.657-1.373.626-.829-.046-1.537.214-2.163.848-.133-.782-.575-1.248-1.247-1.548-.352-.155-.708-.311-.955-.65-.172-.24-.219-.509-.305-.774-.055-.16-.11-.323-.293-.35-.2-.031-.278.136-.356.276-.313.572-.434 1.202-.422 1.84.027 1.436.633 2.58 1.838 3.393.137.094.172.187.129.323-.082.28-.18.553-.266.833-.055.179-.137.218-.328.14a5.5 5.5 0 0 1-1.737-1.179c-.857-.828-1.631-1.743-2.597-2.46a12 12 0 0 0-.689-.47c-.985-.957.13-1.743.387-1.836.27-.098.094-.433-.778-.428-.872.003-1.67.295-2.687.685a3 3 0 0 1-.465.136a9.6 9.6 0 0 0-2.883-.101c-1.885.21-3.39 1.1-4.497 2.622C.082 8.776-.231 10.854.152 13.02c.403 2.284 1.568 4.175 3.36 5.653 1.857 1.533 3.997 2.284 6.438 2.14 1.482-.085 3.132-.284 4.994-1.86.47.234.962.328 1.78.398.629.058 1.235-.031 1.705-.129.735-.155.684-.836.418-.961-2.155-1.004-1.682-.595-2.112-.926 1.095-1.295 2.768-3.598 3.284-6.733.05-.346.115-.834.108-1.114-.004-.171.035-.238.23-.257a4.2 4.2 0 0 0 1.545-.475c1.397-.763 1.96-2.016 2.093-3.517.02-.23-.004-.467-.247-.588M11.58 18.168c-2.088-1.642-3.101-2.183-3.52-2.16-.39.024-.32.472-.234.763.09.288.207.487.371.74.114.167.192.416-.113.603-.673.416-1.842-.14-1.897-.168-1.361-.801-2.5-1.86-3.301-3.306-.775-1.393-1.225-2.888-1.299-4.482-.02-.385.094-.522.477-.592a4.7 4.7 0 0 1 1.53-.038c2.131.311 3.946 1.264 5.467 2.774.868.86 1.525 1.887 2.202 2.89.72 1.066 1.494 2.082 2.48 2.915.348.291.626.513.892.677-.802.09-2.14.109-3.055-.615zm1.001-6.44a.306.306 0 0 1 .415-.287a.3.3 0 0 1 .113.074a.3.3 0 0 1 .086.214c0 .17-.136.307-.308.307a.303.303 0 0 1-.306-.307m3.11 1.596c-.2.081-.4.151-.591.16a1.25 1.25 0 0 1-.798-.254c-.274-.23-.47-.358-.551-.758a1.7 1.7 0 0 1 .015-.588c.07-.327-.007-.537-.238-.727-.188-.156-.426-.199-.689-.199a.6.6 0 0 1-.254-.078a.253.253 0 0 1-.114-.358a1 1 0 0 1 .192-.21c.356-.202.767-.136 1.146.016.352.144.618.408 1.001.782.392.451.462.576.685.915.176.264.336.536.446.848.066.194-.02.353-.25.45',
+    ],
+  },
 }
 
 const props = withDefaults(defineProps<{
diff --git a/frontend/src/composables/useChannelMonitorFormat.ts b/frontend/src/composables/useChannelMonitorFormat.ts
index bd094a9332e8..04fc0546f55a 100644
--- a/frontend/src/composables/useChannelMonitorFormat.ts
+++ b/frontend/src/composables/useChannelMonitorFormat.ts
@@ -11,12 +11,17 @@
  */
 
 import { useI18n } from 'vue-i18n'
-import type { MonitorStatus, Provider } from '@/api/admin/channelMonitor'
+import type { CheckMode, MonitorStatus, Provider } from '@/api/admin/channelMonitor'
 import {
   PROVIDER_OPENAI,
   PROVIDER_ANTHROPIC,
   PROVIDER_GEMINI,
   PROVIDER_GROK,
+  PROVIDER_ANTIGRAVITY,
+  PROVIDER_KIMI,
+  PROVIDER_ZHIPU,
+  PROVIDER_DEEPSEEK,
+  PROVIDERS,
   STATUS_OPERATIONAL,
   STATUS_DEGRADED,
   STATUS_FAILED,
@@ -58,17 +63,19 @@ export function useChannelMonitorFormat() {
   }
 
   function providerLabel(p: Provider | string): string {
-    if (
-      p === PROVIDER_OPENAI ||
-      p === PROVIDER_ANTHROPIC ||
-      p === PROVIDER_GEMINI ||
-      p === PROVIDER_GROK
-    ) {
+    if (PROVIDERS.includes(p as Provider)) {
       return t(`monitorCommon.providers.${p}`)
     }
     return p || '-'
   }
 
+  function checkModeLabel(m: CheckMode | string): string {
+    if (m === 'probe' || m === 'quota' || m === 'quota_probe') {
+      return t(`monitorCommon.checkMode.${m}`)
+    }
+    return m || '-'
+  }
+
   function providerBadgeClass(p: Provider | string): string {
     switch (p) {
       case PROVIDER_OPENAI:
@@ -79,6 +86,16 @@ export function useChannelMonitorFormat() {
         return 'bg-sky-100 text-sky-700 dark:bg-sky-500/15 dark:text-sky-300'
       case PROVIDER_GROK:
         return 'bg-zinc-100 text-zinc-700 dark:bg-zinc-500/15 dark:text-zinc-300'
+      // 配色与 utils/platformColors.ts 的平台色对齐：antigravity=purple /
+      // kimi=pink / zhipu=indigo / deepseek=teal。
+      case PROVIDER_ANTIGRAVITY:
+        return 'bg-purple-100 text-purple-700 dark:bg-purple-500/15 dark:text-purple-300'
+      case PROVIDER_KIMI:
+        return 'bg-pink-100 text-pink-700 dark:bg-pink-500/15 dark:text-pink-300'
+      case PROVIDER_ZHIPU:
+        return 'bg-indigo-100 text-indigo-700 dark:bg-indigo-500/15 dark:text-indigo-300'
+      case PROVIDER_DEEPSEEK:
+        return 'bg-teal-100 text-teal-700 dark:bg-teal-500/15 dark:text-teal-300'
       default:
         return NEUTRAL_BADGE
     }
@@ -107,6 +124,22 @@ export function useChannelMonitorFormat() {
         return active
           ? 'border-zinc-500 bg-zinc-50 text-zinc-800 dark:bg-zinc-500/15 dark:text-zinc-200 dark:border-zinc-400'
           : 'border-gray-200 bg-white text-gray-600 hover:border-zinc-400 hover:text-zinc-800 dark:border-dark-700 dark:bg-dark-800 dark:text-gray-400 dark:hover:border-zinc-500/50'
+      case PROVIDER_ANTIGRAVITY:
+        return active
+          ? 'border-purple-500 bg-purple-50 text-purple-700 dark:bg-purple-500/15 dark:text-purple-300 dark:border-purple-400'
+          : 'border-gray-200 bg-white text-gray-600 hover:border-purple-300 hover:text-purple-700 dark:border-dark-700 dark:bg-dark-800 dark:text-gray-400 dark:hover:border-purple-500/50'
+      case PROVIDER_KIMI:
+        return active
+          ? 'border-pink-500 bg-pink-50 text-pink-700 dark:bg-pink-500/15 dark:text-pink-300 dark:border-pink-400'
+          : 'border-gray-200 bg-white text-gray-600 hover:border-pink-300 hover:text-pink-700 dark:border-dark-700 dark:bg-dark-800 dark:text-gray-400 dark:hover:border-pink-500/50'
+      case PROVIDER_ZHIPU:
+        return active
+          ? 'border-indigo-500 bg-indigo-50 text-indigo-700 dark:bg-indigo-500/15 dark:text-indigo-300 dark:border-indigo-400'
+          : 'border-gray-200 bg-white text-gray-600 hover:border-indigo-300 hover:text-indigo-700 dark:border-dark-700 dark:bg-dark-800 dark:text-gray-400 dark:hover:border-indigo-500/50'
+      case PROVIDER_DEEPSEEK:
+        return active
+          ? 'border-teal-500 bg-teal-50 text-teal-700 dark:bg-teal-500/15 dark:text-teal-300 dark:border-teal-400'
+          : 'border-gray-200 bg-white text-gray-600 hover:border-teal-300 hover:text-teal-700 dark:border-dark-700 dark:bg-dark-800 dark:text-gray-400 dark:hover:border-teal-500/50'
       default:
         return active
           ? 'border-gray-400 bg-gray-50 text-gray-700 dark:border-dark-500 dark:bg-dark-700 dark:text-gray-200'
@@ -147,6 +180,7 @@ export function useChannelMonitorFormat() {
     statusLabel,
     statusBadgeClass,
     providerLabel,
+    checkModeLabel,
     providerBadgeClass,
     providerPickerClass,
     formatLatency,
@@ -180,6 +214,14 @@ export function providerGradient(provider: string): string {
       return 'bg-gradient-to-br from-sky-50 to-indigo-100 dark:from-sky-500/10 dark:to-indigo-500/20'
     case PROVIDER_GROK:
       return 'bg-gradient-to-br from-zinc-50 to-neutral-200 dark:from-zinc-500/10 dark:to-neutral-500/20'
+    case PROVIDER_ANTIGRAVITY:
+      return 'bg-gradient-to-br from-purple-50 to-purple-100 dark:from-purple-500/10 dark:to-purple-500/20'
+    case PROVIDER_KIMI:
+      return 'bg-gradient-to-br from-pink-50 to-pink-100 dark:from-pink-500/10 dark:to-pink-500/20'
+    case PROVIDER_ZHIPU:
+      return 'bg-gradient-to-br from-indigo-50 to-indigo-100 dark:from-indigo-500/10 dark:to-indigo-500/20'
+    case PROVIDER_DEEPSEEK:
+      return 'bg-gradient-to-br from-teal-50 to-teal-100 dark:from-teal-500/10 dark:to-teal-500/20'
     default:
       return 'bg-gradient-to-br from-gray-100 to-gray-200 dark:from-dark-700 dark:to-dark-600'
   }
diff --git a/frontend/src/constants/channelMonitor.ts b/frontend/src/constants/channelMonitor.ts
index 36d41e3e72ef..576177f8449a 100644
--- a/frontend/src/constants/channelMonitor.ts
+++ b/frontend/src/constants/channelMonitor.ts
@@ -7,16 +7,29 @@
  * `useChannelMonitorFormat`.
  */
 
-import type { APIMode, Provider, MonitorStatus } from '@/api/admin/channelMonitor'
+import type { APIMode, CheckMode, Provider, MonitorStatus } from '@/api/admin/channelMonitor'
 
 export const PROVIDER_OPENAI: Provider = 'openai'
 export const PROVIDER_ANTHROPIC: Provider = 'anthropic'
 export const PROVIDER_GEMINI: Provider = 'gemini'
 export const PROVIDER_GROK: Provider = 'grok'
+export const PROVIDER_ANTIGRAVITY: Provider = 'antigravity'
+export const PROVIDER_KIMI: Provider = 'kimi'
+export const PROVIDER_ZHIPU: Provider = 'zhipu'
+export const PROVIDER_DEEPSEEK: Provider = 'deepseek'
 
 export const DEFAULT_GROK_ENDPOINT = 'https://api.x.ai'
 export const DEFAULT_GROK_MODEL = 'grok-4.5'
 
+/** 国产 provider 的官方 endpoint（探活模式预填；配额模式可留空）。 */
+export const DEFAULT_KIMI_ENDPOINT = 'https://api.moonshot.cn'
+export const DEFAULT_ZHIPU_ENDPOINT = 'https://open.bigmodel.cn'
+export const DEFAULT_DEEPSEEK_ENDPOINT = 'https://api.deepseek.com'
+
+export const CHECK_MODE_PROBE: CheckMode = 'probe'
+export const CHECK_MODE_QUOTA: CheckMode = 'quota'
+export const CHECK_MODE_QUOTA_PROBE: CheckMode = 'quota_probe'
+
 export const API_MODE_CHAT_COMPLETIONS: APIMode = 'chat_completions'
 export const API_MODE_RESPONSES: APIMode = 'responses'
 
@@ -25,6 +38,19 @@ export const PROVIDERS: readonly Provider[] = [
   PROVIDER_ANTHROPIC,
   PROVIDER_GEMINI,
   PROVIDER_GROK,
+  PROVIDER_ANTIGRAVITY,
+  PROVIDER_KIMI,
+  PROVIDER_ZHIPU,
+  PROVIDER_DEEPSEEK,
+]
+
+/** 仅支持配额模式（无探活 adapter）的 provider。 */
+export const QUOTA_ONLY_PROVIDERS: readonly Provider[] = [PROVIDER_ANTIGRAVITY]
+
+export const CHECK_MODES: readonly CheckMode[] = [
+  CHECK_MODE_PROBE,
+  CHECK_MODE_QUOTA,
+  CHECK_MODE_QUOTA_PROBE,
 ]
 
 export const API_MODES: readonly APIMode[] = [
diff --git a/frontend/src/i18n/locales/en/admin/channels.ts b/frontend/src/i18n/locales/en/admin/channels.ts
index bed17bc0301a..0dc9e80b02f0 100644
--- a/frontend/src/i18n/locales/en/admin/channels.ts
+++ b/frontend/src/i18n/locales/en/admin/channels.ts
@@ -512,6 +512,7 @@ export default {
       deleteConfirm: 'Are you sure you want to delete monitor "{name}"? This action cannot be undone.',
       nameRequired: 'Please enter a monitor name',
       primaryModelRequired: 'Please enter a primary model',
+      linkedAccountRequired: 'Please select a linked account',
       columns: {
         name: 'Name',
         provider: 'Provider',
@@ -525,6 +526,17 @@ export default {
         name: 'Name',
         namePlaceholder: 'Enter monitor name',
         provider: 'Platform',
+        checkMode: 'Check Mode',
+        checkModeProbe: 'Probe',
+        checkModeProbeHint: 'Sends a lightweight LLM request to measure availability and latency',
+        checkModeQuota: 'Quota',
+        checkModeQuotaHint: 'Only queries the linked account usage windows / balance, no LLM requests',
+        checkModeQuotaProbe: 'Probe + Quota',
+        checkModeQuotaProbeHint: 'Probes the channel and attaches the quota snapshot to the primary model result',
+        linkedAccount: 'Linked Account',
+        linkedAccountPlaceholder: 'Select an account',
+        linkedAccountHint: 'Quota data comes from the selected account (reuses the account-side usage/balance queries)',
+        linkedAccountEmpty: 'No accounts on this platform yet. Add one in Account Management first',
         apiMode: 'OpenAI protocol',
         apiModeChatCompletions: 'OpenAI Compatible',
         apiModeChatCompletionsHint: 'Use /v1/chat/completions with messages; works for most compatible providers.',
diff --git a/frontend/src/i18n/locales/en/admin/settings.ts b/frontend/src/i18n/locales/en/admin/settings.ts
index 9aff76d251c1..424cca1ded1f 100644
--- a/frontend/src/i18n/locales/en/admin/settings.ts
+++ b/frontend/src/i18n/locales/en/admin/settings.ts
@@ -31,6 +31,9 @@ export default {
           hideThroughput: 'Hide throughput rates from users (RPM / TPM)',
           hideThroughputHint:
             'When on, the user Channel Monitor page and user APIs omit RPM and TPM so fleet volume cannot be reverse-estimated from rates × window. Admins still see full metrics. Error rates, latency, and cache rates remain visible.',
+          showQuota: 'Show channel usage/balance to users',
+          showQuotaHint:
+            'When on, quota-mode channel monitors expose the linked account usage windows/balance on the user Channel Status page. Disabled by default; admins always see it.',
         },
         availableChannels: {
           title: 'Available Channels',
diff --git a/frontend/src/i18n/locales/en/dashboard.ts b/frontend/src/i18n/locales/en/dashboard.ts
index 4204a9d679c5..9f49522a84a8 100644
--- a/frontend/src/i18n/locales/en/dashboard.ts
+++ b/frontend/src/i18n/locales/en/dashboard.ts
@@ -430,7 +430,39 @@ export default {
       openai: 'OpenAI',
       anthropic: 'Anthropic',
       gemini: 'Gemini',
-      grok: 'Grok'
+      grok: 'Grok',
+      antigravity: 'Antigravity',
+      kimi: 'Kimi',
+      zhipu: 'Zhipu GLM',
+      deepseek: 'DeepSeek'
+    },
+    // Check modes (how a monitor performs its checks)
+    checkMode: {
+      probe: 'Probe',
+      quota: 'Quota',
+      quota_probe: 'Probe + Quota'
+    },
+    // Quota snapshot rendering (MonitorQuotaView, shared by admin + user views)
+    quota: {
+      unavailable: 'Quota unavailable',
+      resetSoon: 'resetting',
+      windows: {
+        '5h': '5h',
+        '7d': '7d',
+        '7dSonnet': '7d Sonnet',
+        '7dFable': '7d Fable',
+        weekly: 'Weekly',
+        daily: 'Daily',
+        '30d': '30d',
+        total: 'Total'
+      },
+      labels: {
+        requests: 'Requests',
+        tokens: 'Tokens',
+        shared: 'Shared',
+        pro: 'Pro',
+        flash: 'Flash'
+      }
     },
     extraModelsHeader: 'Extra Models',
     extraModelsEmpty: 'No extra models',
diff --git a/frontend/src/i18n/locales/zh/admin/channels.ts b/frontend/src/i18n/locales/zh/admin/channels.ts
index 12cb22d565d3..f18787ac3b9b 100644
--- a/frontend/src/i18n/locales/zh/admin/channels.ts
+++ b/frontend/src/i18n/locales/zh/admin/channels.ts
@@ -512,6 +512,7 @@ export default {
       deleteConfirm: '确定要删除监控「{name}」吗？此操作不可撤销。',
       nameRequired: '请输入监控名称',
       primaryModelRequired: '请输入主模型',
+      linkedAccountRequired: '请选择关联账号',
       columns: {
         name: '名称',
         provider: '供应商',
@@ -525,6 +526,17 @@ export default {
         name: '名称',
         namePlaceholder: '输入监控名称',
         provider: '平台',
+        checkMode: '检查方式',
+        checkModeProbe: '探活',
+        checkModeProbeHint: '向上游发送轻量 LLM 请求，检测可用性与延迟',
+        checkModeQuota: '配额',
+        checkModeQuotaHint: '只查询关联账号的用量滚动窗口/余额，不产生 LLM 请求',
+        checkModeQuotaProbe: '探活 + 配额',
+        checkModeQuotaProbeHint: '探活的同时查询配额，用量快照附加在主模型结果上',
+        linkedAccount: '关联账号',
+        linkedAccountPlaceholder: '选择账号',
+        linkedAccountHint: '配额数据来自所选账号（复用账号管理侧的用量/余额查询）',
+        linkedAccountEmpty: '当前平台暂无账号，请先在账号管理中添加',
         apiMode: 'OpenAI 协议',
         apiModeChatCompletions: 'OpenAI Compatible',
         apiModeChatCompletionsHint: '使用 /v1/chat/completions，发送 messages；适合大多数兼容站。',
diff --git a/frontend/src/i18n/locales/zh/admin/settings.ts b/frontend/src/i18n/locales/zh/admin/settings.ts
index 82a932e5641f..41d7b673d1f6 100644
--- a/frontend/src/i18n/locales/zh/admin/settings.ts
+++ b/frontend/src/i18n/locales/zh/admin/settings.ts
@@ -31,6 +31,9 @@ export default {
           hideThroughput: '对用户隐藏吞吐速率（RPM / TPM）',
           hideThroughputHint:
             '开启后，用户端渠道监控页面与用户 API 不返回 RPM/TPM，避免用「速率 × 时间窗」反推集群规模。管理员仍可见完整指标；错误率、延迟、缓存率照常展示。',
+          showQuota: '向用户展示渠道用量/余额',
+          showQuotaHint:
+            '开启后，配额模式的渠道监控会在用户端渠道状态页展示关联账号的用量滚动窗口/余额。默认关闭；管理员始终可见。',
         },
         availableChannels: {
           title: '可用渠道',
diff --git a/frontend/src/i18n/locales/zh/dashboard.ts b/frontend/src/i18n/locales/zh/dashboard.ts
index 1d9b8f457465..bcd31fa1f5ef 100644
--- a/frontend/src/i18n/locales/zh/dashboard.ts
+++ b/frontend/src/i18n/locales/zh/dashboard.ts
@@ -435,7 +435,39 @@ export default {
       openai: 'OpenAI',
       anthropic: 'Anthropic',
       gemini: 'Gemini',
-      grok: 'Grok'
+      grok: 'Grok',
+      antigravity: 'Antigravity',
+      kimi: 'Kimi',
+      zhipu: '智谱 GLM',
+      deepseek: 'DeepSeek'
+    },
+    // 检查模式（监控条目的工作方式）
+    checkMode: {
+      probe: '探活',
+      quota: '配额',
+      quota_probe: '探活 + 配额'
+    },
+    // 配额快照展示（MonitorQuotaView，管理端与用户端共用）
+    quota: {
+      unavailable: '配额信息不可用',
+      resetSoon: '即将重置',
+      windows: {
+        '5h': '5 小时',
+        '7d': '7 天',
+        '7dSonnet': '7 天 Sonnet',
+        '7dFable': '7 天 Fable',
+        weekly: '周',
+        daily: '日',
+        '30d': '30 天',
+        total: '总量'
+      },
+      labels: {
+        requests: '请求',
+        tokens: 'Token',
+        shared: '共享',
+        pro: 'Pro',
+        flash: 'Flash'
+      }
     },
     extraModelsHeader: '附加模型',
     extraModelsEmpty: '无附加模型',
diff --git a/frontend/src/types/index.ts b/frontend/src/types/index.ts
index fa162814d61b..eaa585f15c90 100644
--- a/frontend/src/types/index.ts
+++ b/frontend/src/types/index.ts
@@ -268,6 +268,8 @@ export interface PublicSettings {
   channel_monitor_default_interval_seconds: number
   /** When true, user monitor hides RPM/TPM so scale cannot be reverse-estimated. */
   channel_monitor_hide_throughput?: boolean
+  /** When true, user monitor shows account quota/balance snapshots (default off). */
+  channel_monitor_show_quota?: boolean
   available_channels_enabled: boolean
   model_plaza_enabled: boolean
   model_plaza_require_auth: boolean
diff --git a/frontend/src/utils/featureFlags.ts b/frontend/src/utils/featureFlags.ts
index 0fc9d8a3ad88..f2e36b4a6060 100644
--- a/frontend/src/utils/featureFlags.ts
+++ b/frontend/src/utils/featureFlags.ts
@@ -186,3 +186,14 @@ export function isChannelMonitorThroughputHidden(): boolean {
   const appStore = useAppStore()
   return Boolean(appStore.cachedPublicSettings?.channel_monitor_hide_throughput)
 }
+
+/**
+ * Show quota/balance snapshots on the user-facing monitor page
+ * (channel_monitor_show_quota, default off). The backend strips
+ * latest_quota server-side when the switch is off; this flag is
+ * defense-in-depth only. Admin views always show quota.
+ */
+export function isChannelMonitorQuotaVisible(): boolean {
+  const appStore = useAppStore()
+  return appStore.cachedPublicSettings?.channel_monitor_show_quota === true
+}
diff --git a/frontend/src/views/admin/SettingsView.vue b/frontend/src/views/admin/SettingsView.vue
index bb4d021b25de..80af0b194041 100644
--- a/frontend/src/views/admin/SettingsView.vue
+++ b/frontend/src/views/admin/SettingsView.vue
@@ -7089,6 +7089,18 @@
                 </div>
                 <Toggle v-model="form.channel_monitor_hide_throughput" />
               </div>
+
+              <div v-if="form.channel_monitor_mode === 'v1'" class="flex items-start justify-between gap-4">
+                <div class="min-w-0">
+                  <p class="text-sm font-medium text-gray-900 dark:text-white">
+                    {{ t('admin.settings.features.channelMonitor.showQuota') }}
+                  </p>
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    {{ t('admin.settings.features.channelMonitor.showQuotaHint') }}
+                  </p>
+                </div>
+                <Toggle v-model="form.channel_monitor_show_quota" />
+              </div>
             </div>
           </div>
         </div>
@@ -9420,6 +9432,7 @@ type SettingsForm = Omit<
 > & {
   /** Form always binds a concrete boolean (SystemSettings marks this optional). */
   channel_monitor_hide_throughput: boolean;
+  channel_monitor_show_quota: boolean;
   smtp_password: string;
   turnstile_secret_key: string;
   tencent_captcha_app_secret_key: string;
@@ -9732,6 +9745,7 @@ const form = reactive<SettingsForm>({
   channel_monitor_mode: 'v1' as 'v1' | 'v2',
   channel_monitor_default_interval_seconds: 60,
   channel_monitor_hide_throughput: false,
+  channel_monitor_show_quota: false,
   // Available Channels feature switch
   available_channels_enabled: false,
   // Model Plaza feature switches + description
@@ -10733,6 +10747,9 @@ async function loadSettings() {
     form.channel_monitor_hide_throughput = Boolean(
       settings.channel_monitor_hide_throughput
     );
+    form.channel_monitor_show_quota = Boolean(
+      settings.channel_monitor_show_quota
+    );
     form.login_agreement_updated_at =
       settings.login_agreement_updated_at || "2026-03-31";
     form.login_agreement_documents =
@@ -11384,6 +11401,7 @@ async function saveSettings() {
       channel_monitor_default_interval_seconds:
         Number(form.channel_monitor_default_interval_seconds) || 60,
       channel_monitor_hide_throughput: Boolean(form.channel_monitor_hide_throughput),
+      channel_monitor_show_quota: Boolean(form.channel_monitor_show_quota),
       // Available Channels feature switch
       available_channels_enabled: form.available_channels_enabled,
       // Model Plaza feature switches + description

From 302a10b88608392617dbe8e228711a86a236ac0d Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 05:10:10 +0000
Subject: [PATCH 54/68] =?UTF-8?q?test(monitor-ui):=20=E9=85=8D=E9=A2=9D?=
 =?UTF-8?q?=E8=A7=86=E5=9B=BE=E6=B8=B2=E6=9F=93=E4=B8=8E=E5=BC=80=E5=85=B3?=
 =?UTF-8?q?=E9=97=A8=E6=8E=A7=E7=94=A8=E4=BE=8B?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 新增 MonitorQuotaView.spec：tier window/label 的 i18n 映射与未知
  token 透出、阈值配色（≥90 红/≥75 黄/其余绿）、进度条宽度钳制、
  多币种余额与单余额回退、plan_level 徽章、错误截断与失败态不渲染 tiers
- 新增 MonitorCard.quota.spec：latest_quota 受系统开关门控，
  无快照或开关关闭时不渲染
- 既有工厂/断言补 check_mode、account_id；provider 网格断言 4→8
---
 .../admin/monitor/MonitorActionsCell.spec.ts  |   2 +
 .../common/__tests__/MonitorQuotaView.spec.ts | 128 ++++++++++++++++++
 .../user/__tests__/MonitorCard.quota.spec.ts  |  91 +++++++++++++
 .../ChannelMonitorView.duplicate.spec.ts      |   2 +
 .../__tests__/ChannelMonitorView.grok.spec.ts |   2 +-
 5 files changed, 224 insertions(+), 1 deletion(-)
 create mode 100644 frontend/src/components/common/__tests__/MonitorQuotaView.spec.ts
 create mode 100644 frontend/src/components/user/__tests__/MonitorCard.quota.spec.ts

diff --git a/frontend/src/components/admin/monitor/MonitorActionsCell.spec.ts b/frontend/src/components/admin/monitor/MonitorActionsCell.spec.ts
index 31f9eca73825..a7989ffeaa00 100644
--- a/frontend/src/components/admin/monitor/MonitorActionsCell.spec.ts
+++ b/frontend/src/components/admin/monitor/MonitorActionsCell.spec.ts
@@ -34,6 +34,8 @@ function makeMonitor(overrides: Partial<ChannelMonitor> = {}): ChannelMonitor {
     extra_headers: {},
     body_override_mode: 'off',
     body_override: null,
+    check_mode: 'probe',
+    account_id: null,
     ...overrides,
   }
 }
diff --git a/frontend/src/components/common/__tests__/MonitorQuotaView.spec.ts b/frontend/src/components/common/__tests__/MonitorQuotaView.spec.ts
new file mode 100644
index 000000000000..9f305d2156e2
--- /dev/null
+++ b/frontend/src/components/common/__tests__/MonitorQuotaView.spec.ts
@@ -0,0 +1,128 @@
+import { mount } from '@vue/test-utils'
+import { describe, expect, it, vi } from 'vitest'
+
+import type { MonitorQuotaSnapshot } from '@/api/admin/channelMonitor'
+import MonitorQuotaView from '../MonitorQuotaView.vue'
+
+vi.mock('vue-i18n', async () => {
+  const actual = await vi.importActual<typeof import('vue-i18n')>('vue-i18n')
+  return {
+    ...actual,
+    // te() 恒真：已知 token 直接返回 i18n key，便于断言 window/label 映射。
+    useI18n: () => ({ t: (key: string) => key, te: () => true }),
+  }
+})
+
+function makeSnapshot(overrides: Partial<MonitorQuotaSnapshot> = {}): MonitorQuotaSnapshot {
+  return {
+    source: 'usage',
+    success: true,
+    fetched_at: '2026-08-18T00:00:00Z',
+    ...overrides,
+  }
+}
+
+describe('MonitorQuotaView', () => {
+  it('renders nothing without a snapshot', () => {
+    const wrapper = mount(MonitorQuotaView, { props: { snapshot: null } })
+    expect(wrapper.find('[data-testid="monitor-quota-view"]').exists()).toBe(false)
+    expect(wrapper.text()).toBe('')
+  })
+
+  it('maps tier window/label tokens through i18n and colors utilization', () => {
+    const wrapper = mount(MonitorQuotaView, {
+      props: {
+        snapshot: makeSnapshot({
+          tiers: [
+            { window: '5h', used_percent: 42.4 },
+            { window: '7d', label: 'pro', used_percent: 80 },
+            { window: 'weekly', label: 'unknown-token', used_percent: 95 },
+          ],
+        }),
+      },
+    })
+
+    const rows = wrapper.findAll('[data-testid="monitor-quota-view"] .flex.items-center')
+    // tier 行只在 success 且有数据时渲染
+    expect(rows.length).toBeGreaterThanOrEqual(3)
+    const text = wrapper.text()
+    // 已知 window token 走 i18n
+    expect(text).toContain('monitorCommon.quota.windows.5h')
+    expect(text).toContain('monitorCommon.quota.windows.7d')
+    // 已知 label token 拼成 label/window
+    expect(text).toContain('monitorCommon.quota.labels.pro/monitorCommon.quota.windows.7d')
+    // 未知 label 原样透出（前向兼容）
+    expect(text).toContain('unknown-token/monitorCommon.quota.windows.weekly')
+    // 百分比取整
+    expect(text).toContain('42%')
+    expect(text).toContain('80%')
+    expect(text).toContain('95%')
+
+    const html = wrapper.html()
+    // 阈值配色：≥90 红 / ≥75 黄 / 其余绿（与账号页 CNProviderQuotaCell 对齐）
+    expect(html).toContain('bg-emerald-500')
+    expect(html).toContain('bg-amber-500')
+    expect(html).toContain('bg-red-500')
+  })
+
+  it('clamps the tier bar width into 0-100', () => {
+    const wrapper = mount(MonitorQuotaView, {
+      props: {
+        snapshot: makeSnapshot({ tiers: [{ window: '5h', used_percent: 240 }] }),
+      },
+    })
+    expect(wrapper.html()).toContain('width: 100%')
+  })
+
+  it('shows the plan level badge and multi-currency balances', () => {
+    const wrapper = mount(MonitorQuotaView, {
+      props: {
+        snapshot: makeSnapshot({
+          plan_level: 'Max20',
+          balances: [
+            { currency: 'CNY', balance: 12.5 },
+            { currency: 'USD', balance: 0 },
+          ],
+        }),
+      },
+    })
+
+    expect(wrapper.text()).toContain('Max20')
+    expect(wrapper.text()).toContain('12.50 CNY')
+    expect(wrapper.text()).toContain('0.00 USD')
+    // 余额为 0 用红色警示
+    expect(wrapper.html()).toContain('text-red-600')
+  })
+
+  it('falls back to the single balance + currency pair', () => {
+    const wrapper = mount(MonitorQuotaView, {
+      props: { snapshot: makeSnapshot({ balance: 3.2, currency: 'CNY' }) },
+    })
+    expect(wrapper.text()).toContain('3.20 CNY')
+  })
+
+  it('renders a truncated error state when the fetch failed', () => {
+    const longError = 'x'.repeat(60)
+    const wrapper = mount(MonitorQuotaView, {
+      props: {
+        snapshot: makeSnapshot({ success: false, error: longError }),
+      },
+    })
+
+    const error = wrapper.get('[data-testid="monitor-quota-error"]')
+    expect(error.text()).toBe(`${'x'.repeat(48)}…`)
+    expect(error.attributes('title')).toBe(longError)
+  })
+
+  it('keeps failed snapshots from rendering tier rows', () => {
+    const wrapper = mount(MonitorQuotaView, {
+      props: {
+        snapshot: makeSnapshot({
+          success: false,
+          tiers: [{ window: '5h', used_percent: 10 }],
+        }),
+      },
+    })
+    expect(wrapper.text()).not.toContain('10%')
+  })
+})
diff --git a/frontend/src/components/user/__tests__/MonitorCard.quota.spec.ts b/frontend/src/components/user/__tests__/MonitorCard.quota.spec.ts
new file mode 100644
index 000000000000..93a8c7abc3cd
--- /dev/null
+++ b/frontend/src/components/user/__tests__/MonitorCard.quota.spec.ts
@@ -0,0 +1,91 @@
+import { mount } from '@vue/test-utils'
+import { describe, expect, it, vi } from 'vitest'
+
+import type { UserMonitorView } from '@/api/channelMonitor'
+import MonitorCard from '../monitor/MonitorCard.vue'
+
+const { isQuotaVisible } = vi.hoisted(() => ({
+  isQuotaVisible: vi.fn(() => false),
+}))
+
+vi.mock('@/utils/featureFlags', () => ({
+  isChannelMonitorQuotaVisible: () => isQuotaVisible(),
+}))
+
+vi.mock('vue-i18n', async () => {
+  const actual = await vi.importActual<typeof import('vue-i18n')>('vue-i18n')
+  return {
+    ...actual,
+    useI18n: () => ({ t: (key: string) => key, te: () => true }),
+  }
+})
+
+function makeItem(overrides: Partial<UserMonitorView> = {}): UserMonitorView {
+  return {
+    id: 1,
+    name: 'claude-main',
+    provider: 'kimi',
+    group_name: '',
+    primary_model: 'quota',
+    primary_status: 'operational',
+    primary_latency_ms: null,
+    primary_ping_latency_ms: null,
+    availability_7d: 100,
+    extra_models: [],
+    timeline: [],
+    ...overrides,
+  }
+}
+
+function mountCard(item: UserMonitorView) {
+  return mount(MonitorCard, {
+    props: {
+      item,
+      window: '7d',
+      availabilityValue: 100,
+      countdownSeconds: 0,
+    },
+    global: {
+      stubs: {
+        MonitorMetricPair: true,
+        MonitorAvailabilityRow: true,
+        MonitorTimeline: true,
+      },
+    },
+  })
+}
+
+describe('MonitorCard quota snapshot visibility', () => {
+  it('hides the quota block when the system switch is off even if data exists', () => {
+    isQuotaVisible.mockReturnValue(false)
+    const wrapper = mountCard(
+      makeItem({
+        latest_quota: { source: 'cn_quota', success: true, fetched_at: '2026-08-18T00:00:00Z' },
+      }),
+    )
+    expect(wrapper.find('[data-testid="monitor-quota-view"]').exists()).toBe(false)
+  })
+
+  it('renders the quota block when the switch is on and a snapshot exists', () => {
+    isQuotaVisible.mockReturnValue(true)
+    const wrapper = mountCard(
+      makeItem({
+        latest_quota: {
+          source: 'cn_quota',
+          success: true,
+          plan_level: 'kimi-plus',
+          tiers: [{ window: 'daily', label: 'requests', used_percent: 60 }],
+          fetched_at: '2026-08-18T00:00:00Z',
+        },
+      }),
+    )
+    expect(wrapper.find('[data-testid="monitor-quota-view"]').exists()).toBe(true)
+    expect(wrapper.text()).toContain('kimi-plus')
+  })
+
+  it('never renders the quota block without a snapshot', () => {
+    isQuotaVisible.mockReturnValue(true)
+    const wrapper = mountCard(makeItem())
+    expect(wrapper.find('[data-testid="monitor-quota-view"]').exists()).toBe(false)
+  })
+})
diff --git a/frontend/src/views/admin/__tests__/ChannelMonitorView.duplicate.spec.ts b/frontend/src/views/admin/__tests__/ChannelMonitorView.duplicate.spec.ts
index 3eeb1d611fc5..5dd97fd534d2 100644
--- a/frontend/src/views/admin/__tests__/ChannelMonitorView.duplicate.spec.ts
+++ b/frontend/src/views/admin/__tests__/ChannelMonitorView.duplicate.spec.ts
@@ -96,6 +96,8 @@ function makeMonitor(overrides: Partial<ChannelMonitor> = {}): ChannelMonitor {
     extra_headers: {},
     body_override_mode: 'off',
     body_override: null,
+    check_mode: 'probe',
+    account_id: null,
     ...overrides,
   }
 }
diff --git a/frontend/src/views/admin/__tests__/ChannelMonitorView.grok.spec.ts b/frontend/src/views/admin/__tests__/ChannelMonitorView.grok.spec.ts
index f3a47ffa6195..94f619aea9c6 100644
--- a/frontend/src/views/admin/__tests__/ChannelMonitorView.grok.spec.ts
+++ b/frontend/src/views/admin/__tests__/ChannelMonitorView.grok.spec.ts
@@ -93,7 +93,7 @@ describe('channel monitor Grok provider', () => {
 
     expect(PROVIDERS).toContain(PROVIDER_GROK)
     const providerButtons = wrapper.findAll('[data-testid^="monitor-provider-"]')
-    expect(providerButtons).toHaveLength(4)
+    expect(providerButtons).toHaveLength(8)
     expect(providerButtons[0].element.parentElement?.className).toContain('grid-cols-2')
     expect(providerButtons[0].element.parentElement?.className).toContain('sm:grid-cols-4')
 

From 7376f4a48cc0bdcf78232e1ed77063f0e0152c5a Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 05:57:41 +0000
Subject: [PATCH 55/68] =?UTF-8?q?fix(lint):=20gofmt=20=E5=AF=B9=E9=BD=90?=
 =?UTF-8?q?=20+=20=E7=A7=BB=E9=99=A4=E6=9C=AA=E4=BD=BF=E7=94=A8=E7=9A=84?=
 =?UTF-8?q?=20isSupportedProvider?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 三个文件补 gofmt 规范对齐（结构体尾部注释、const 块字段对齐）
- isSupportedProvider 在 validateProvider 改用能力集合后无引用，删除
---
 .../internal/domain/channel_monitor_quota.go   |  8 ++++----
 .../service/channel_monitor_checker.go         |  7 -------
 .../internal/service/channel_monitor_const.go  | 18 +++++++++---------
 .../internal/service/channel_monitor_types.go  |  4 ++--
 4 files changed, 15 insertions(+), 22 deletions(-)

diff --git a/backend/internal/domain/channel_monitor_quota.go b/backend/internal/domain/channel_monitor_quota.go
index fed21dedb797..72c989c38fb3 100644
--- a/backend/internal/domain/channel_monitor_quota.go
+++ b/backend/internal/domain/channel_monitor_quota.go
@@ -46,10 +46,10 @@ type MonitorQuotaSnapshot struct {
 	Source    string             `json:"source"`
 	Success   bool               `json:"success"`
 	Tiers     []MonitorQuotaTier `json:"tiers,omitempty"`
-	Balance   *float64           `json:"balance,omitempty"`           // cn_balance 主余额
-	Balances  []MonitorBalance   `json:"balances,omitempty"`          // 多币种余额（如 DeepSeek CNY+USD）
-	Currency  string             `json:"currency,omitempty"`          // 主余额币种
-	PlanLevel string             `json:"plan_level,omitempty"`        // 套餐等级（如智谱 level）
+	Balance   *float64           `json:"balance,omitempty"`    // cn_balance 主余额
+	Balances  []MonitorBalance   `json:"balances,omitempty"`   // 多币种余额（如 DeepSeek CNY+USD）
+	Currency  string             `json:"currency,omitempty"`   // 主余额币种
+	PlanLevel string             `json:"plan_level,omitempty"` // 套餐等级（如智谱 level）
 	// CredentialInvalid 上游 401/403 鉴权失败（区别于网络/解析错误），
 	// 检测状态据此推导 failed 而非 error。
 	CredentialInvalid bool      `json:"credential_invalid,omitempty"`
diff --git a/backend/internal/service/channel_monitor_checker.go b/backend/internal/service/channel_monitor_checker.go
index bc8aa7064584..71177db0bdb8 100644
--- a/backend/internal/service/channel_monitor_checker.go
+++ b/backend/internal/service/channel_monitor_checker.go
@@ -271,13 +271,6 @@ func providerAdapterFor(provider, apiMode string) (providerAdapter, string, bool
 	return adapter, MonitorAPIModeChatCompletions, ok
 }
 
-// isSupportedProvider 校验 provider 字符串是否在探活 adapter 表中。
-// 注意：完整的 provider 合法性（含仅配额模式的 antigravity）走 validate.go 的 monitorProviders。
-func isSupportedProvider(p string) bool {
-	_, ok := providerAdapters[p]
-	return ok
-}
-
 // callProvider 通过 providerAdapters 分发到具体实现。
 // opts 承载用户的自定义 headers / body 覆盖（可为 nil）。
 //
diff --git a/backend/internal/service/channel_monitor_const.go b/backend/internal/service/channel_monitor_const.go
index 4ed6a026ea6d..6427c8914de2 100644
--- a/backend/internal/service/channel_monitor_const.go
+++ b/backend/internal/service/channel_monitor_const.go
@@ -61,21 +61,21 @@ const (
 	// MonitorProviderOpenAI 等 provider 字符串常量（也是 ent enum 的实际值）。
 	// 后 4 个 provider（antigravity/kimi/zhipu/deepseek）为配额模式引入：
 	// antigravity 无探活 adapter（仅配额），其余 3 个复用 OpenAI 兼容探活。
-	MonitorProviderOpenAI     = "openai"
-	MonitorProviderAnthropic  = "anthropic"
-	MonitorProviderGemini     = "gemini"
-	MonitorProviderGrok       = "grok"
+	MonitorProviderOpenAI      = "openai"
+	MonitorProviderAnthropic   = "anthropic"
+	MonitorProviderGemini      = "gemini"
+	MonitorProviderGrok        = "grok"
 	MonitorProviderAntigravity = "antigravity"
-	MonitorProviderKimi       = "kimi"
-	MonitorProviderZhipu      = "zhipu"
-	MonitorProviderDeepseek   = "deepseek"
+	MonitorProviderKimi        = "kimi"
+	MonitorProviderZhipu       = "zhipu"
+	MonitorProviderDeepseek    = "deepseek"
 
 	// MonitorCheckMode 检测模式（channel_monitors.check_mode）。
 	//   probe       - LLM 探活（默认，原有行为）
 	//   quota       - 仅查关联账号用量/余额，零 LLM 成本
 	//   quota_probe - 探活 + 配额并存（配额快照挂到主模型历史行）
-	MonitorCheckModeProbe     = "probe"
-	MonitorCheckModeQuota     = "quota"
+	MonitorCheckModeProbe      = "probe"
+	MonitorCheckModeQuota      = "quota"
 	MonitorCheckModeQuotaProbe = "quota_probe"
 
 	// MonitorDefaultQuotaModel 是 quota 模式监控未显式指定模型时占位的虚拟模型名
diff --git a/backend/internal/service/channel_monitor_types.go b/backend/internal/service/channel_monitor_types.go
index 298e037ee118..138593e0bd65 100644
--- a/backend/internal/service/channel_monitor_types.go
+++ b/backend/internal/service/channel_monitor_types.go
@@ -51,8 +51,8 @@ type ChannelMonitor struct {
 
 	// 配额模式（check_mode = quota / quota_probe）：
 	// 关联已有账号复用账号侧用量服务，Endpoint/APIKey 可为空（quota 模式）。
-	CheckMode string  // probe（默认）/ quota / quota_probe；空串按 probe 处理
-	AccountID *int64  // 关联账号 ID；账号删除后被 DB 置空（监控保留并报「账号未关联」）
+	CheckMode string // probe（默认）/ quota / quota_probe；空串按 probe 处理
+	AccountID *int64 // 关联账号 ID；账号删除后被 DB 置空（监控保留并报「账号未关联」）
 
 	// 请求自定义快照（来自模板拷贝 or 用户手填，运行时直接读取）
 	TemplateID       *int64            // 仅用于 UI 分组 + 一键应用，运行时不用

From ebcae03afda0ddec1ff744973ea3d3cced12c3dd Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 06:29:43 +0000
Subject: [PATCH 56/68] =?UTF-8?q?fix(lint):=20=E8=A1=A5=E9=BD=90=20setting?=
 =?UTF-8?q?=5Fpublic.go=20=E4=B8=8E=E9=85=8D=E9=A2=9D=20fetcher=20?=
 =?UTF-8?q?=E6=B5=8B=E8=AF=95=E7=9A=84=20gofmt=20=E5=AF=B9=E9=BD=90?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../channel_monitor_quota_fetcher_test.go     | 22 +++++++++----------
 backend/internal/service/setting_public.go    | 14 ++++++------
 2 files changed, 18 insertions(+), 18 deletions(-)

diff --git a/backend/internal/service/channel_monitor_quota_fetcher_test.go b/backend/internal/service/channel_monitor_quota_fetcher_test.go
index 5807058d9053..c9337c766d93 100644
--- a/backend/internal/service/channel_monitor_quota_fetcher_test.go
+++ b/backend/internal/service/channel_monitor_quota_fetcher_test.go
@@ -282,18 +282,18 @@ func TestUsageQuotaTiers_MapsAllWindowKinds(t *testing.T) {
 	remaining := int64(400)
 	resetUnix := int64(1777283883)
 	usage := &UsageInfo{
-		FiveHour:           &UsageProgress{Utilization: 50},
-		SevenDay:           &UsageProgress{Utilization: 60},
-		SevenDaySonnet:     &UsageProgress{Utilization: 70},
-		SevenDayFable:      &UsageProgress{Utilization: 80},
-		ThirtyDay:          &UsageProgress{Utilization: 20},
-		GeminiSharedDaily:  &UsageProgress{Utilization: 11},
-		GeminiProDaily:     &UsageProgress{Utilization: 22},
-		GeminiFlashDaily:   &UsageProgress{Utilization: 33},
-		GrokRequestQuota:   &xai.QuotaWindow{Limit: &limit, Remaining: &remaining, ResetUnix: &resetUnix},
-		GrokTokenQuota:     &xai.QuotaWindow{Limit: &limit, Remaining: &remaining, ResetAt: "2026-08-19T00:00:00Z"},
+		FiveHour:          &UsageProgress{Utilization: 50},
+		SevenDay:          &UsageProgress{Utilization: 60},
+		SevenDaySonnet:    &UsageProgress{Utilization: 70},
+		SevenDayFable:     &UsageProgress{Utilization: 80},
+		ThirtyDay:         &UsageProgress{Utilization: 20},
+		GeminiSharedDaily: &UsageProgress{Utilization: 11},
+		GeminiProDaily:    &UsageProgress{Utilization: 22},
+		GeminiFlashDaily:  &UsageProgress{Utilization: 33},
+		GrokRequestQuota:  &xai.QuotaWindow{Limit: &limit, Remaining: &remaining, ResetUnix: &resetUnix},
+		GrokTokenQuota:    &xai.QuotaWindow{Limit: &limit, Remaining: &remaining, ResetAt: "2026-08-19T00:00:00Z"},
 		AntigravityQuota: map[string]*AntigravityModelQuota{
-			"gemini-3-pro": {Utilization: 45},
+			"gemini-3-pro":   {Utilization: 45},
 			"gemini-3-flash": {Utilization: 55},
 		},
 	}
diff --git a/backend/internal/service/setting_public.go b/backend/internal/service/setting_public.go
index 4679d0048467..81d74aff0de0 100644
--- a/backend/internal/service/setting_public.go
+++ b/backend/internal/service/setting_public.go
@@ -616,13 +616,13 @@ type PublicSettingsInjectionPayload struct {
 	ChannelMonitorHideThroughput bool `json:"channel_monitor_hide_throughput"`
 	// ChannelMonitorShowQuota gates the user-facing quota/balance display on
 	// monitors; fail-closed (absent/false = hidden). Admin UI always shows it.
-	ChannelMonitorShowQuota bool `json:"channel_monitor_show_quota"`
-	AvailableChannelsEnabled     bool `json:"available_channels_enabled"`
-	ModelPlazaEnabled            bool `json:"model_plaza_enabled"`
-	ModelPlazaRequireAuth        bool `json:"model_plaza_require_auth"`
-	AffiliateEnabled             bool `json:"affiliate_enabled"`
-	RiskControlEnabled           bool `json:"risk_control_enabled"`
-	AllowUserViewErrorRequests   bool `json:"allow_user_view_error_requests"`
+	ChannelMonitorShowQuota    bool `json:"channel_monitor_show_quota"`
+	AvailableChannelsEnabled   bool `json:"available_channels_enabled"`
+	ModelPlazaEnabled          bool `json:"model_plaza_enabled"`
+	ModelPlazaRequireAuth      bool `json:"model_plaza_require_auth"`
+	AffiliateEnabled           bool `json:"affiliate_enabled"`
+	RiskControlEnabled         bool `json:"risk_control_enabled"`
+	AllowUserViewErrorRequests bool `json:"allow_user_view_error_requests"`
 }
 
 // GetPublicSettingsForInjection returns public settings in a format suitable for HTML injection.

From 7e579cb28df5f2e7ed9ef84bc8bca6370acfa8f7 Mon Sep 17 00:00:00 2001
From: hansnow <hansnow2012@gmail.com>
Date: Tue, 18 Aug 2026 14:42:30 +0800
Subject: [PATCH 57/68] fix(openai): adapt client tools in WS HTTP bridge

---
 .../openai_gateway_grok_tool_protocol.go      | 26 +++++--
 .../internal/service/openai_ws_http_bridge.go | 13 +++-
 .../service/openai_ws_http_bridge_test.go     | 75 +++++++++++++++++++
 3 files changed, 108 insertions(+), 6 deletions(-)

diff --git a/backend/internal/service/openai_gateway_grok_tool_protocol.go b/backend/internal/service/openai_gateway_grok_tool_protocol.go
index 1a8dfc35a6bd..f1683a7adb46 100644
--- a/backend/internal/service/openai_gateway_grok_tool_protocol.go
+++ b/backend/internal/service/openai_gateway_grok_tool_protocol.go
@@ -15,12 +15,12 @@ import (
 
 const grokResponsesClientToolMappingContextKey = "grok_responses_client_tool_mapping"
 
-func adaptGrokResponsesClientTools(body []byte) ([]byte, apicompat.ResponsesClientToolMapping, error) {
+func adaptResponsesClientToolsForFunctionUpstream(body []byte, upstream string) ([]byte, apicompat.ResponsesClientToolMapping, error) {
 	decoder := json.NewDecoder(bytes.NewReader(body))
 	decoder.UseNumber()
 	var requestBody map[string]any
 	if err := decoder.Decode(&requestBody); err != nil {
-		return body, apicompat.ResponsesClientToolMapping{}, fmt.Errorf("decode Grok Responses client tools: %w", err)
+		return body, apicompat.ResponsesClientToolMapping{}, fmt.Errorf("decode %s Responses client tools: %w", upstream, err)
 	}
 
 	mapping, changed, err := apicompat.AdaptResponsesClientTools(requestBody)
@@ -32,15 +32,23 @@ func adaptGrokResponsesClientTools(body []byte) ([]byte, apicompat.ResponsesClie
 	}
 	rebuilt, err := marshalOpenAIUpstreamJSON(requestBody)
 	if err != nil {
-		return body, apicompat.ResponsesClientToolMapping{}, fmt.Errorf("encode Grok Responses client tools: %w", err)
+		return body, apicompat.ResponsesClientToolMapping{}, fmt.Errorf("encode %s Responses client tools: %w", upstream, err)
 	}
 	return rebuilt, mapping, nil
 }
 
-func hasGrokResponsesClientToolMapping(mapping apicompat.ResponsesClientToolMapping) bool {
+func adaptGrokResponsesClientTools(body []byte) ([]byte, apicompat.ResponsesClientToolMapping, error) {
+	return adaptResponsesClientToolsForFunctionUpstream(body, "Grok")
+}
+
+func hasResponsesClientToolMapping(mapping apicompat.ResponsesClientToolMapping) bool {
 	return len(mapping.CustomTools) > 0 || mapping.ToolSearch || len(mapping.NamespaceTools) > 0
 }
 
+func hasGrokResponsesClientToolMapping(mapping apicompat.ResponsesClientToolMapping) bool {
+	return hasResponsesClientToolMapping(mapping)
+}
+
 func setGrokResponsesClientToolMapping(c *gin.Context, mapping apicompat.ResponsesClientToolMapping) {
 	if c == nil {
 		return
@@ -97,7 +105,7 @@ func (b *grokResponsesClientToolStreamBody) Close() error {
 	return sourceErr
 }
 
-func newGrokResponsesClientToolStreamBody(
+func newResponsesClientToolStreamBody(
 	source io.ReadCloser,
 	mapping apicompat.ResponsesClientToolMapping,
 	maxLineSize int,
@@ -108,6 +116,14 @@ func newGrokResponsesClientToolStreamBody(
 	return body
 }
 
+func newGrokResponsesClientToolStreamBody(
+	source io.ReadCloser,
+	mapping apicompat.ResponsesClientToolMapping,
+	maxLineSize int,
+) io.ReadCloser {
+	return newResponsesClientToolStreamBody(source, mapping, maxLineSize)
+}
+
 func transformGrokResponsesClientToolStream(
 	source io.ReadCloser,
 	destination *io.PipeWriter,
diff --git a/backend/internal/service/openai_ws_http_bridge.go b/backend/internal/service/openai_ws_http_bridge.go
index 87abf1ef04f7..98eee9c7db0c 100644
--- a/backend/internal/service/openai_ws_http_bridge.go
+++ b/backend/internal/service/openai_ws_http_bridge.go
@@ -12,6 +12,7 @@ import (
 	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
 	"github.com/gin-gonic/gin"
 	"github.com/tidwall/gjson"
 )
@@ -186,6 +187,13 @@ func (s *OpenAIGatewayService) proxyOpenAIWSHTTPBridgeTurn(
 	if err != nil {
 		return nil, fmt.Errorf("prepare http bridge body: %w", err)
 	}
+	var clientToolMapping apicompat.ResponsesClientToolMapping
+	if account.Platform == PlatformOpenAI && account.Type == AccountTypeAPIKey {
+		body, clientToolMapping, err = adaptResponsesClientToolsForFunctionUpstream(body, "OpenAI WS HTTP bridge")
+		if err != nil {
+			return nil, fmt.Errorf("adapt OpenAI WS HTTP bridge client tools: %w", err)
+		}
+	}
 
 	upstreamCtx, releaseUpstreamCtx := detachUpstreamContext(ctx)
 	var upstreamReq *http.Request
@@ -329,11 +337,14 @@ func (s *OpenAIGatewayService) proxyOpenAIWSHTTPBridgeTurn(
 		return result
 	}
 
-	scanner := bufio.NewScanner(resp.Body)
 	maxLineSize := defaultMaxLineSize
 	if s.cfg != nil && s.cfg.Gateway.MaxLineSize > 0 {
 		maxLineSize = s.cfg.Gateway.MaxLineSize
 	}
+	if hasResponsesClientToolMapping(clientToolMapping) {
+		resp.Body = newResponsesClientToolStreamBody(resp.Body, clientToolMapping, maxLineSize)
+	}
+	scanner := bufio.NewScanner(resp.Body)
 	scanBuf := getSSEScannerBuf64K()
 	scanner.Buffer(scanBuf[:0], maxLineSize)
 	defer putSSEScannerBuf64K(scanBuf)
diff --git a/backend/internal/service/openai_ws_http_bridge_test.go b/backend/internal/service/openai_ws_http_bridge_test.go
index 1e352f012efb..887d495202e7 100644
--- a/backend/internal/service/openai_ws_http_bridge_test.go
+++ b/backend/internal/service/openai_ws_http_bridge_test.go
@@ -41,6 +41,81 @@ func TestPrepareOpenAIWSHTTPBridgeBodyStripsWSFields(t *testing.T) {
 	require.Equal(t, "hi", gjson.GetBytes(body, "input").String())
 }
 
+func TestProxyOpenAIWSHTTPBridgeTurnAPIKeyAdaptsClientTools(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	sse := strings.Join([]string{
+		`data: {"type":"response.output_item.added","sequence_number":0,"output_index":0,"item":{"type":"function_call","id":"item_exec","call_id":"call_exec","name":"exec","status":"in_progress"}}`,
+		``,
+		`data: {"type":"response.function_call_arguments.done","sequence_number":1,"output_index":0,"item_id":"item_exec","call_id":"call_exec","name":"exec","arguments":"{\"input\":\"pwd\"}"}`,
+		``,
+		`data: {"type":"response.output_item.done","sequence_number":2,"output_index":0,"item":{"type":"function_call","id":"item_exec","call_id":"call_exec","name":"exec","arguments":"{\"input\":\"pwd\"}","status":"completed"}}`,
+		``,
+		`data: {"type":"response.completed","sequence_number":3,"response":{"id":"resp_tools","status":"completed","output":[{"type":"function_call","id":"item_exec","call_id":"call_exec","name":"exec","arguments":"{\"input\":\"pwd\"}","status":"completed"}],"usage":{"input_tokens":1,"output_tokens":1}}}`,
+		``,
+	}, "\n")
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}},
+		Body:       io.NopCloser(strings.NewReader(sse)),
+	}}
+	svc := &OpenAIGatewayService{
+		cfg:          &config.Config{Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize}},
+		httpUpstream: upstream,
+	}
+	account := &Account{ID: 5659, Platform: PlatformOpenAI, Type: AccountTypeAPIKey, Concurrency: 1}
+	payload := []byte(`{
+		"type":"response.create","model":"gpt-5","stream":true,
+		"tools":[{"type":"custom","name":"exec","description":"Run a command"}],
+		"input":[
+			{"type":"custom_tool_call","id":"previous_item","call_id":"previous_call","name":"exec","input":"echo ready"},
+			{"type":"custom_tool_call_output","call_id":"previous_call","output":"ready"}
+		]
+	}`)
+	recorder := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(recorder)
+	c.Request = httptest.NewRequest(http.MethodGet, "/v1/responses", nil)
+	var events [][]byte
+
+	result, err := svc.proxyOpenAIWSHTTPBridgeTurn(
+		context.Background(), c, account, "test-token", payload, len(payload),
+		"gpt-5", "", "", "", "", 2,
+		func(message []byte) error {
+			events = append(events, append([]byte(nil), message...))
+			return nil
+		},
+	)
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.Equal(t, "function", gjson.GetBytes(upstream.lastBody, "tools.0.type").String())
+	require.Equal(t, "function_call", gjson.GetBytes(upstream.lastBody, "input.0.type").String())
+	require.JSONEq(t, `{"input":"echo ready"}`, gjson.GetBytes(upstream.lastBody, "input.0.arguments").String())
+	require.False(t, gjson.GetBytes(upstream.lastBody, "input.0.input").Exists())
+	require.Equal(t, "function_call_output", gjson.GetBytes(upstream.lastBody, "input.1.type").String())
+
+	var outputDone, completed []byte
+	for _, event := range events {
+		switch gjson.GetBytes(event, "type").String() {
+		case "response.output_item.done":
+			outputDone = event
+		case "response.completed":
+			completed = event
+		}
+	}
+	require.NotEmpty(t, outputDone)
+	require.Equal(t, "custom_tool_call", gjson.GetBytes(outputDone, "item.type").String())
+	require.Equal(t, "pwd", gjson.GetBytes(outputDone, "item.input").String())
+	require.False(t, gjson.GetBytes(outputDone, "item.arguments").Exists())
+	require.NotEmpty(t, completed)
+	require.Equal(t, "custom_tool_call", gjson.GetBytes(completed, "response.output.0.type").String())
+	require.Equal(t, "pwd", gjson.GetBytes(completed, "response.output.0.input").String())
+	require.True(t, result.wsReplayInputExists)
+	require.Len(t, result.wsReplayInput, 1)
+	require.Equal(t, "custom_tool_call", gjson.GetBytes(result.wsReplayInput[0], "type").String())
+	require.Equal(t, "pwd", gjson.GetBytes(result.wsReplayInput[0], "input").String())
+}
+
 func TestOpenAIWSHTTPBridgeDecisionKeepsSmallFramesOnWS(t *testing.T) {
 	svc := &OpenAIGatewayService{
 		cfg: &config.Config{

From 269fbcac0e5b78dd2167c8b15267ed3a9fe7efa5 Mon Sep 17 00:00:00 2001
From: IanShaw027 <ianshaw027@gmail.com>
Date: Tue, 18 Aug 2026 14:50:57 +0800
Subject: [PATCH 58/68] =?UTF-8?q?feat:=20Grok=20=E7=94=A8=E9=87=8F?=
 =?UTF-8?q?=E6=9D=A1=E8=A1=A5=E9=BD=90=E6=9C=AC=E7=AB=99=2024h/7d/30d=20?=
 =?UTF-8?q?=E8=81=9A=E5=90=88?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

把已有的本地窗口统计接到 Free 24h 与付费 7d/30d 进度条，避免只显示官方百分比。
---
 .../components/account/AccountUsageCell.vue   |  16 +-
 .../__tests__/AccountUsageCell.spec.ts        | 151 ++++++++++++++++++
 2 files changed, 165 insertions(+), 2 deletions(-)

diff --git a/frontend/src/components/account/AccountUsageCell.vue b/frontend/src/components/account/AccountUsageCell.vue
index 8a7820cfee15..3291b9b30d13 100644
--- a/frontend/src/components/account/AccountUsageCell.vue
+++ b/frontend/src/components/account/AccountUsageCell.vue
@@ -365,6 +365,7 @@
             label="24h"
             :title="t('admin.accounts.usageWindow.grokFreeQuota24hHint', { limit: formatCompactNumber(grokFreeTokenBar.limit) })"
             :utilization="grokFreeTokenBar.utilization"
+            :window-stats="grokFreeQuotaUsage"
             :show-now-when-idle="true"
             color="emerald"
           />
@@ -378,6 +379,7 @@
             label="7d"
             :utilization="grokWeeklyBillingBar.utilization"
             :resets-at="grokWeeklyBillingBar.resetsAt"
+            :window-stats="grokWeeklyBillingBar.windowStats"
             :show-now-when-idle="true"
             color="indigo"
           />
@@ -386,6 +388,7 @@
             label="30d"
             :utilization="grokMonthlyBillingBar.utilization"
             :resets-at="grokMonthlyBillingBar.resetsAt"
+            :window-stats="grokMonthlyBillingBar.windowStats"
             :show-now-when-idle="true"
             color="indigo"
           />
@@ -1108,9 +1111,16 @@ const geminiUsageBars = computed(() => {
 interface GrokQuotaBarInfo {
   utilization: number
   resetsAt: string | null
+  windowStats?: WindowStats | null
 }
 
 const grokBilling = computed(() => usageInfo.value?.grok_billing || null)
+const grokLocalUsage7d = computed(() => (
+  usageInfo.value?.grok_local_usage_7d || usageInfo.value?.seven_day?.window_stats || null
+))
+const grokLocalUsageMonthly = computed(() => (
+  usageInfo.value?.grok_local_usage_monthly || usageInfo.value?.thirty_day?.window_stats || null
+))
 const grokWeeklyBillingBar = computed((): GrokQuotaBarInfo | null => {
   const billing = grokBilling.value
   if (billing?.period_type?.toLowerCase() !== 'weekly' || billing.usage_percent == null) {
@@ -1118,7 +1128,8 @@ const grokWeeklyBillingBar = computed((): GrokQuotaBarInfo | null => {
   }
   return {
     utilization: Math.min(100, Math.max(0, billing.usage_percent)),
-    resetsAt: billing.period_end || null
+    resetsAt: billing.period_end || null,
+    windowStats: grokLocalUsage7d.value
   }
 })
 // Monthly used/limit % from billing probe (used_percent or derived from cents).
@@ -1142,7 +1153,8 @@ const grokMonthlyBillingBar = computed((): GrokQuotaBarInfo | null => {
   }
   return {
     utilization: Math.min(100, Math.max(0, utilization)),
-    resetsAt: billing.billing_period_end || billing.period_end || null
+    resetsAt: billing.billing_period_end || billing.period_end || null,
+    windowStats: grokLocalUsageMonthly.value
   }
 })
 const formatGrokMoney = (value?: number | null) => {
diff --git a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
index dc071b6de57a..3e048a53462c 100644
--- a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
+++ b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
@@ -1017,6 +1017,157 @@ describe('AccountUsageCell', () => {
     expect(wrapper.text()).toContain('24h|100')
   })
 
+  it('Grok Free 24h bar shows rolling local usage chips', async () => {
+    getUsage.mockResolvedValue({
+      grok_free_token_limit: 1_000_000,
+      grok_billing: { period_type: 'weekly', usage_percent: null, plan: '' },
+      grok_local_usage: {
+        requests: 2,
+        tokens: 250_000,
+        cost: 0,
+        standard_cost: 0
+      },
+      grok_local_usage_24h: {
+        requests: 12,
+        tokens: 750_000,
+        cost: 0.12,
+        standard_cost: 0.12,
+        user_cost: 0.04
+      }
+    })
+
+    const wrapper = mount(AccountUsageCell, {
+      props: {
+        account: makeAccount({ id: 4410, platform: 'grok', type: 'oauth', extra: {} })
+      },
+      global: {
+        stubs: {
+          UsageProgressBar: {
+            props: ['label', 'utilization', 'windowStats'],
+            template: '<div class="usage-bar">{{ label }}|{{ utilization }}|{{ windowStats?.tokens }}</div>'
+          },
+          AccountQuotaInfo: true
+        }
+      }
+    })
+
+    await flushPromises()
+    expect(wrapper.text()).toContain('24h|75|750000')
+    expect(wrapper.text()).not.toContain('|250000')
+    expect(wrapper.text()).not.toContain('7d|')
+  })
+
+  it('Grok SuperGrok and Heavy bars show period-aligned local 7d and 30d usage', async () => {
+    getUsage.mockResolvedValue({
+      subscription_tier: 'SuperGrok Heavy',
+      grok_billing: {
+        period_type: 'weekly',
+        usage_percent: 37,
+        used_percent: 12,
+        monthly_limit_cents: 150_000,
+        period_end: '2026-07-16T03:25:00Z',
+        billing_period_end: '2026-08-01T00:00:00Z',
+        plan: 'SuperGrok Heavy'
+      },
+      grok_local_usage: {
+        requests: 1,
+        tokens: 99,
+        cost: 0,
+        standard_cost: 0
+      },
+      grok_local_usage_24h: {
+        requests: 2,
+        tokens: 100,
+        cost: 0,
+        standard_cost: 0
+      },
+      grok_local_usage_7d: {
+        requests: 8,
+        tokens: 2_200_000,
+        cost: 4.42,
+        standard_cost: 4.42
+      },
+      grok_local_usage_monthly: {
+        requests: 20,
+        tokens: 8_000_000,
+        cost: 18.5,
+        standard_cost: 18.5
+      }
+    })
+
+    const wrapper = mount(AccountUsageCell, {
+      props: {
+        account: makeAccount({ id: 4411, platform: 'grok', type: 'oauth', extra: {} })
+      },
+      global: {
+        stubs: {
+          UsageProgressBar: {
+            props: ['label', 'utilization', 'windowStats'],
+            template: '<div class="usage-bar">{{ label }}|{{ utilization }}|{{ windowStats?.tokens }}</div>'
+          },
+          AccountQuotaInfo: true
+        }
+      }
+    })
+
+    await flushPromises()
+    expect(wrapper.text()).toContain('7d|37|2200000')
+    expect(wrapper.text()).toContain('30d|12|8000000')
+    expect(wrapper.text()).not.toContain('|99')
+    expect(wrapper.text()).not.toContain('|100')
+    expect(wrapper.text()).not.toContain('24h|')
+  })
+
+  it('Grok paid bars fall back to official seven_day and thirty_day window_stats', async () => {
+    getUsage.mockResolvedValue({
+      subscription_tier: 'SuperGrok',
+      grok_billing: {
+        period_type: 'weekly',
+        usage_percent: 20,
+        used_percent: 8,
+        monthly_limit_cents: 25_000,
+        plan: 'SuperGrok'
+      },
+      seven_day: {
+        utilization: 20,
+        window_stats: {
+          requests: 6,
+          tokens: 1_500_000,
+          cost: 3.1,
+          standard_cost: 3.1
+        }
+      },
+      thirty_day: {
+        utilization: 8,
+        window_stats: {
+          requests: 14,
+          tokens: 4_400_000,
+          cost: 9.2,
+          standard_cost: 9.2
+        }
+      }
+    })
+
+    const wrapper = mount(AccountUsageCell, {
+      props: {
+        account: makeAccount({ id: 4412, platform: 'grok', type: 'oauth', extra: {} })
+      },
+      global: {
+        stubs: {
+          UsageProgressBar: {
+            props: ['label', 'utilization', 'windowStats'],
+            template: '<div class="usage-bar">{{ label }}|{{ utilization }}|{{ windowStats?.tokens }}</div>'
+          },
+          AccountQuotaInfo: true
+        }
+      }
+    })
+
+    await flushPromises()
+    expect(wrapper.text()).toContain('7d|20|1500000')
+    expect(wrapper.text()).toContain('30d|8|4400000')
+  })
+
   it('Key 账号在 today stats loading 时显示骨架屏', async () => {
 		const wrapper = mount(AccountUsageCell, {
 		  props: {

From fd42d3722d11ae4980bddd718e94105e42b5ce4f Mon Sep 17 00:00:00 2001
From: IanShaw027 <ianshaw027@gmail.com>
Date: Tue, 18 Aug 2026 14:55:08 +0800
Subject: [PATCH 59/68] fix: hide Grok prepaid and used/limit when they are
 empty

Skip the prepaid chip without a positive balance, and skip used/limit when monthly limit is 0.
---
 .../components/account/AccountUsageCell.vue   | 35 +++++---
 .../__tests__/AccountUsageCell.spec.ts        | 88 +++++++++++++++++++
 2 files changed, 109 insertions(+), 14 deletions(-)

diff --git a/frontend/src/components/account/AccountUsageCell.vue b/frontend/src/components/account/AccountUsageCell.vue
index 3291b9b30d13..966b0a5e084f 100644
--- a/frontend/src/components/account/AccountUsageCell.vue
+++ b/frontend/src/components/account/AccountUsageCell.vue
@@ -397,12 +397,16 @@
             class="flex flex-wrap items-center gap-1 text-[10px] text-gray-500 dark:text-gray-400"
           >
             <span
+              v-if="grokPrepaidMoneyLine.showPrepaid"
               class="rounded bg-emerald-50 px-1 py-0.5 text-emerald-700 dark:bg-emerald-900/30 dark:text-emerald-300"
               :title="t('admin.accounts.usageWindow.grokPrepaid')"
             >
               {{ t('admin.accounts.usageWindow.grokPrepaid') }} ${{ grokPrepaidMoneyLine.prepaid }}
             </span>
-            <span :title="t('admin.accounts.usageWindow.grokMonthlyLimit')">
+            <span
+              v-if="grokPrepaidMoneyLine.showUsedLimit"
+              :title="t('admin.accounts.usageWindow.grokMonthlyLimit')"
+            >
               {{ t('admin.accounts.usageWindow.grokUsed') }}
               {{ grokPrepaidMoneyLine.used }}/{{ grokPrepaidMoneyLine.limit }}
             </span>
@@ -1164,30 +1168,33 @@ const formatGrokMoney = (value?: number | null) => {
   if (value >= 10) return value.toFixed(1)
   return value.toFixed(2)
 }
-// Prepaid money line for paid Grok: show when prepaid_balance is present.
-// Monthly used/limit numbers are optional context; primary progress is the 30d bar.
+// Prepaid chip only when there is a positive prepaid balance.
+// Used/limit only when monthly limit is a positive number (0 means unlimited / unset).
 const grokPrepaidMoneyLine = computed(() => {
   const billing = grokBilling.value
   if (!billing) return null
   const prepaid = billing.prepaid_balance
-  // "只针对预付": only render when prepaid field exists (including $0.00).
-  if (prepaid == null || !Number.isFinite(prepaid)) return null
+  const showPrepaid = prepaid != null && Number.isFinite(prepaid) && prepaid > 0
+  const limitRaw =
+    billing.monthly_limit != null
+      ? billing.monthly_limit
+      : billing.monthly_limit_cents != null
+        ? billing.monthly_limit_cents / 100
+        : null
+  const showUsedLimit = limitRaw != null && Number.isFinite(limitRaw) && limitRaw > 0
+  if (!showPrepaid && !showUsedLimit) return null
   const used =
     billing.monthly_used != null
       ? billing.monthly_used
       : billing.used_cents != null
         ? billing.used_cents / 100
         : 0
-  const limit =
-    billing.monthly_limit != null
-      ? billing.monthly_limit
-      : billing.monthly_limit_cents != null
-        ? billing.monthly_limit_cents / 100
-        : 0
   return {
-    prepaid: formatGrokMoney(prepaid),
-    used: formatGrokMoney(used),
-    limit: formatGrokMoney(limit)
+    showPrepaid,
+    showUsedLimit,
+    prepaid: showPrepaid ? formatGrokMoney(prepaid) : null,
+    used: showUsedLimit ? formatGrokMoney(used) : null,
+    limit: showUsedLimit ? formatGrokMoney(limitRaw) : null
   }
 })
 const grokPlanLabelIsFree = (value: string) => value.includes('free') || value.includes('basic')
diff --git a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
index 3e048a53462c..1ad6866dd7e1 100644
--- a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
+++ b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
@@ -1168,6 +1168,94 @@ describe('AccountUsageCell', () => {
     expect(wrapper.text()).toContain('30d|8|4400000')
   })
 
+  it('Grok paid hides zero prepaid and hides used/limit when monthly limit is 0', async () => {
+    getUsage.mockResolvedValue({
+      subscription_tier: 'SuperGrok',
+      grok_billing: {
+        period_type: 'weekly',
+        usage_percent: 20,
+        prepaid_balance: 0,
+        monthly_limit: 0,
+        monthly_used: 3.5,
+        plan: 'SuperGrok'
+      }
+    })
+
+    const wrapper = mount(AccountUsageCell, {
+      props: {
+        account: makeAccount({ id: 4413, platform: 'grok', type: 'oauth', extra: {} })
+      },
+      global: {
+        stubs: {
+          UsageProgressBar: true,
+          AccountQuotaInfo: true
+        }
+      }
+    })
+
+    await flushPromises()
+    expect(wrapper.text()).not.toContain('admin.accounts.usageWindow.grokPrepaid')
+    expect(wrapper.text()).not.toContain('admin.accounts.usageWindow.grokUsed')
+    expect(wrapper.text()).not.toContain('3.5/0')
+  })
+
+  it('Grok paid shows used/limit without prepaid, and prepaid without a zero monthly limit', async () => {
+    getUsage.mockResolvedValueOnce({
+      subscription_tier: 'SuperGrok',
+      grok_billing: {
+        period_type: 'weekly',
+        usage_percent: 20,
+        monthly_limit: 25,
+        monthly_used: 3.5,
+        plan: 'SuperGrok'
+      }
+    })
+
+    const usedOnly = mount(AccountUsageCell, {
+      props: {
+        account: makeAccount({ id: 4414, platform: 'grok', type: 'oauth', extra: {} })
+      },
+      global: {
+        stubs: {
+          UsageProgressBar: true,
+          AccountQuotaInfo: true
+        }
+      }
+    })
+    await flushPromises()
+    expect(usedOnly.text()).not.toContain('admin.accounts.usageWindow.grokPrepaid')
+    expect(usedOnly.text()).toContain('admin.accounts.usageWindow.grokUsed')
+    expect(usedOnly.text()).toContain('3.50/25.0')
+
+    getUsage.mockResolvedValueOnce({
+      subscription_tier: 'SuperGrok Heavy',
+      grok_billing: {
+        period_type: 'weekly',
+        usage_percent: 20,
+        prepaid_balance: 12.5,
+        monthly_limit: 0,
+        monthly_used: 8,
+        plan: 'SuperGrok Heavy'
+      }
+    })
+    const prepaidOnly = mount(AccountUsageCell, {
+      props: {
+        account: makeAccount({ id: 4415, platform: 'grok', type: 'oauth', extra: {} })
+      },
+      global: {
+        stubs: {
+          UsageProgressBar: true,
+          AccountQuotaInfo: true
+        }
+      }
+    })
+    await flushPromises()
+    expect(prepaidOnly.text()).toContain('admin.accounts.usageWindow.grokPrepaid')
+    expect(prepaidOnly.text()).toContain('$12.5')
+    expect(prepaidOnly.text()).not.toContain('admin.accounts.usageWindow.grokUsed')
+    expect(prepaidOnly.text()).not.toContain('8.00/0')
+  })
+
   it('Key 账号在 today stats loading 时显示骨架屏', async () => {
 		const wrapper = mount(AccountUsageCell, {
 		  props: {

From 9617775f9a5b4999e7ee59e15d96184253f8f2e4 Mon Sep 17 00:00:00 2001
From: yaxin <dev@sub2api.org>
Date: Tue, 18 Aug 2026 15:02:25 +0800
Subject: [PATCH 60/68] fix(repo): tolerate ErrTxStarted for tx-bound clients
 and harden test stubs

- user_repo.create(): keep the TxFromContext fast path, but restore
  tolerance for dbent.ErrTxStarted in the self-owned-transaction branch.
  ent's Client.Tx only inspects the driver type, so a repository built
  from a tx-bound client (client-injected transactions, e.g. the
  integration fixture testEntTx + tx.Client()) hits ErrTxStarted; reuse
  that client instead of failing. Fixes the two red integration tests in
  allowed_groups_contract_integration_test.go.
- createUserAndClaimInvitation: roll back via defer (matching the OAuth
  registration precedent) so a panic inside the transaction cannot leak
  the connection.
- settingRepoStub: guard call counters and state with a mutex; the new
  concurrency regression test exercises it from multiple goroutines and
  the unsynchronized counters were flagged by -race.
---
 backend/internal/repository/user_repo.go         | 16 +++++++++++-----
 backend/internal/service/auth_service.go         |  2 +-
 .../service/auth_service_register_test.go        |  6 ++++++
 3 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/backend/internal/repository/user_repo.go b/backend/internal/repository/user_repo.go
index 4b8841bb6fac..c072f09b2790 100644
--- a/backend/internal/repository/user_repo.go
+++ b/backend/internal/repository/user_repo.go
@@ -82,13 +82,19 @@ func (r *userRepository) create(ctx context.Context, userIn *service.User, guard
 		txClient = existingTx.Client()
 	} else {
 		tx, err := r.client.Tx(ctx)
-		if err != nil {
+		switch {
+		case errors.Is(err, dbent.ErrTxStarted):
+			// r.client 本身已是事务绑定 client（client 注入式事务，如集成测试
+			// 夹具 tx.Client()）：直接复用，提交/回滚由 client 的持有方负责。
+			txClient = r.client
+		case err != nil:
 			return err
+		default:
+			ownedTx = tx
+			defer func() { _ = ownedTx.Rollback() }()
+			txClient = tx.Client()
+			txCtx = dbent.NewTxContext(ctx, tx)
 		}
-		ownedTx = tx
-		defer func() { _ = ownedTx.Rollback() }()
-		txClient = tx.Client()
-		txCtx = dbent.NewTxContext(ctx, tx)
 	}
 
 	lockKeys := []string{normalizedEmailUniquenessLockKey(userIn.Email)}
diff --git a/backend/internal/service/auth_service.go b/backend/internal/service/auth_service.go
index 72adbe890232..f17cfbb4d784 100644
--- a/backend/internal/service/auth_service.go
+++ b/backend/internal/service/auth_service.go
@@ -1317,9 +1317,9 @@ func (s *AuthService) createUserAndClaimInvitation(ctx context.Context, user *Us
 		logger.LegacyPrintf("service.auth", "[Auth] Failed to start registration transaction: %v", err)
 		return ErrServiceUnavailable
 	}
+	defer func() { _ = tx.Rollback() }()
 	execCtx := dbent.NewTxContext(ctx, tx)
 	if err := commitUser(execCtx); err != nil {
-		_ = tx.Rollback()
 		return err
 	}
 	if err := tx.Commit(); err != nil {
diff --git a/backend/internal/service/auth_service_register_test.go b/backend/internal/service/auth_service_register_test.go
index 04d7efaeb960..c29fdac0c38b 100644
--- a/backend/internal/service/auth_service_register_test.go
+++ b/backend/internal/service/auth_service_register_test.go
@@ -5,6 +5,7 @@ package service
 import (
 	"context"
 	"errors"
+	"sync"
 	"testing"
 	"time"
 
@@ -14,6 +15,7 @@ import (
 )
 
 type settingRepoStub struct {
+	mu               sync.Mutex
 	values           map[string]string
 	err              error
 	getValueCalls    int
@@ -25,6 +27,8 @@ func (s *settingRepoStub) Get(ctx context.Context, key string) (*Setting, error)
 }
 
 func (s *settingRepoStub) GetValue(ctx context.Context, key string) (string, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
 	s.getValueCalls++
 	if s.err != nil {
 		return "", s.err
@@ -40,6 +44,8 @@ func (s *settingRepoStub) Set(ctx context.Context, key, value string) error {
 }
 
 func (s *settingRepoStub) GetMultiple(ctx context.Context, keys []string) (map[string]string, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
 	s.getMultipleCalls++
 	if s.err != nil {
 		return nil, s.err

From 1ba92449c7372493b2c94d42e290c0959f4cf760 Mon Sep 17 00:00:00 2001
From: yaxin <dev@sub2api.org>
Date: Tue, 18 Aug 2026 15:05:54 +0800
Subject: [PATCH 61/68] fix(gemini): wire includeServerSideToolInvocations into
 the typed transform path

The struct field alone never reached the wire: the raw passthrough
pipeline is covered by enableMixedGeminiToolInvocations (#5711), but
TransformClaudeToGeminiWithOptions builds GeminiToolConfig from scratch
and never set the flag, so gemini-* models entering through the Claude
format gateway could still hit the upstream 400 from issue #5709.

- Set IncludeServerSideToolInvocations=true when the built tool
  declarations mix functionDeclarations with googleSearch, matching the
  raw-path injection semantics.
- Replace the marshal-roundtrip-only test with behavior tests that
  drive TransformClaudeToGeminiWithOptions: mixed tools set the flag,
  function-only and web-search-only requests leave it unset.
---
 .../pkg/antigravity/request_transformer.go    | 22 +++++++
 .../antigravity/request_transformer_test.go   | 66 ++++++++++++++-----
 2 files changed, 71 insertions(+), 17 deletions(-)

diff --git a/backend/internal/pkg/antigravity/request_transformer.go b/backend/internal/pkg/antigravity/request_transformer.go
index 7d0a707959b8..9da3cc33a297 100644
--- a/backend/internal/pkg/antigravity/request_transformer.go
+++ b/backend/internal/pkg/antigravity/request_transformer.go
@@ -156,6 +156,13 @@ func TransformClaudeToGeminiWithOptions(claudeReq *ClaudeRequest, projectID, map
 				Mode: "VALIDATED",
 			},
 		}
+		// 内置工具（googleSearch）与函数调用混用时，上游要求显式开启
+		// includeServerSideToolInvocations，否则返回 400（issue #5709）。
+		// 与 raw 透传路的 enableMixedGeminiToolInvocations 注入保持同一语义。
+		if hasMixedToolInvocations(tools) {
+			enabled := true
+			innerRequest.ToolConfig.IncludeServerSideToolInvocations = &enabled
+		}
 	}
 
 	if systemInstruction != nil {
@@ -703,6 +710,21 @@ func isWebSearchTool(tool ClaudeTool) bool {
 	}
 }
 
+// hasMixedToolInvocations 判断构建后的工具声明是否同时包含函数声明与内置工具
+// （googleSearch）。仅在两者并存时需要开启 includeServerSideToolInvocations。
+func hasMixedToolInvocations(declarations []GeminiToolDeclaration) bool {
+	hasFunc, hasBuiltin := false, false
+	for _, d := range declarations {
+		if len(d.FunctionDeclarations) > 0 {
+			hasFunc = true
+		}
+		if d.GoogleSearch != nil {
+			hasBuiltin = true
+		}
+	}
+	return hasFunc && hasBuiltin
+}
+
 // buildTools 构建 tools
 func buildTools(tools []ClaudeTool) []GeminiToolDeclaration {
 	if len(tools) == 0 {
diff --git a/backend/internal/pkg/antigravity/request_transformer_test.go b/backend/internal/pkg/antigravity/request_transformer_test.go
index 008e54724a93..05764ef7ff60 100644
--- a/backend/internal/pkg/antigravity/request_transformer_test.go
+++ b/backend/internal/pkg/antigravity/request_transformer_test.go
@@ -567,25 +567,57 @@ func TestTransformClaudeToGeminiWithOptions_PreservesWebSearchAlongsideFunctions
 }
 
 func TestGeminiToolConfig_IncludeServerSideToolInvocations(t *testing.T) {
-	t.Run("serialize and deserialize toolConfig with includeServerSideToolInvocations", func(t *testing.T) {
-		trueVal := true
-		cfg := GeminiToolConfig{
-			FunctionCallingConfig: &GeminiFunctionCallingConfig{
-				Mode: "VALIDATED",
-			},
-			IncludeServerSideToolInvocations: &trueVal,
-		}
+	functionTool := ClaudeTool{
+		Name:        "get_weather",
+		Description: "Get weather information",
+		InputSchema: map[string]any{"type": "object"},
+	}
+	webSearchTool := ClaudeTool{
+		Type: "web_search_20250305",
+		Name: "web_search",
+	}
 
-		data, err := json.Marshal(cfg)
+	transform := func(t *testing.T, tools []ClaudeTool) (V1InternalRequest, string) {
+		t.Helper()
+		body, err := TransformClaudeToGeminiWithOptions(&ClaudeRequest{
+			Model: "claude-3-5-sonnet-latest",
+			Messages: []ClaudeMessage{
+				{
+					Role:    "user",
+					Content: json.RawMessage(`[{"type":"text","text":"hello"}]`),
+				},
+			},
+			Tools: tools,
+		}, "project-1", "gemini-2.5-flash", DefaultTransformOptions())
 		require.NoError(t, err)
-		require.Contains(t, string(data), `"includeServerSideToolInvocations":true`)
 
-		var decoded GeminiToolConfig
-		err = json.Unmarshal(data, &decoded)
-		require.NoError(t, err)
-		require.NotNil(t, decoded.IncludeServerSideToolInvocations)
-		require.True(t, *decoded.IncludeServerSideToolInvocations)
-		require.Equal(t, "VALIDATED", decoded.FunctionCallingConfig.Mode)
+		var req V1InternalRequest
+		require.NoError(t, json.Unmarshal(body, &req))
+		return req, string(body)
+	}
+
+	t.Run("mixed builtin and function tools enable server-side tool invocations", func(t *testing.T) {
+		req, raw := transform(t, []ClaudeTool{functionTool, webSearchTool})
+
+		require.NotNil(t, req.Request.ToolConfig)
+		require.NotNil(t, req.Request.ToolConfig.IncludeServerSideToolInvocations)
+		require.True(t, *req.Request.ToolConfig.IncludeServerSideToolInvocations)
+		require.Contains(t, raw, `"includeServerSideToolInvocations":true`)
 	})
-}
 
+	t.Run("function tools only leave the flag unset", func(t *testing.T) {
+		req, raw := transform(t, []ClaudeTool{functionTool})
+
+		require.NotNil(t, req.Request.ToolConfig)
+		require.Nil(t, req.Request.ToolConfig.IncludeServerSideToolInvocations)
+		require.NotContains(t, raw, "includeServerSideToolInvocations")
+	})
+
+	t.Run("web search only leaves the flag unset", func(t *testing.T) {
+		req, raw := transform(t, []ClaudeTool{webSearchTool})
+
+		require.NotNil(t, req.Request.ToolConfig)
+		require.Nil(t, req.Request.ToolConfig.IncludeServerSideToolInvocations)
+		require.NotContains(t, raw, "includeServerSideToolInvocations")
+	})
+}

From a341239596caa92d3a546c84126283787de5670e Mon Sep 17 00:00:00 2001
From: yaxin <dev@sub2api.org>
Date: Tue, 18 Aug 2026 15:20:08 +0800
Subject: [PATCH 62/68] fix(fingerprint): align credential-face identity with
 the real client and de-drift models version

- Replace ApplyCodexCanonicalIdentity with CodexCanonicalAuthIdentity /
  ApplyCodexCanonicalAuthIdentity: the credential face (auth.openai.com
  token exchange / refresh / PAT whoami) now sends the originator +
  canonical User-Agent pair and no version header, matching codex-rs
  default_headers(); the version gate (#3901) only exists on the
  /backend-api/codex inference face. whoami keeps its original header
  shape (originator + UA) with the canonical UA source.
- Token exchange and refresh send the full pair instead of a bare UA,
  eliminating the half-identity (UA without originator) combination no
  real client ever emits.
- Codex models manifest: the Version header now follows the client's
  own client_version when it is valid and >= the upstream floor (same
  source as the query param, restoring the pre-refactor consistency),
  falling back to the canonical version otherwise; the query param
  keeps its verbatim passthrough contract.
- Drop the now-unreferenced openAICodexProbeVersion constant and its
  vacuous consistency assertions; probes resolve their version through
  resolveCodexOutboundIdentity at runtime.
---
 .../repository/openai_oauth_service.go        |  8 +++++--
 .../repository/openai_oauth_service_test.go   | 16 ++++++++++++--
 .../internal/service/account_usage_service.go | 17 +++++++--------
 .../service/openai_capacity_shed_test.go      |  1 -
 .../internal/service/openai_codex_identity.go | 21 +++++++++++++------
 .../service/openai_codex_identity_test.go     |  5 +++--
 .../service/openai_codex_models_service.go    | 10 ++++++++-
 .../openai_codex_models_service_test.go       |  4 ++--
 .../service/openai_codex_pat_service.go       |  2 +-
 .../openai_codex_version_consistency_test.go  |  3 ---
 10 files changed, 58 insertions(+), 29 deletions(-)

diff --git a/backend/internal/repository/openai_oauth_service.go b/backend/internal/repository/openai_oauth_service.go
index 32b1652f6554..e7a34ad6a04a 100644
--- a/backend/internal/repository/openai_oauth_service.go
+++ b/backend/internal/repository/openai_oauth_service.go
@@ -46,9 +46,11 @@ func (s *openaiOAuthService) ExchangeCode(ctx context.Context, code, codeVerifie
 
 	var tokenResp openai.TokenResponse
 
+	authUA, authOriginator := service.CodexCanonicalAuthIdentity()
 	resp, err := client.R().
 		SetContext(ctx).
-		SetHeader("User-Agent", service.CodexCanonicalUserAgent()).
+		SetHeader("User-Agent", authUA).
+		SetHeader("originator", authOriginator).
 		SetFormDataFromValues(formData).
 		SetSuccessResult(&tokenResp).
 		Post(s.tokenURL)
@@ -94,9 +96,11 @@ func (s *openaiOAuthService) refreshTokenWithClientID(ctx context.Context, refre
 
 	var tokenResp openai.TokenResponse
 
+	authUA, authOriginator := service.CodexCanonicalAuthIdentity()
 	resp, err := client.R().
 		SetContext(ctx).
-		SetHeader("User-Agent", service.CodexCanonicalUserAgent()).
+		SetHeader("User-Agent", authUA).
+		SetHeader("originator", authOriginator).
 		SetFormDataFromValues(formData).
 		SetSuccessResult(&tokenResp).
 		Post(s.tokenURL)
diff --git a/backend/internal/repository/openai_oauth_service_test.go b/backend/internal/repository/openai_oauth_service_test.go
index b3feda7483d4..d43b208d4e86 100644
--- a/backend/internal/repository/openai_oauth_service_test.go
+++ b/backend/internal/repository/openai_oauth_service_test.go
@@ -78,11 +78,17 @@ func (s *OpenAIOAuthServiceSuite) TestExchangeCode_DefaultRedirectURI() {
 			w.WriteHeader(http.StatusBadRequest)
 			return
 		}
-		if got := r.Header.Get("User-Agent"); got != service.CodexCanonicalUserAgent() {
+		wantUA, wantOriginator := service.CodexCanonicalAuthIdentity()
+		if got := r.Header.Get("User-Agent"); got != wantUA {
 			errCh <- "user-agent mismatch"
 			w.WriteHeader(http.StatusBadRequest)
 			return
 		}
+		if got := r.Header.Get("originator"); got != wantOriginator {
+			errCh <- "originator mismatch"
+			w.WriteHeader(http.StatusBadRequest)
+			return
+		}
 
 		w.Header().Set("Content-Type", "application/json")
 		_, _ = io.WriteString(w, `{"access_token":"at","refresh_token":"rt","token_type":"bearer","expires_in":3600}`)
@@ -127,11 +133,17 @@ func (s *OpenAIOAuthServiceSuite) TestRefreshToken_FormFields() {
 			w.WriteHeader(http.StatusBadRequest)
 			return
 		}
-		if got := r.Header.Get("User-Agent"); got != service.CodexCanonicalUserAgent() {
+		wantUA, wantOriginator := service.CodexCanonicalAuthIdentity()
+		if got := r.Header.Get("User-Agent"); got != wantUA {
 			errCh <- "user-agent mismatch"
 			w.WriteHeader(http.StatusBadRequest)
 			return
 		}
+		if got := r.Header.Get("originator"); got != wantOriginator {
+			errCh <- "originator mismatch"
+			w.WriteHeader(http.StatusBadRequest)
+			return
+		}
 
 		w.Header().Set("Content-Type", "application/json")
 		_, _ = io.WriteString(w, `{"access_token":"at2","refresh_token":"rt2","token_type":"bearer","expires_in":3600}`)
diff --git a/backend/internal/service/account_usage_service.go b/backend/internal/service/account_usage_service.go
index 3cdc71337752..9d82d5caa169 100644
--- a/backend/internal/service/account_usage_service.go
+++ b/backend/internal/service/account_usage_service.go
@@ -105,15 +105,14 @@ type antigravityUsageCache struct {
 }
 
 const (
-	apiCacheTTL             = 3 * time.Minute
-	apiErrorCacheTTL        = 1 * time.Minute        // 负缓存 TTL：429 等错误缓存 1 分钟
-	antigravityErrorTTL     = 1 * time.Minute        // Antigravity 错误缓存 TTL（可恢复错误）
-	apiQueryMaxJitter       = 800 * time.Millisecond // 用量查询最大随机延迟
-	windowStatsCacheTTL     = 1 * time.Minute
-	openAIProbeCacheTTL     = 10 * time.Minute
-	grokProbeRetryTTL       = 1 * time.Minute
-	grokFreeQuotaWindow     = 24 * time.Hour
-	openAICodexProbeVersion = codexCLIVersion // 编译期兜底；运行时探针 version 走 CodexCanonicalClientVersion
+	apiCacheTTL         = 3 * time.Minute
+	apiErrorCacheTTL    = 1 * time.Minute        // 负缓存 TTL：429 等错误缓存 1 分钟
+	antigravityErrorTTL = 1 * time.Minute        // Antigravity 错误缓存 TTL（可恢复错误）
+	apiQueryMaxJitter   = 800 * time.Millisecond // 用量查询最大随机延迟
+	windowStatsCacheTTL = 1 * time.Minute
+	openAIProbeCacheTTL = 10 * time.Minute
+	grokProbeRetryTTL   = 1 * time.Minute
+	grokFreeQuotaWindow = 24 * time.Hour
 )
 
 // UsageCache 封装账户使用量相关的缓存
diff --git a/backend/internal/service/openai_capacity_shed_test.go b/backend/internal/service/openai_capacity_shed_test.go
index e68a48db1cb5..eef8261bb665 100644
--- a/backend/internal/service/openai_capacity_shed_test.go
+++ b/backend/internal/service/openai_capacity_shed_test.go
@@ -250,7 +250,6 @@ func TestCodexOutboundVersionHasSingleSource(t *testing.T) {
 		strings.HasPrefix(codexCLIUserAgent, openai.CodexDefaultOriginator+"/"+codexCLIVersion+" "),
 		"codexCLIUserAgent=%q 必须以 codexCLIVersion=%q 作为版本段", codexCLIUserAgent, codexCLIVersion,
 	)
-	require.Equal(t, codexCLIVersion, openAICodexProbeVersion)
 	require.GreaterOrEqual(t, CompareVersions(codexCLIVersion, codexUpstreamMinVersion), 0,
 		"codexCLIVersion=%q 不得低于上游最低门槛 %q", codexCLIVersion, codexUpstreamMinVersion,
 	)
diff --git a/backend/internal/service/openai_codex_identity.go b/backend/internal/service/openai_codex_identity.go
index 7d9edc4e28a6..2cb8a45c402a 100644
--- a/backend/internal/service/openai_codex_identity.go
+++ b/backend/internal/service/openai_codex_identity.go
@@ -83,15 +83,24 @@ func CodexCanonicalUserAgent() string {
 	return resolveCodexOutboundIdentity("").userAgent
 }
 
-// ApplyCodexCanonicalIdentity 为无账号句柄的出站请求写入与推理同源的身份三元组。
-func ApplyCodexCanonicalIdentity(h http.Header) {
+// CodexCanonicalAuthIdentity 返回凭据面（auth.openai.com：换 Token / 刷新 / whoami）
+// 出站请求的身份对：规范 User-Agent 与配套 originator，与推理解析链同源。
+// 凭据面不发 version 头——真实 Codex 客户端在该面只携带 originator 与 User-Agent
+// （codex-rs login/default_client.rs 的 default_headers()），version 门槛
+// （issue #3901）只存在于 /backend-api/codex 推理面。
+func CodexCanonicalAuthIdentity() (userAgent, originator string) {
+	identity := resolveCodexOutboundIdentity("")
+	return identity.userAgent, identity.originator
+}
+
+// ApplyCodexCanonicalAuthIdentity 为凭据面出站请求写入身份对（不含 version）。
+func ApplyCodexCanonicalAuthIdentity(h http.Header) {
 	if h == nil {
 		return
 	}
-	identity := resolveCodexOutboundIdentity("")
-	h.Set("user-agent", identity.userAgent)
-	h.Set("originator", identity.originator)
-	h.Set("version", identity.version)
+	userAgent, originator := CodexCanonicalAuthIdentity()
+	h.Set("user-agent", userAgent)
+	h.Set("originator", originator)
 }
 
 // CodexCanonicalClientVersion 返回当前生效的 Codex 客户端版本号。
diff --git a/backend/internal/service/openai_codex_identity_test.go b/backend/internal/service/openai_codex_identity_test.go
index 782064643159..15768e5ea5e0 100644
--- a/backend/internal/service/openai_codex_identity_test.go
+++ b/backend/internal/service/openai_codex_identity_test.go
@@ -333,10 +333,11 @@ func TestCodexCanonicalUserAgentFollowsResolver(t *testing.T) {
 	require.Equal(t, "0.200.1", CodexCanonicalClientVersion())
 
 	h := make(http.Header)
-	ApplyCodexCanonicalIdentity(h)
+	ApplyCodexCanonicalAuthIdentity(h)
 	require.Equal(t, "codex_cli_rs", h.Get("originator"))
 	require.Equal(t, "codex_cli_rs/0.200.1"+codexCLIUserAgentSuffix, h.Get("user-agent"))
-	require.Equal(t, "0.200.1", h.Get("version"))
+	// 凭据面不发 version 头（真实客户端在 auth.openai.com 只带 originator + UA）。
+	require.Empty(t, h.Get("version"))
 }
 
 func TestCodexCanonicalUserAgentFallsBackWithoutResolver(t *testing.T) {
diff --git a/backend/internal/service/openai_codex_models_service.go b/backend/internal/service/openai_codex_models_service.go
index 8b5bcda91552..ee5200301704 100644
--- a/backend/internal/service/openai_codex_models_service.go
+++ b/backend/internal/service/openai_codex_models_service.go
@@ -311,7 +311,15 @@ func (s *OpenAIGatewayService) FetchCodexModelsManifest(ctx context.Context, acc
 	identity := resolveCodexOutboundIdentity(overrideUA)
 	headers.Set("Originator", identity.originator)
 	headers.Set("User-Agent", identity.userAgent)
-	headers.Set("Version", identity.version)
+	// Version 头优先与 client_version 查询参数同源：客户端自报版本合法且不低于上游
+	// 门槛时原样使用；否则回退规范版本，避免陈旧 version 触发上游 404（issue #3901）。
+	// client_version 查询参数本身始终按客户端原值透传（内容协商语义，契约见
+	// TestFetchCodexModelsManifestPassthrough）。
+	headerVersion := NormalizeCodexClientVersion(clientVersion)
+	if headerVersion == "" || CompareVersions(headerVersion, codexUpstreamMinVersion) < 0 {
+		headerVersion = identity.version
+	}
+	headers.Set("Version", headerVersion)
 
 	proxyURL := ""
 	if account.ProxyID != nil && account.Proxy != nil {
diff --git a/backend/internal/service/openai_codex_models_service_test.go b/backend/internal/service/openai_codex_models_service_test.go
index c7eb7e29ec52..8f4c4957559f 100644
--- a/backend/internal/service/openai_codex_models_service_test.go
+++ b/backend/internal/service/openai_codex_models_service_test.go
@@ -451,8 +451,8 @@ func TestFetchCodexModelsManifestAPIKeyCustomUpstream(t *testing.T) {
 	if gotRequest.Header.Get("Originator") != openai.CodexDefaultOriginator {
 		t.Errorf("originator header: got %q", gotRequest.Header.Get("Originator"))
 	}
-	if gotRequest.Header.Get("Version") != CodexCanonicalClientVersion() {
-		t.Errorf("version header: got %q", gotRequest.Header.Get("Version"))
+	if gotRequest.Header.Get("Version") != "0.144.0" {
+		t.Errorf("version header must match the client_version query param: got %q", gotRequest.Header.Get("Version"))
 	}
 	if gotRequest.Header.Get("User-Agent") != CodexCanonicalUserAgent() {
 		t.Errorf("user-agent header: got %q", gotRequest.Header.Get("User-Agent"))
diff --git a/backend/internal/service/openai_codex_pat_service.go b/backend/internal/service/openai_codex_pat_service.go
index f4c4cc9f1138..fa89aa962dd6 100644
--- a/backend/internal/service/openai_codex_pat_service.go
+++ b/backend/internal/service/openai_codex_pat_service.go
@@ -58,7 +58,7 @@ func (s *OpenAIOAuthService) ValidateCodexPersonalAccessToken(ctx context.Contex
 	}
 	req.Header.Set("authorization", "Bearer "+accessToken)
 	req.Header.Set("accept", "application/json")
-	ApplyCodexCanonicalIdentity(req.Header)
+	ApplyCodexCanonicalAuthIdentity(req.Header)
 
 	resp, err := client.Do(req)
 	if err != nil {
diff --git a/backend/internal/service/openai_codex_version_consistency_test.go b/backend/internal/service/openai_codex_version_consistency_test.go
index aede2ae4329c..fe95994237ce 100644
--- a/backend/internal/service/openai_codex_version_consistency_test.go
+++ b/backend/internal/service/openai_codex_version_consistency_test.go
@@ -11,9 +11,6 @@ import (
 )
 
 func TestCodexVersionConstants_Consistency(t *testing.T) {
-	require.Equal(t, codexCLIVersion, openAICodexProbeVersion,
-		"codexCLIVersion and openAICodexProbeVersion must stay in sync")
-
 	require.True(t, strings.Contains(codexCLIUserAgent, openai.CodexDefaultOriginator+"/"+codexCLIVersion),
 		"codexCLIUserAgent must embed codexCLIVersion")
 

From c253bd2c72dcce1aee21a4bc671ad23eb1bf5a34 Mon Sep 17 00:00:00 2001
From: hansnow <hansnow2012@gmail.com>
Date: Tue, 18 Aug 2026 15:48:02 +0800
Subject: [PATCH 63/68] fix(openai): restore client tools in terminal events

---
 .../pkg/apicompat/responses_client_tools.go   | 11 +++-
 .../apicompat/responses_client_tools_test.go  | 27 ++++++++++
 .../openai_gateway_grok_tool_protocol.go      | 12 ++---
 .../service/openai_ws_http_bridge_test.go     | 54 +++++++++++++++++++
 4 files changed, 97 insertions(+), 7 deletions(-)

diff --git a/backend/internal/pkg/apicompat/responses_client_tools.go b/backend/internal/pkg/apicompat/responses_client_tools.go
index daffc984d5d0..c4b1e0018edd 100644
--- a/backend/internal/pkg/apicompat/responses_client_tools.go
+++ b/backend/internal/pkg/apicompat/responses_client_tools.go
@@ -430,7 +430,7 @@ func (r *ResponsesClientToolStreamRestorer) RestoreEvent(payload []byte) ([][]by
 	if err := json.Unmarshal(payload, &wire); err != nil {
 		return nil, false, err
 	}
-	if wire.Type == "response.completed" || wire.Type == "response.incomplete" || wire.Type == "response.failed" {
+	if isResponsesClientToolTerminalEvent(wire.Type) {
 		restored, changed, err := RestoreResponsesClientToolPayload(payload, r.adapter)
 		if err != nil {
 			return nil, false, err
@@ -465,6 +465,15 @@ func (r *ResponsesClientToolStreamRestorer) RestoreEvent(payload []byte) ([][]by
 	return result, true, nil
 }
 
+func isResponsesClientToolTerminalEvent(typ string) bool {
+	switch strings.TrimSpace(typ) {
+	case "response.completed", "response.done", "response.incomplete", "response.failed", "response.cancelled", "response.canceled":
+		return true
+	default:
+		return false
+	}
+}
+
 func (r *ResponsesClientToolStreamRestorer) clientToolEventPayload(payload []byte) bool {
 	var raw struct {
 		ItemID      string `json:"item_id"`
diff --git a/backend/internal/pkg/apicompat/responses_client_tools_test.go b/backend/internal/pkg/apicompat/responses_client_tools_test.go
index 91ee0890a624..cc1ed467ba7c 100644
--- a/backend/internal/pkg/apicompat/responses_client_tools_test.go
+++ b/backend/internal/pkg/apicompat/responses_client_tools_test.go
@@ -174,3 +174,30 @@ func TestResponsesClientToolStreamRestorer_RawEventsPreserveUnknownFieldsAndOutp
 	require.Len(t, done, 2)
 	require.Equal(t, "pwd", done[1].Input)
 }
+
+func TestResponsesClientToolStreamRestorer_RestoresAllTerminalEvents(t *testing.T) {
+	for _, eventType := range []string{
+		"response.completed",
+		"response.done",
+		"response.incomplete",
+		"response.failed",
+		"response.cancelled",
+		"response.canceled",
+	} {
+		t.Run(eventType, func(t *testing.T) {
+			restorer := NewResponsesClientToolStreamRestorer(ResponsesClientToolMapping{CustomTools: map[string]bool{"exec": true}})
+			payload := []byte(`{"type":"` + eventType + `","sequence_number":7,"response":{"id":"resp_tools","output":[{"type":"function_call","id":"item_exec","call_id":"call_exec","name":"exec","arguments":"{\"input\":\"pwd\"}"}]}}`)
+
+			restored, changed, err := restorer.RestoreEvent(payload)
+
+			require.NoError(t, err)
+			require.True(t, changed)
+			require.Len(t, restored, 1)
+			require.Equal(t, eventType, gjson.GetBytes(restored[0], "type").String())
+			require.Equal(t, int64(7), gjson.GetBytes(restored[0], "sequence_number").Int())
+			require.Equal(t, "custom_tool_call", gjson.GetBytes(restored[0], "response.output.0.type").String())
+			require.Equal(t, "pwd", gjson.GetBytes(restored[0], "response.output.0.input").String())
+			require.False(t, gjson.GetBytes(restored[0], "response.output.0.arguments").Exists())
+		})
+	}
+}
diff --git a/backend/internal/service/openai_gateway_grok_tool_protocol.go b/backend/internal/service/openai_gateway_grok_tool_protocol.go
index f1683a7adb46..11122172ae9c 100644
--- a/backend/internal/service/openai_gateway_grok_tool_protocol.go
+++ b/backend/internal/service/openai_gateway_grok_tool_protocol.go
@@ -91,12 +91,12 @@ func restoreGrokResponsesClientToolPayload(c *gin.Context, payload []byte) ([]by
 	return restored, err
 }
 
-type grokResponsesClientToolStreamBody struct {
+type responsesClientToolStreamBody struct {
 	*io.PipeReader
 	source io.Closer
 }
 
-func (b *grokResponsesClientToolStreamBody) Close() error {
+func (b *responsesClientToolStreamBody) Close() error {
 	readerErr := b.PipeReader.Close()
 	sourceErr := b.source.Close()
 	if readerErr != nil {
@@ -111,8 +111,8 @@ func newResponsesClientToolStreamBody(
 	maxLineSize int,
 ) io.ReadCloser {
 	reader, writer := io.Pipe()
-	body := &grokResponsesClientToolStreamBody{PipeReader: reader, source: source}
-	go transformGrokResponsesClientToolStream(source, writer, mapping, maxLineSize)
+	body := &responsesClientToolStreamBody{PipeReader: reader, source: source}
+	go transformResponsesClientToolStream(source, writer, mapping, maxLineSize)
 	return body
 }
 
@@ -124,7 +124,7 @@ func newGrokResponsesClientToolStreamBody(
 	return newResponsesClientToolStreamBody(source, mapping, maxLineSize)
 }
 
-func transformGrokResponsesClientToolStream(
+func transformResponsesClientToolStream(
 	source io.ReadCloser,
 	destination *io.PipeWriter,
 	mapping apicompat.ResponsesClientToolMapping,
@@ -208,7 +208,7 @@ func transformGrokResponsesClientToolStream(
 				payloads, _, err = restorer.RestoreEvent(payload)
 				if err != nil {
 					_ = buffered.Flush()
-					_ = destination.CloseWithError(fmt.Errorf("restore Grok Responses client tool event: %w", err))
+					_ = destination.CloseWithError(fmt.Errorf("restore Responses client tool event: %w", err))
 					return
 				}
 			}
diff --git a/backend/internal/service/openai_ws_http_bridge_test.go b/backend/internal/service/openai_ws_http_bridge_test.go
index 887d495202e7..54786723166b 100644
--- a/backend/internal/service/openai_ws_http_bridge_test.go
+++ b/backend/internal/service/openai_ws_http_bridge_test.go
@@ -116,6 +116,60 @@ func TestProxyOpenAIWSHTTPBridgeTurnAPIKeyAdaptsClientTools(t *testing.T) {
 	require.Equal(t, "pwd", gjson.GetBytes(result.wsReplayInput[0], "input").String())
 }
 
+func TestProxyOpenAIWSHTTPBridgeTurnAPIKeyRestoresClientToolsInResponseDone(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	sse := strings.Join([]string{
+		`data: {"type":"response.output_item.added","sequence_number":0,"output_index":0,"item":{"type":"function_call","id":"item_exec","call_id":"call_exec","name":"exec","status":"in_progress"}}`,
+		``,
+		`data: {"type":"response.function_call_arguments.done","sequence_number":1,"output_index":0,"item_id":"item_exec","call_id":"call_exec","name":"exec","arguments":"{\"input\":\"pwd\"}"}`,
+		``,
+		`data: {"type":"response.done","sequence_number":2,"response":{"id":"resp_tools","status":"completed","output":[{"type":"function_call","id":"item_exec","call_id":"call_exec","name":"exec","arguments":"{\"input\":\"pwd\"}","status":"completed"}],"usage":{"input_tokens":1,"output_tokens":1}}}`,
+		``,
+	}, "\n")
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}},
+		Body:       io.NopCloser(strings.NewReader(sse)),
+	}}
+	svc := &OpenAIGatewayService{
+		cfg:          &config.Config{Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize}},
+		httpUpstream: upstream,
+	}
+	account := &Account{ID: 5764, Platform: PlatformOpenAI, Type: AccountTypeAPIKey, Concurrency: 1}
+	payload := []byte(`{
+		"type":"response.create","model":"gpt-5","stream":true,
+		"tools":[{"type":"custom","name":"exec","description":"Run a command"}],
+		"input":"run pwd"
+	}`)
+	recorder := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(recorder)
+	c.Request = httptest.NewRequest(http.MethodGet, "/v1/responses", nil)
+	var events [][]byte
+
+	result, err := svc.proxyOpenAIWSHTTPBridgeTurn(
+		context.Background(), c, account, "test-token", payload, len(payload),
+		"gpt-5", "", "", "", "", 1,
+		func(message []byte) error {
+			events = append(events, append([]byte(nil), message...))
+			return nil
+		},
+	)
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.Len(t, events, 4)
+	terminal := events[len(events)-1]
+	require.Equal(t, "response.done", gjson.GetBytes(terminal, "type").String())
+	require.Equal(t, int64(3), gjson.GetBytes(terminal, "sequence_number").Int())
+	require.Equal(t, "custom_tool_call", gjson.GetBytes(terminal, "response.output.0.type").String())
+	require.Equal(t, "pwd", gjson.GetBytes(terminal, "response.output.0.input").String())
+	require.False(t, gjson.GetBytes(terminal, "response.output.0.arguments").Exists())
+	require.True(t, result.wsReplayInputExists)
+	require.Len(t, result.wsReplayInput, 1)
+	require.Equal(t, "custom_tool_call", gjson.GetBytes(result.wsReplayInput[0], "type").String())
+}
+
 func TestOpenAIWSHTTPBridgeDecisionKeepsSmallFramesOnWS(t *testing.T) {
 	svc := &OpenAIGatewayService{
 		cfg: &config.Config{

From 8f6f459835779efa2e3308b90addfa20e81a4a66 Mon Sep 17 00:00:00 2001
From: shaw <dev@sub2api.org>
Date: Tue, 18 Aug 2026 17:12:01 +0800
Subject: [PATCH 64/68] fix(channels): support kimi/zhipu/deepseek platforms in
 channel pricing

- ChannelsView platformOrder now includes the three CN provider
  platforms so channel pricing can be configured for them; composite
  group expansion/attachment stays limited to the five main platforms
  (matches backend isConcreteRequestPlatform and composite-routes
  target_platform validation)
- SyncPricingModels maps kimi->moonshot, zhipu->zhipu,
  deepseek->deepseek; also fixes gemini mapping to "google" which
  matched zero catalog entries (provider key is "gemini")
- Add CN platform colors to channel pricing model tag classes
---
 backend/internal/handler/admin/channel_handler.go  |  5 ++++-
 .../internal/handler/admin/channel_handler_test.go |  2 +-
 frontend/src/components/admin/channel/types.ts     |  6 ++++++
 frontend/src/views/admin/ChannelsView.vue          | 14 ++++++++++----
 4 files changed, 21 insertions(+), 6 deletions(-)

diff --git a/backend/internal/handler/admin/channel_handler.go b/backend/internal/handler/admin/channel_handler.go
index 90c1a69b6d07..dfce024fa357 100644
--- a/backend/internal/handler/admin/channel_handler.go
+++ b/backend/internal/handler/admin/channel_handler.go
@@ -568,9 +568,12 @@ func (h *ChannelHandler) GetModelDefaultPricing(c *gin.Context) {
 var platformToLiteLLMProvider = map[string]string{
 	service.PlatformAnthropic:   "anthropic",
 	service.PlatformOpenAI:      "openai",
-	service.PlatformGemini:      "google",
+	service.PlatformGemini:      "gemini",
 	service.PlatformAntigravity: "anthropic",
 	service.PlatformGrok:        "xai",
+	service.PlatformKimi:        "moonshot",
+	service.PlatformZhipu:       "zhipu",
+	service.PlatformDeepseek:    "deepseek",
 }
 
 // SyncPricingModels 返回 LiteLLM 定价目录中指定平台的最新模型列表
diff --git a/backend/internal/handler/admin/channel_handler_test.go b/backend/internal/handler/admin/channel_handler_test.go
index 7f9c050fa524..6c5ddedc92e7 100644
--- a/backend/internal/handler/admin/channel_handler_test.go
+++ b/backend/internal/handler/admin/channel_handler_test.go
@@ -502,7 +502,7 @@ func TestSyncPricingModels_ValidPlatform_EmptyService(t *testing.T) {
 	svc := service.NewPricingService(nil, nil)
 	router := setupSyncPricingModelsRouter(svc)
 
-	for _, platform := range []string{"anthropic", "openai", "gemini", "antigravity"} {
+	for _, platform := range []string{"anthropic", "openai", "gemini", "antigravity", "grok", "kimi", "zhipu", "deepseek"} {
 		req := httptest.NewRequest(http.MethodGet, "/channels/pricing/sync-models?platform="+platform, nil)
 		w := httptest.NewRecorder()
 		router.ServeHTTP(w, req)
diff --git a/frontend/src/components/admin/channel/types.ts b/frontend/src/components/admin/channel/types.ts
index 8809eff072c0..322a407c0c2c 100644
--- a/frontend/src/components/admin/channel/types.ts
+++ b/frontend/src/components/admin/channel/types.ts
@@ -368,6 +368,9 @@ export function getPlatformTagClass(platform: string): string {
     case 'gemini': return 'bg-blue-100 text-blue-700 dark:bg-blue-900/30 dark:text-blue-400'
     case 'antigravity': return 'bg-purple-100 text-purple-700 dark:bg-purple-900/30 dark:text-purple-400'
     case 'grok': return 'bg-slate-100 text-slate-700 dark:bg-slate-800 dark:text-slate-300'
+    case 'kimi': return 'bg-pink-100 text-pink-700 dark:bg-pink-900/30 dark:text-pink-400'
+    case 'zhipu': return 'bg-indigo-100 text-indigo-700 dark:bg-indigo-900/30 dark:text-indigo-400'
+    case 'deepseek': return 'bg-teal-100 text-teal-700 dark:bg-teal-900/30 dark:text-teal-400'
     default: return 'bg-gray-100 text-gray-700 dark:bg-gray-900/30 dark:text-gray-400'
   }
 }
@@ -380,6 +383,9 @@ export function getPlatformTextClass(platform: string): string {
     case 'gemini': return 'text-blue-700 dark:text-blue-400'
     case 'antigravity': return 'text-purple-700 dark:text-purple-400'
     case 'grok': return 'text-slate-700 dark:text-slate-300'
+    case 'kimi': return 'text-pink-700 dark:text-pink-400'
+    case 'zhipu': return 'text-indigo-700 dark:text-indigo-400'
+    case 'deepseek': return 'text-teal-700 dark:text-teal-400'
     default: return ''
   }
 }
diff --git a/frontend/src/views/admin/ChannelsView.vue b/frontend/src/views/admin/ChannelsView.vue
index d431cb129337..e3a7ca826229 100644
--- a/frontend/src/views/admin/ChannelsView.vue
+++ b/frontend/src/views/admin/ChannelsView.vue
@@ -762,7 +762,10 @@ const form = reactive({
 let abortController: AbortController | null = null
 
 // ── Platform config ──
-const platformOrder: GroupPlatform[] = ['anthropic', 'openai', 'gemini', 'antigravity', 'grok']
+const platformOrder: GroupPlatform[] = ['anthropic', 'openai', 'gemini', 'antigravity', 'grok', 'kimi', 'zhipu', 'deepseek']
+// composite 分组仅覆盖主平台（与后端 isConcreteRequestPlatform / composite-routes target_platform 一致），
+// 不含国产供应商平台。
+const compositePlatforms: GroupPlatform[] = ['anthropic', 'openai', 'gemini', 'antigravity', 'grok']
 
 // ── Helpers ──
 function formatDate(value: string): string {
@@ -801,7 +804,9 @@ function togglePlatform(platform: GroupPlatform) {
 }
 
 function getGroupsForPlatform(platform: GroupPlatform): AdminGroup[] {
-  return allGroups.value.filter(g => g.platform === platform || g.platform === 'composite')
+  return allGroups.value.filter(
+    g => g.platform === platform || (g.platform === 'composite' && compositePlatforms.includes(platform))
+  )
 }
 
 // ── Group helpers ──
@@ -1183,7 +1188,7 @@ function apiToForm(channel: Channel): PlatformSection[] {
   for (const gid of channel.group_ids || []) {
     const p = groupPlatformMap.get(gid)
     if (p === 'composite') {
-      platformOrder.forEach(platform => activePlatforms.add(platform))
+      compositePlatforms.forEach(platform => activePlatforms.add(platform))
     } else if (p) {
       activePlatforms.add(p)
     }
@@ -1202,7 +1207,8 @@ function apiToForm(channel: Channel): PlatformSection[] {
 
     const groupIds = (channel.group_ids || []).filter(gid => {
       const groupPlatform = groupPlatformMap.get(gid)
-      return groupPlatform === platform || groupPlatform === 'composite'
+      return groupPlatform === platform ||
+        (groupPlatform === 'composite' && compositePlatforms.includes(platform))
     })
     const mapping = (channel.model_mapping || {})[platform] || {}
     const pricing = (channel.model_pricing || [])

From 22df600d090479ea42070eb79c42c520ea9e1eab Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 09:14:20 +0000
Subject: [PATCH 65/68] =?UTF-8?q?fix(channel-monitor):=20=E9=85=8D?=
 =?UTF-8?q?=E9=A2=9D=E5=BF=AB=E7=85=A7=E8=AF=86=E5=88=AB=E5=80=BC=E9=80=9A?=
 =?UTF-8?q?=E9=81=93=E5=A4=B1=E8=B4=A5=E5=B9=B6=E5=8A=A0=2060s=20=E8=B4=9F?=
 =?UTF-8?q?=E7=BC=93=E5=AD=98=E4=B8=8E=20singleflight?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- fetchUsage 显式识别 UsageInfo.Error/ErrorCode/NeedsReauth/IsBanned/IsForbidden
  （antigravity/grok 等平台失败不走 Go error 通道，此前会被误判为 operational）
- 凭据失效语义（401/403）标记 CredentialInvalid → failed 状态；
  grok quota_unknown 已知未知态豁免不判失败
- 失败快照进 60s 负缓存，避免故障/凭据失效期间以最小 15s 间隔打上游
- 同账号并发抓取由 singleflight 合并（脱离调用方 ctx，45s 总超时兜底）
---
 .../internal/service/channel_monitor_const.go |   6 +
 .../service/channel_monitor_quota_fetcher.go  |  80 +++++++++-
 .../channel_monitor_quota_fetcher_test.go     | 144 ++++++++++++++++--
 3 files changed, 213 insertions(+), 17 deletions(-)

diff --git a/backend/internal/service/channel_monitor_const.go b/backend/internal/service/channel_monitor_const.go
index 6427c8914de2..bf7676ed5068 100644
--- a/backend/internal/service/channel_monitor_const.go
+++ b/backend/internal/service/channel_monitor_const.go
@@ -85,6 +85,12 @@ const (
 	// monitorQuotaFetchCacheTTL 配额快照缓存时长。多个监控可能关联同一账号，
 	// 而 interval 最小 15s 且国产配额服务无缓存，TTL 防止打爆上游配额端点。
 	monitorQuotaFetchCacheTTL = 5 * time.Minute
+	// monitorQuotaErrorCacheTTL 失败快照的负缓存时长：失败也短缓存，避免
+	// 故障/凭据失效期间每次调度（最小 15s）都带真实凭据打上游；到期自动重试。
+	monitorQuotaErrorCacheTTL = 60 * time.Second
+	// monitorQuotaFetchTimeout singleflight 内单次配额抓取的总超时
+	// （脱离调用方 ctx，防止某个监控的取消波及共享同一账号的其他监控）。
+	monitorQuotaFetchTimeout = 45 * time.Second
 	// monitorQuotaDegradedUsedPercent 任一用量窗口使用率超过该阈值时，
 	// 配额检查状态记为 degraded（对齐账号页展示阈值）。
 	monitorQuotaDegradedUsedPercent = 90.0
diff --git a/backend/internal/service/channel_monitor_quota_fetcher.go b/backend/internal/service/channel_monitor_quota_fetcher.go
index 9fa990337f0a..a4fa992c3069 100644
--- a/backend/internal/service/channel_monitor_quota_fetcher.go
+++ b/backend/internal/service/channel_monitor_quota_fetcher.go
@@ -12,6 +12,7 @@ import (
 
 	"github.com/Wei-Shaw/sub2api/internal/domain"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/xai"
+	"golang.org/x/sync/singleflight"
 )
 
 // 渠道监控「配额模式」的配额抓取器。
@@ -26,7 +27,9 @@ import (
 // 由 deriveQuotaCheckResult 推导为 failed/error 状态。
 //
 // 多个监控可能关联同一账号，而 interval 最小 15s 且国产配额服务自身无缓存，
-// 所以成功快照统一带 monitorQuotaFetchCacheTTL 缓存，防止打爆上游配额端点。
+// 所以快照统一带 TTL 缓存（成功 monitorQuotaFetchCacheTTL、失败
+// monitorQuotaErrorCacheTTL 负缓存），防止打爆上游配额端点；同账号的并发
+// 抓取由 singleflight 合并为一次上游查询。
 
 // monitorUsageSource 海外平台账号用量查询（AccountUsageService 天然满足）。
 type monitorUsageSource interface {
@@ -48,15 +51,17 @@ type monitorAccountSource interface {
 	GetByID(ctx context.Context, id int64) (*Account, error)
 }
 
-// ChannelMonitorQuotaFetcher 配额抓取器（带成功快照 TTL 缓存）。
+// ChannelMonitorQuotaFetcher 配额抓取器（成功/失败快照均带 TTL 缓存，
+// 同账号并发抓取由 singleflight 合并）。
 type ChannelMonitorQuotaFetcher struct {
 	usage     monitorUsageSource
 	cnQuota   monitorCNQuotaSource
 	cnBalance monitorCNBalanceSource
 	accounts  monitorAccountSource
 
-	mu    sync.Mutex
-	cache map[int64]monitorQuotaCacheEntry
+	mu     sync.Mutex
+	cache  map[int64]monitorQuotaCacheEntry
+	flight singleflight.Group
 }
 
 type monitorQuotaCacheEntry struct {
@@ -111,11 +116,31 @@ func (f *ChannelMonitorQuotaFetcher) Fetch(ctx context.Context, accountID int64)
 		return cached
 	}
 
-	snapshot := f.fetchUncached(ctx, accountID, now)
-	if snapshot.Success {
-		f.storeSnapshot(accountID, snapshot, now.Add(monitorQuotaFetchCacheTTL))
+	// singleflight 合并同账号并发抓取；脱离调用方 ctx（仿 CN 配额服务），
+	// 避免某个监控的取消波及共享同一账号的其他监控。
+	key := "monitor-quota:" + strconv.FormatInt(accountID, 10)
+	ch := f.flight.DoChan(key, func() (any, error) {
+		fetchCtx, cancel := context.WithTimeout(context.Background(), monitorQuotaFetchTimeout)
+		defer cancel()
+		snapshot := f.fetchUncached(fetchCtx, accountID, time.Now())
+		// 失败也进短 TTL 负缓存：凭据失效/故障期间不必每次调度都打上游。
+		ttl := monitorQuotaFetchCacheTTL
+		if !snapshot.Success {
+			ttl = monitorQuotaErrorCacheTTL
+		}
+		f.storeSnapshot(accountID, snapshot, time.Now().Add(ttl))
+		return snapshot, nil
+	})
+	select {
+	case <-ctx.Done():
+		return quotaErrorSnapshot("usage", "context canceled", now)
+	case res := <-ch:
+		snapshot, ok := res.Val.(*domain.MonitorQuotaSnapshot)
+		if res.Err != nil || !ok || snapshot == nil {
+			return quotaErrorSnapshot("usage", "quota fetch failed", now)
+		}
+		return snapshot
 	}
-	return snapshot
 }
 
 func (f *ChannelMonitorQuotaFetcher) cachedSnapshot(accountID int64, now time.Time) (*domain.MonitorQuotaSnapshot, bool) {
@@ -175,6 +200,20 @@ func (f *ChannelMonitorQuotaFetcher) fetchUsage(ctx context.Context, accountID i
 			FetchedAt:         now,
 		}
 	}
+	if usage == nil {
+		return quotaErrorSnapshot("usage", "usage service returned no data", now)
+	}
+	// openai/gemini/antigravity/grok 的失败多走「值通道」（err==nil 但错误
+	// 降级在 UsageInfo 字段里），必须显式识别，否则会被误判为 operational。
+	if failed, credInvalid, msg := usageFailureInfo(usage); failed {
+		return &domain.MonitorQuotaSnapshot{
+			Source:            "usage",
+			Success:           false,
+			CredentialInvalid: credInvalid,
+			Error:             truncateMessage(sanitizeErrorMessage(msg)),
+			FetchedAt:         now,
+		}
+	}
 	snapshot := &domain.MonitorQuotaSnapshot{
 		Source:    "usage",
 		Success:   true,
@@ -387,6 +426,31 @@ func isCredentialErrorMessage(msg string) bool {
 		strings.Contains(msg, "authentication")
 }
 
+// usageFailureInfo 识别 GetUsage 经「值通道」返回的失败：antigravity/grok
+// 等平台 err==nil 但把错误降级在 UsageInfo 字段里（Error/ErrorCode/状态标记）。
+// 返回 failed=false 表示可用；credentialInvalid 表示凭据失效（401/403 语义，
+// 推导为 failed 状态）；msg 为失败摘要。
+//
+// grok 的 ErrorCode=quota_unknown 是「尚未观测到计费快照/限流头」的已知未知态，
+// 不是失败（严格按 ErrorCode 判会把健康 grok 账号永久判 error），显式豁免。
+func usageFailureInfo(usage *UsageInfo) (failed, credentialInvalid bool, msg string) {
+	if usage == nil {
+		return false, false, ""
+	}
+	if usage.ErrorCode == "quota_unknown" {
+		return false, false, ""
+	}
+	failed = usage.Error != "" || usage.NeedsReauth || usage.IsBanned ||
+		usage.IsForbidden || usage.ErrorCode != ""
+	if !failed {
+		return false, false, ""
+	}
+	credentialInvalid = usage.NeedsReauth || usage.IsBanned || usage.IsForbidden ||
+		usage.ErrorCode == errorCodeUnauthenticated || usage.ErrorCode == errorCodeForbidden
+	msg = firstNonEmpty(usage.Error, usage.ForbiddenReason, usage.ErrorCode, "usage fetch failed")
+	return failed, credentialInvalid, msg
+}
+
 // deriveQuotaCheckResult 把配额快照推导为检测状态（复用既有 status 枚举，
 // 时间线/可用率机制自动生效）：
 //   - 查询成功且无告警        → operational
diff --git a/backend/internal/service/channel_monitor_quota_fetcher_test.go b/backend/internal/service/channel_monitor_quota_fetcher_test.go
index c9337c766d93..f2e6ff998cf1 100644
--- a/backend/internal/service/channel_monitor_quota_fetcher_test.go
+++ b/backend/internal/service/channel_monitor_quota_fetcher_test.go
@@ -5,6 +5,7 @@ package service
 import (
 	"context"
 	"errors"
+	"sync"
 	"testing"
 	"time"
 
@@ -16,18 +17,33 @@ import (
 // --- fetcher 依赖 stub ---
 
 type stubMonitorUsageSource struct {
-	usage   *UsageInfo
-	err     error
+	usage *UsageInfo
+	err   error
+	// block 非 nil 时 GetUsage 阻塞在该 channel 上，用于并发/singleflight 测试。
+	block chan struct{}
+
+	mu      sync.Mutex
 	calls   int
 	lastCtx context.Context
 }
 
 func (s *stubMonitorUsageSource) GetUsage(ctx context.Context, accountID int64, force ...bool) (*UsageInfo, error) {
+	s.mu.Lock()
 	s.calls++
 	s.lastCtx = ctx
+	s.mu.Unlock()
+	if s.block != nil {
+		<-s.block
+	}
 	return s.usage, s.err
 }
 
+func (s *stubMonitorUsageSource) getCalls() int {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.calls
+}
+
 type stubMonitorCNQuotaSource struct {
 	result *CNProviderQuotaProbeResult
 	err    error
@@ -110,7 +126,7 @@ func TestQuotaFetcher_OverseasAccountUsesUsageService(t *testing.T) {
 	require.NotEmpty(t, fiveHour.ResetAt)
 
 	require.Equal(t, "7d", snapshot.Tiers[1].Window)
-	require.Equal(t, 1, usage.calls)
+	require.Equal(t, 1, usage.getCalls())
 	require.Equal(t, 0, cnQuota.calls)
 }
 
@@ -182,7 +198,7 @@ func TestQuotaFetcher_AccountMissingYieldsLinkedAccountSnapshot(t *testing.T) {
 
 	require.False(t, snapshot.Success)
 	require.Equal(t, "linked account not found", snapshot.Error)
-	require.Equal(t, 0, usage.calls) // 未走到数据源
+	require.Equal(t, 0, usage.getCalls()) // 未走到数据源
 }
 
 func TestQuotaFetcher_UsageAuthErrorMarksCredentialInvalid(t *testing.T) {
@@ -197,6 +213,71 @@ func TestQuotaFetcher_UsageAuthErrorMarksCredentialInvalid(t *testing.T) {
 	require.Contains(t, snapshot.Error, "401")
 }
 
+// 值通道失败：antigravity/grok 等平台 err==nil 但错误降级在 UsageInfo 字段里，
+// 必须识别为失败快照，否则会被误判为 operational。
+func TestQuotaFetcher_UsageValueChannelFailureYieldsFailureSnapshot(t *testing.T) {
+	fetcher, usage, _, _, accounts := newQuotaFetcherTestSetup(t)
+
+	// 凭据失效（401 语义）→ failed。
+	accounts.accounts[3] = &Account{ID: 3, Platform: domain.PlatformAnthropic}
+	usage.usage = &UsageInfo{Error: "usage API error: HTTP 401", ErrorCode: errorCodeUnauthenticated, NeedsReauth: true}
+	snapshot := fetcher.Fetch(context.Background(), 3)
+	require.False(t, snapshot.Success)
+	require.True(t, snapshot.CredentialInvalid)
+	require.Contains(t, snapshot.Error, "401")
+	require.Equal(t, MonitorStatusFailed, deriveQuotaCheckResult(snapshot, "quota", time.Now()).Status)
+
+	// 限流等非凭据失败 → error（而非 operational）。
+	accounts.accounts[13] = &Account{ID: 13, Platform: domain.PlatformAnthropic}
+	usage.usage = &UsageInfo{Error: "usage API error: HTTP 429", ErrorCode: errorCodeRateLimited}
+	snapshot = fetcher.Fetch(context.Background(), 13)
+	require.False(t, snapshot.Success)
+	require.False(t, snapshot.CredentialInvalid)
+	require.Contains(t, snapshot.Error, "429")
+	require.Equal(t, MonitorStatusError, deriveQuotaCheckResult(snapshot, "quota", time.Now()).Status)
+
+	// grok 已知未知态（尚未观测到计费/限流头）不算失败。
+	accounts.accounts[14] = &Account{ID: 14, Platform: domain.PlatformGrok}
+	usage.usage = &UsageInfo{ErrorCode: "quota_unknown", Error: "Grok quota is unknown until billing is probed"}
+	snapshot = fetcher.Fetch(context.Background(), 14)
+	require.True(t, snapshot.Success)
+	require.Empty(t, snapshot.Error)
+	require.Empty(t, snapshot.Tiers)
+	require.Equal(t, MonitorStatusOperational, deriveQuotaCheckResult(snapshot, "quota", time.Now()).Status)
+}
+
+func TestUsageFailureInfo_ClassificationMatrix(t *testing.T) {
+	cases := []struct {
+		name              string
+		usage             *UsageInfo
+		failed            bool
+		credentialInvalid bool
+		msg               string
+	}{
+		{name: "nil usage", usage: nil},
+		{name: "healthy empty", usage: &UsageInfo{}},
+		{name: "error text only", usage: &UsageInfo{Error: "boom"}, failed: true, msg: "boom"},
+		{name: "needs reauth", usage: &UsageInfo{NeedsReauth: true}, failed: true, credentialInvalid: true, msg: "usage fetch failed"},
+		{name: "banned", usage: &UsageInfo{IsBanned: true}, failed: true, credentialInvalid: true, msg: "usage fetch failed"},
+		{name: "forbidden with reason", usage: &UsageInfo{IsForbidden: true, ForbiddenReason: "usage limited"}, failed: true, credentialInvalid: true, msg: "usage limited"},
+		{name: "error code unauthenticated", usage: &UsageInfo{ErrorCode: errorCodeUnauthenticated}, failed: true, credentialInvalid: true, msg: errorCodeUnauthenticated},
+		{name: "error code forbidden", usage: &UsageInfo{ErrorCode: errorCodeForbidden}, failed: true, credentialInvalid: true, msg: errorCodeForbidden},
+		{name: "error code rate limited", usage: &UsageInfo{ErrorCode: errorCodeRateLimited}, failed: true, msg: errorCodeRateLimited},
+		{name: "error code network error", usage: &UsageInfo{ErrorCode: errorCodeNetworkError}, failed: true, msg: errorCodeNetworkError},
+		{name: "grok quota unknown exempted", usage: &UsageInfo{ErrorCode: "quota_unknown", Error: "Grok quota is unknown until billing is probed"}},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			failed, credentialInvalid, msg := usageFailureInfo(tc.usage)
+			require.Equal(t, tc.failed, failed)
+			require.Equal(t, tc.credentialInvalid, credentialInvalid)
+			if tc.msg != "" {
+				require.Equal(t, tc.msg, msg)
+			}
+		})
+	}
+}
+
 func TestQuotaFetcher_CNQuotaCredentialInvalidFlagPropagates(t *testing.T) {
 	fetcher, _, cnQuota, _, accounts := newQuotaFetcherTestSetup(t)
 	accounts.accounts[5] = &Account{
@@ -251,7 +332,7 @@ func TestQuotaFetcher_CachesSuccessSnapshotPerAccount(t *testing.T) {
 		snapshot := fetcher.Fetch(context.Background(), 8)
 		require.True(t, snapshot.Success)
 	}
-	require.Equal(t, 1, usage.calls, "success snapshots should be served from cache")
+	require.Equal(t, 1, usage.getCalls(), "success snapshots should be served from cache")
 
 	// 缓存过期后重新拉取。
 	fetcher.mu.Lock()
@@ -261,18 +342,63 @@ func TestQuotaFetcher_CachesSuccessSnapshotPerAccount(t *testing.T) {
 	fetcher.mu.Unlock()
 
 	_ = fetcher.Fetch(context.Background(), 8)
-	require.Equal(t, 2, usage.calls)
+	require.Equal(t, 2, usage.getCalls())
 }
 
-func TestQuotaFetcher_DoesNotCacheFailures(t *testing.T) {
+func TestQuotaFetcher_CachesFailureSnapshotWithShortTTL(t *testing.T) {
 	fetcher, usage, _, _, accounts := newQuotaFetcherTestSetup(t)
 	accounts.accounts[4] = &Account{ID: 4, Platform: domain.PlatformOpenAI}
 	usage.err = errors.New("boom")
 
+	for i := 0; i < 2; i++ {
+		snapshot := fetcher.Fetch(context.Background(), 4)
+		require.False(t, snapshot.Success)
+	}
+	require.Equal(t, 1, usage.getCalls(), "failure snapshots should be served from the short negative cache")
+
+	// 失败快照的 TTL 是负缓存时长（而非成功 TTL）。
+	fetcher.mu.Lock()
+	entry := fetcher.cache[4]
+	require.WithinDuration(t, entry.snapshot.FetchedAt.Add(monitorQuotaErrorCacheTTL), entry.expiry, time.Second)
+	entry.expiry = time.Now().Add(-time.Second)
+	fetcher.cache[4] = entry
+	fetcher.mu.Unlock()
+
 	_ = fetcher.Fetch(context.Background(), 4)
-	_ = fetcher.Fetch(context.Background(), 4)
+	require.Equal(t, 2, usage.getCalls(), "expired negative cache should refetch")
+}
+
+func TestQuotaFetcher_ConcurrentFetchesShareSingleFlight(t *testing.T) {
+	fetcher, usage, _, _, accounts := newQuotaFetcherTestSetup(t)
+	accounts.accounts[12] = &Account{ID: 12, Platform: domain.PlatformOpenAI}
+	usage.usage = &UsageInfo{FiveHour: &UsageProgress{Utilization: 10}}
+	usage.block = make(chan struct{})
+
+	var wg sync.WaitGroup
+	snapshots := make([]*domain.MonitorQuotaSnapshot, 5)
+	for i := 0; i < 5; i++ {
+		wg.Add(1)
+		go func(idx int) {
+			defer wg.Done()
+			snapshots[idx] = fetcher.Fetch(context.Background(), 12)
+		}(i)
+	}
+
+	// 上游被 block 卡住时，5 个并发 Fetch 应只产生 1 次真实查询。
+	require.Eventually(t, func() bool { return usage.getCalls() == 1 },
+		5*time.Second, 10*time.Millisecond)
+	close(usage.block)
+	wg.Wait()
+
+	for _, snapshot := range snapshots {
+		require.NotNil(t, snapshot)
+		require.True(t, snapshot.Success)
+	}
+	require.Equal(t, 1, usage.getCalls())
 
-	require.Equal(t, 2, usage.calls, "failed snapshots must not be cached")
+	// 成功快照已缓存：再取一次仍不打上游。
+	_ = fetcher.Fetch(context.Background(), 12)
+	require.Equal(t, 1, usage.getCalls())
 }
 
 // --- UsageInfo → tiers 归一 ---

From 03c3f3b6fcf51e2c7c1dcc7a6c2fdbd4931bf67d Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 09:14:31 +0000
Subject: [PATCH 66/68] =?UTF-8?q?feat(ui):=20Select=20=E7=BB=84=E4=BB=B6?=
 =?UTF-8?q?=E6=94=AF=E6=8C=81=E5=8F=AF=E9=80=89=E8=BF=9C=E7=A8=8B=E6=90=9C?=
 =?UTF-8?q?=E7=B4=A2=EF=BC=88remote/loading=20props=20+=20search=20?=
 =?UTF-8?q?=E4=BA=8B=E4=BB=B6=EF=BC=89?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- remote 模式始终显示搜索框且不做本地过滤，输入 300ms 防抖后 emit('search')
- 关闭/卸载时取消排队中的防抖定时器，避免尾随空 query
- loading 时空态显示加载文案；未设 remote 时行为不变（回归用例保障）
---
 frontend/src/components/common/Select.vue     |  39 ++++++-
 .../common/__tests__/Select.spec.ts           | 104 ++++++++++++++++++
 2 files changed, 140 insertions(+), 3 deletions(-)

diff --git a/frontend/src/components/common/Select.vue b/frontend/src/components/common/Select.vue
index 606e6300bf9f..a534f5c246a6 100644
--- a/frontend/src/components/common/Select.vue
+++ b/frontend/src/components/common/Select.vue
@@ -111,7 +111,7 @@
 
             <!-- Empty state -->
             <div v-if="filteredOptions.length === 0" class="select-empty">
-              {{ emptyTextDisplay }}
+              {{ props.loading ? t('common.loading') : emptyTextDisplay }}
             </div>
           </div>
         </div>
@@ -154,11 +154,16 @@ interface Props {
   id?: string
   ariaLabel?: string
   ariaDescribedby?: string
+  /** 远程搜索模式：输入不在本地过滤 options，而是防抖后 emit('search', query)，由父组件请求数据更新 options */
+  remote?: boolean
+  /** 远程搜索模式下的加载态：options 为空时下拉显示 loading 文案 */
+  loading?: boolean
 }
 
 interface Emits {
   (e: 'update:modelValue', value: string | number | boolean | null): void
   (e: 'change', value: string | number | boolean | null, option: SelectOption | null): void
+  (e: 'search', query: string): void
 }
 
 const props = withDefaults(defineProps<Props>(), {
@@ -169,7 +174,9 @@ const props = withDefaults(defineProps<Props>(), {
   creatablePrefix: '',
   clearable: false,
   valueKey: 'value',
-  labelKey: 'label'
+  labelKey: 'label',
+  remote: false,
+  loading: false
 })
 
 const emit = defineEmits<Emits>()
@@ -192,7 +199,13 @@ const placeholderText = computed(() => props.placeholder ?? t('common.selectOpti
 const searchPlaceholderText = computed(() => props.searchPlaceholder ?? t('common.searchPlaceholder'))
 const emptyTextDisplay = computed(() => props.emptyText ?? t('common.noOptionsFound'))
 
+// 远程搜索的防抖间隔（对齐 OpenAIFastPolicyUserSelector 的 300ms 惯例）。
+const REMOTE_SEARCH_DEBOUNCE_MS = 300
+let remoteSearchTimer: ReturnType<typeof setTimeout> | null = null
+
 const isSearchable = computed(() => {
+  // 远程搜索模式始终显示搜索框（选项只是服务端结果的一页）。
+  if (props.remote) return true
   if (props.searchable === 'auto') return props.options.length > 5
   return props.searchable
 })
@@ -276,7 +289,8 @@ const hasValue = computed(
 
 const filteredOptions = computed(() => {
   let opts = props.options as any[]
-  if (isSearchable.value && searchQuery.value) {
+  // 远程搜索模式不在本地过滤（选项即服务端搜索结果的一页）。
+  if (isSearchable.value && searchQuery.value && !props.remote) {
     const query = searchQuery.value.toLowerCase()
     opts = opts.filter((opt) => {
       // Match label
@@ -377,11 +391,26 @@ watch(isOpen, (open) => {
   } else {
     searchQuery.value = ''
     focusedIndex.value = -1
+    // 关闭时取消仍在排队的远程搜索（避免关闭后尾随 emit 一次 search(''))。
+    if (remoteSearchTimer) {
+      clearTimeout(remoteSearchTimer)
+      remoteSearchTimer = null
+    }
     window.removeEventListener('scroll', updateTriggerRect, { capture: true })
     window.removeEventListener('resize', calculateDropdownPosition)
   }
 })
 
+// 远程搜索：输入防抖后交给父组件请求（!isOpen 抑制关闭重置 searchQuery 触发的空 query）。
+watch(searchQuery, (query) => {
+  if (!props.remote || !isOpen.value) return
+  if (remoteSearchTimer) clearTimeout(remoteSearchTimer)
+  remoteSearchTimer = setTimeout(() => {
+    remoteSearchTimer = null
+    emit('search', query.trim())
+  }, REMOTE_SEARCH_DEBOUNCE_MS)
+})
+
 const selectOption = (option: any) => {
   const value = getOptionValue(option) ?? null
   emit('update:modelValue', value)
@@ -467,6 +496,10 @@ onUnmounted(() => {
   document.removeEventListener('click', handleClickOutside)
   window.removeEventListener('scroll', updateTriggerRect, { capture: true })
   window.removeEventListener('resize', calculateDropdownPosition)
+  if (remoteSearchTimer) {
+    clearTimeout(remoteSearchTimer)
+    remoteSearchTimer = null
+  }
 })
 </script>
 
diff --git a/frontend/src/components/common/__tests__/Select.spec.ts b/frontend/src/components/common/__tests__/Select.spec.ts
index 0817cad1a9fc..6cd401015224 100644
--- a/frontend/src/components/common/__tests__/Select.spec.ts
+++ b/frontend/src/components/common/__tests__/Select.spec.ts
@@ -61,6 +61,7 @@ afterEach(() => {
   unmountWrapper = undefined
   document.body.innerHTML = ''
   setViewportWidth(originalInnerWidth)
+  vi.useRealTimers()
   vi.restoreAllMocks()
 })
 
@@ -113,3 +114,106 @@ describe('Select dropdown viewport constraints', () => {
     expect(dropdown?.style.maxWidth).toBe('0px')
   })
 })
+
+describe('Select remote search', () => {
+  const mountRemoteSelect = (props: Record<string, unknown> = {}) => {
+    const wrapper = mount(Select, {
+      props: {
+        modelValue: null,
+        remote: true,
+        options: [
+          { value: 'alpha', label: 'Alpha account' },
+          { value: 'beta', label: 'Beta account' },
+        ],
+        ...props,
+      },
+    })
+    unmountWrapper = () => wrapper.unmount()
+    return wrapper
+  }
+
+  const openDropdown = async () => {
+    const dropdown = document.body.querySelector<HTMLElement>('.select-dropdown-portal')
+    expect(dropdown).not.toBeNull()
+    return dropdown as HTMLElement
+  }
+
+  const typeSearchQuery = async (query: string) => {
+    const dropdown = await openDropdown()
+    const input = dropdown.querySelector<HTMLInputElement>('.select-search-input')
+    expect(input).not.toBeNull()
+    input!.value = query
+    input!.dispatchEvent(new Event('input'))
+    await nextTick()
+  }
+
+  it('emits debounced search events and skips local filtering in remote mode', async () => {
+    vi.useFakeTimers()
+    const wrapper = mountRemoteSelect()
+    await wrapper.get('button').trigger('click')
+    await nextTick()
+
+    await typeSearchQuery('zzz')
+
+    // 防抖窗口内不触发。
+    expect(wrapper.emitted('search')).toBeUndefined()
+    await vi.advanceTimersByTimeAsync(300)
+
+    expect(wrapper.emitted('search')).toEqual([['zzz']])
+    // 远程模式不做本地过滤：无命中的 query 下选项仍完整展示（由父组件更新 options）。
+    const dropdown = await openDropdown()
+    const labels = [...dropdown.querySelectorAll('.select-option-label')].map((el) => el.textContent)
+    expect(labels).toContain('Alpha account')
+    expect(labels).toContain('Beta account')
+  })
+
+  it('does not emit search when the dropdown closes and the query resets', async () => {
+    vi.useFakeTimers()
+    const wrapper = mountRemoteSelect()
+    await wrapper.get('button').trigger('click')
+    await nextTick()
+
+    await typeSearchQuery('hidden')
+
+    // 关闭下拉：排队中的防抖定时器应被取消，也不应因 query 重置而尾随 emit。
+    await wrapper.get('button').trigger('click')
+    await nextTick()
+    await vi.advanceTimersByTimeAsync(300)
+
+    expect(wrapper.emitted('search')).toBeUndefined()
+  })
+
+  it('shows the loading text instead of empty text while loading with no options', async () => {
+    const wrapper = mountRemoteSelect({ options: [], loading: true })
+    await wrapper.get('button').trigger('click')
+    await nextTick()
+
+    const dropdown = await openDropdown()
+    expect(dropdown.querySelector('.select-empty')?.textContent).toContain('common.loading')
+  })
+
+  it('keeps local filtering and emits nothing when remote is not set', async () => {
+    vi.useFakeTimers()
+    const wrapper = mount(Select, {
+      props: {
+        modelValue: null,
+        searchable: true,
+        options: [
+          { value: 'alpha', label: 'Alpha account' },
+          { value: 'beta', label: 'Beta account' },
+        ],
+      },
+    })
+    unmountWrapper = () => wrapper.unmount()
+    await wrapper.get('button').trigger('click')
+    await nextTick()
+
+    await typeSearchQuery('alpha')
+    await vi.advanceTimersByTimeAsync(300)
+
+    expect(wrapper.emitted('search')).toBeUndefined()
+    const dropdown = await openDropdown()
+    const labels = [...dropdown.querySelectorAll('.select-option-label')].map((el) => el.textContent)
+    expect(labels).toEqual(['Alpha account'])
+  })
+})

From 5cbd0c96a2eec8133cc504946767fcc7e83b9f1a Mon Sep 17 00:00:00 2001
From: Randark <randark@randark.site>
Date: Tue, 18 Aug 2026 09:14:44 +0000
Subject: [PATCH 67/68] =?UTF-8?q?fix(monitor-ui):=20=E5=85=B3=E8=81=94?=
 =?UTF-8?q?=E8=B4=A6=E5=8F=B7=E9=80=89=E6=8B=A9=E5=99=A8=E6=94=B9=E6=9C=8D?=
 =?UTF-8?q?=E5=8A=A1=E7=AB=AF=E6=90=9C=E7=B4=A2+=E5=9B=9E=E5=A1=AB?=
 =?UTF-8?q?=EF=BC=8COpenAI=20=E9=85=8D=E9=A2=9D=E6=A8=A1=E5=BC=8F=E5=8A=A0?=
 =?UTF-8?q?=E6=B6=88=E8=80=97=E6=8F=90=E7=A4=BA?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 选择器改为 /admin/accounts?platform=&search= 服务端搜索（50 条/页，
  300ms 防抖由 Select 承担），AbortController + 序列守卫防乱序覆盖
- 已绑定账号不在结果页时用 getById 回填为固定选项，删除「不在列表即
  静默清空绑定」；绑定仅在显式切换 provider 或账号确实失效时清空（带提示）
- OpenAI 平台配额模式显示 Codex 探测额度消耗提示；checkModeQuotaHint
  修正为「不发送探活请求」不再宣称「不产生 LLM 请求」
- grok spec 补齐缺失的 accounts mock；新增账号选择器 7 用例
  （加载/防抖搜索/回填保留/失效清空/提示门控/固定选项/切平台清空）
---
 .../admin/monitor/MonitorFormDialog.vue       | 134 ++++++--
 .../src/i18n/locales/en/admin/channels.ts     |   4 +-
 .../src/i18n/locales/zh/admin/channels.ts     |   4 +-
 .../__tests__/ChannelMonitorView.grok.spec.ts |  10 +-
 .../MonitorFormDialog.accountSelector.spec.ts | 311 ++++++++++++++++++
 5 files changed, 437 insertions(+), 26 deletions(-)
 create mode 100644 frontend/src/views/admin/__tests__/MonitorFormDialog.accountSelector.spec.ts

diff --git a/frontend/src/components/admin/monitor/MonitorFormDialog.vue b/frontend/src/components/admin/monitor/MonitorFormDialog.vue
index b67bcb775a63..ea73b929ed05 100644
--- a/frontend/src/components/admin/monitor/MonitorFormDialog.vue
+++ b/frontend/src/components/admin/monitor/MonitorFormDialog.vue
@@ -56,13 +56,24 @@
         <label class="input-label">
           {{ t('admin.channelMonitor.form.linkedAccount') }} <span class="text-red-500">*</span>
         </label>
-        <Select
-          v-model="accountSelectValue"
-          :options="accountOptions"
-          :placeholder="t('admin.channelMonitor.form.linkedAccountPlaceholder')"
-        />
+        <div data-testid="monitor-linked-account">
+          <Select
+            v-model="accountSelectValue"
+            :options="accountOptions"
+            :placeholder="t('admin.channelMonitor.form.linkedAccountPlaceholder')"
+            remote
+            :loading="accountsLoading"
+            @search="onAccountSearch"
+          />
+        </div>
         <p class="mt-1 text-xs text-gray-400">{{ t('admin.channelMonitor.form.linkedAccountHint') }}</p>
-        <p v-if="accountOptions.length === 0 && !accountsLoading" class="mt-1 text-xs text-amber-600 dark:text-amber-400">
+        <p v-if="form.provider === PROVIDER_OPENAI" class="mt-1 text-xs text-amber-600 dark:text-amber-400">
+          {{ t('admin.channelMonitor.form.openAIQuotaProbeHint') }}
+        </p>
+        <p v-if="accountHydrationFailed" class="mt-1 text-xs text-amber-600 dark:text-amber-400">
+          {{ t('admin.channelMonitor.form.linkedAccountMissing') }}
+        </p>
+        <p v-if="accountOptions.length === 0 && !accountsLoading && !accountSearchQuery" class="mt-1 text-xs text-amber-600 dark:text-amber-400">
           {{ t('admin.channelMonitor.form.linkedAccountEmpty') }}
         </p>
       </div>
@@ -524,13 +535,27 @@ interface LinkedAccount {
 
 const linkedAccounts = ref<LinkedAccount[]>([])
 const accountsLoading = ref(false)
-
-const accountOptions = computed(() =>
-  linkedAccounts.value.map((a) => ({
+// 当前搜索词（用于空态文案区分「平台无账号」与「搜索无命中」）。
+const accountSearchQuery = ref('')
+// 已绑定账号回填失败（getById 失败或平台失配）时提示用户重新选择。
+const accountHydrationFailed = ref(false)
+// 固定选项：已绑定/已选中但不在当前结果页里的账号，保证搜索后 label 仍可见。
+const pinnedAccount = ref<LinkedAccount | null>(null)
+let accountSearchSeq = 0
+let accountSearchAbort: AbortController | null = null
+const hydrationAttempted = new Set<number>()
+
+const accountOptions = computed(() => {
+  const opts = linkedAccounts.value.map((a) => ({
     value: String(a.id),
     label: `${a.name} (#${a.id})`,
-  })),
-)
+  }))
+  const pinned = pinnedAccount.value
+  if (pinned && !linkedAccounts.value.some((a) => a.id === pinned.id)) {
+    opts.unshift({ value: String(pinned.id), label: `${pinned.name} (#${pinned.id})` })
+  }
+  return opts
+})
 
 // Select 组件绑定 string，与 number | null 互转。
 const accountSelectValue = computed<string>({
@@ -538,37 +563,95 @@ const accountSelectValue = computed<string>({
   set: (raw: string) => {
     if (raw === '') {
       form.account_id = null
+      pinnedAccount.value = null
+      accountHydrationFailed.value = false
       return
     }
     const id = Number(raw)
-    if (Number.isFinite(id)) form.account_id = id
+    if (Number.isFinite(id)) {
+      form.account_id = id
+      pinnedAccount.value = linkedAccounts.value.find((a) => a.id === id) ?? pinnedAccount.value
+    }
   },
 })
 
-// 拉取当前 provider 平台的账号（大分页一次取齐，客户端不再过滤）。
+// 服务端搜索当前 provider 平台的账号（支持关键字，避免大分页截断取不齐）。
+// seq + abort 防止快速切换 provider / 连续输入时乱序响应覆盖新结果。
 // 失败不阻塞表单：下拉为空 + 空态提示。
-async function loadLinkedAccounts() {
+async function loadLinkedAccounts(search = '') {
   if (!usesQuotaMode.value || !props.show) return
+  accountSearchQuery.value = search
+  const seq = ++accountSearchSeq
+  accountSearchAbort?.abort()
+  const controller = new AbortController()
+  accountSearchAbort = controller
   accountsLoading.value = true
   try {
-    const res = await adminAPI.accounts.list(1, 200, { platform: form.provider })
+    const res = await adminAPI.accounts.list(
+      1,
+      50,
+      { platform: form.provider, ...(search ? { search } : {}) },
+      { signal: controller.signal },
+    )
+    if (seq !== accountSearchSeq) return
     linkedAccounts.value = (res.items || []).map((a) => ({ id: a.id, name: a.name }))
-    // 关联账号必须与 provider 同平台：平台切换后原选择失效，自动清掉。
-    if (form.account_id != null && !linkedAccounts.value.some((a) => a.id === form.account_id)) {
-      form.account_id = null
-    }
+    await ensureSelectedAccountHydrated()
   } catch (err: unknown) {
+    if (controller.signal.aborted) return
     console.warn('load linked accounts failed', err)
-    linkedAccounts.value = []
+    if (!search) linkedAccounts.value = []
   } finally {
-    accountsLoading.value = false
+    if (seq === accountSearchSeq) accountsLoading.value = false
   }
 }
 
+// 编辑已有 quota 监控时，已绑定账号可能不在搜索结果第一页：用 getById
+// 回填为固定选项，绑定不因分页截断而丢失。仅当账号确实无法加载或平台
+// 失配时才清空绑定（带可见提示），否则绑定只在用户显式切换 provider 时清空。
+async function ensureSelectedAccountHydrated() {
+  const id = form.account_id
+  if (id == null || !usesQuotaMode.value) return
+  if (linkedAccounts.value.some((a) => a.id === id) || pinnedAccount.value?.id === id) return
+  if (hydrationAttempted.has(id)) return
+  hydrationAttempted.add(id)
+  try {
+    const account = await adminAPI.accounts.getById(id)
+    if (form.account_id !== id) return
+    if (String(account.platform) !== form.provider) {
+      form.account_id = null
+      pinnedAccount.value = null
+      accountHydrationFailed.value = true
+      return
+    }
+    pinnedAccount.value = { id: account.id, name: account.name }
+  } catch {
+    if (form.account_id === id) {
+      form.account_id = null
+      pinnedAccount.value = null
+      accountHydrationFailed.value = true
+    }
+  }
+}
+
+function onAccountSearch(query: string) {
+  void loadLinkedAccounts(query)
+}
+
 watch(
   () => [props.show, form.provider, form.check_mode] as const,
-  ([show]) => {
-    if (!show) return
+  ([show, provider], prev) => {
+    const [prevShow, prevProvider] = prev ?? []
+    if (!show) {
+      accountSearchAbort?.abort()
+      return
+    }
+    // 弹窗重开 / provider 真正变化时重置回填状态（check_mode 变化不重置，
+    // 避免 probe↔quota 切换时无谓地重拉列表）。
+    if (show !== prevShow || provider !== prevProvider) {
+      hydrationAttempted.clear()
+      accountHydrationFailed.value = false
+      pinnedAccount.value = null
+    }
     void loadLinkedAccounts()
   },
   { immediate: true },
@@ -584,7 +667,10 @@ function selectProvider(provider: Provider) {
   const clearPrevDefaultEndpoint =
     !!PROVIDER_DEFAULT_ENDPOINTS[previousProvider] && form.endpoint === PROVIDER_DEFAULT_ENDPOINTS[previousProvider]
   form.provider = provider
+  // 关联账号与平台绑定：切换 provider 时显式清空（这是唯一主动清空的入口）。
   form.account_id = null
+  pinnedAccount.value = null
+  accountHydrationFailed.value = false
   // antigravity 仅配额模式：切到它时强制 quota（checkModeOptions 同步禁用其余项）。
   if (provider === PROVIDER_ANTIGRAVITY && form.check_mode !== CHECK_MODE_QUOTA) {
     form.check_mode = CHECK_MODE_QUOTA
@@ -628,6 +714,8 @@ function resetForm() {
   form.api_mode = API_MODE_CHAT_COMPLETIONS
   form.check_mode = CHECK_MODE_PROBE
   form.account_id = null
+  pinnedAccount.value = null
+  accountHydrationFailed.value = false
   form.endpoint = ''
   form.api_key = ''
   form.primary_model = ''
diff --git a/frontend/src/i18n/locales/en/admin/channels.ts b/frontend/src/i18n/locales/en/admin/channels.ts
index 0dc9e80b02f0..41793496f0b6 100644
--- a/frontend/src/i18n/locales/en/admin/channels.ts
+++ b/frontend/src/i18n/locales/en/admin/channels.ts
@@ -530,13 +530,15 @@ export default {
         checkModeProbe: 'Probe',
         checkModeProbeHint: 'Sends a lightweight LLM request to measure availability and latency',
         checkModeQuota: 'Quota',
-        checkModeQuotaHint: 'Only queries the linked account usage windows / balance, no LLM requests',
+        checkModeQuotaHint: 'Only queries the linked account usage windows / balance without probe requests',
         checkModeQuotaProbe: 'Probe + Quota',
         checkModeQuotaProbeHint: 'Probes the channel and attaches the quota snapshot to the primary model result',
         linkedAccount: 'Linked Account',
         linkedAccountPlaceholder: 'Select an account',
         linkedAccountHint: 'Quota data comes from the selected account (reuses the account-side usage/balance queries)',
         linkedAccountEmpty: 'No accounts on this platform yet. Add one in Account Management first',
+        linkedAccountMissing: 'The linked account no longer exists or is not accessible. Please re-select an account',
+        openAIQuotaProbeHint: 'Note: on the OpenAI platform the usage query may trigger a Codex probe request that consumes the account\'s own quota (at most once every 10 minutes)',
         apiMode: 'OpenAI protocol',
         apiModeChatCompletions: 'OpenAI Compatible',
         apiModeChatCompletionsHint: 'Use /v1/chat/completions with messages; works for most compatible providers.',
diff --git a/frontend/src/i18n/locales/zh/admin/channels.ts b/frontend/src/i18n/locales/zh/admin/channels.ts
index f18787ac3b9b..60e154858b50 100644
--- a/frontend/src/i18n/locales/zh/admin/channels.ts
+++ b/frontend/src/i18n/locales/zh/admin/channels.ts
@@ -530,13 +530,15 @@ export default {
         checkModeProbe: '探活',
         checkModeProbeHint: '向上游发送轻量 LLM 请求，检测可用性与延迟',
         checkModeQuota: '配额',
-        checkModeQuotaHint: '只查询关联账号的用量滚动窗口/余额，不产生 LLM 请求',
+        checkModeQuotaHint: '只查询关联账号的用量滚动窗口/余额，不发送探活请求',
         checkModeQuotaProbe: '探活 + 配额',
         checkModeQuotaProbeHint: '探活的同时查询配额，用量快照附加在主模型结果上',
         linkedAccount: '关联账号',
         linkedAccountPlaceholder: '选择账号',
         linkedAccountHint: '配额数据来自所选账号（复用账号管理侧的用量/余额查询）',
         linkedAccountEmpty: '当前平台暂无账号，请先在账号管理中添加',
+        linkedAccountMissing: '关联账号已不存在或不可访问，请重新选择账号',
+        openAIQuotaProbeHint: '注意：OpenAI 平台的用量查询可能触发 Codex 探测请求，会消耗账号自身的额度（每 10 分钟最多触发一次）',
         apiMode: 'OpenAI 协议',
         apiModeChatCompletions: 'OpenAI Compatible',
         apiModeChatCompletionsHint: '使用 /v1/chat/completions，发送 messages；适合大多数兼容站。',
diff --git a/frontend/src/views/admin/__tests__/ChannelMonitorView.grok.spec.ts b/frontend/src/views/admin/__tests__/ChannelMonitorView.grok.spec.ts
index 94f619aea9c6..007876a72ffc 100644
--- a/frontend/src/views/admin/__tests__/ChannelMonitorView.grok.spec.ts
+++ b/frontend/src/views/admin/__tests__/ChannelMonitorView.grok.spec.ts
@@ -10,8 +10,10 @@ import {
   PROVIDER_GROK,
 } from '@/constants/channelMonitor'
 
-const { listTemplates } = vi.hoisted(() => ({
+const { listTemplates, accountsList, accountsGetById } = vi.hoisted(() => ({
   listTemplates: vi.fn(),
+  accountsList: vi.fn(),
+  accountsGetById: vi.fn(),
 }))
 
 
@@ -34,6 +36,10 @@ vi.mock('@/api/admin', () => ({
     channelMonitorTemplate: {
       list: listTemplates,
     },
+    accounts: {
+      list: (...args: unknown[]) => accountsList(...args),
+      getById: (...args: unknown[]) => accountsGetById(...args),
+    },
   },
 }))
 
@@ -85,6 +91,8 @@ function mountDialog() {
 describe('channel monitor Grok provider', () => {
   beforeEach(() => {
     listTemplates.mockReset().mockResolvedValue({ items: [] })
+    accountsList.mockReset().mockResolvedValue({ items: [] })
+    accountsGetById.mockReset()
   })
 
   it('offers Grok in the responsive provider grid and prefills its official defaults', async () => {
diff --git a/frontend/src/views/admin/__tests__/MonitorFormDialog.accountSelector.spec.ts b/frontend/src/views/admin/__tests__/MonitorFormDialog.accountSelector.spec.ts
new file mode 100644
index 000000000000..9ff3fcd2b200
--- /dev/null
+++ b/frontend/src/views/admin/__tests__/MonitorFormDialog.accountSelector.spec.ts
@@ -0,0 +1,311 @@
+import { defineComponent, nextTick } from 'vue'
+import { flushPromises, mount, type VueWrapper } from '@vue/test-utils'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+
+import type { ChannelMonitor } from '@/api/admin/channelMonitor'
+import MonitorFormDialog from '@/components/admin/monitor/MonitorFormDialog.vue'
+
+const {
+  listTemplates,
+  accountsList,
+  accountsGetById,
+  monitorCreate,
+  monitorUpdate,
+} = vi.hoisted(() => ({
+  listTemplates: vi.fn(),
+  accountsList: vi.fn(),
+  accountsGetById: vi.fn(),
+  monitorCreate: vi.fn(),
+  monitorUpdate: vi.fn(),
+}))
+
+
+vi.mock('@/utils/featureFlags', () => ({
+  isChannelMonitorV1Mode: () => true,
+  isChannelMonitorV2Mode: () => false,
+  getChannelMonitorMode: () => 'v1' as const,
+}))
+
+vi.mock('@/features/channel-monitor-v2/MonitorSettingsPanel.vue', () => ({
+  default: { name: 'MonitorSettingsPanel', template: '<div data-testid="v2-settings" />' },
+}))
+
+vi.mock('@/api/admin', () => ({
+  adminAPI: {
+    channelMonitor: {
+      create: monitorCreate,
+      update: monitorUpdate,
+    },
+    channelMonitorTemplate: {
+      list: listTemplates,
+    },
+    accounts: {
+      list: (...args: unknown[]) => accountsList(...args),
+      getById: (...args: unknown[]) => accountsGetById(...args),
+    },
+  },
+}))
+
+vi.mock('@/api/keys', () => ({
+  keysAPI: { list: vi.fn() },
+}))
+
+vi.mock('@/api/groups', () => ({
+  userGroupsAPI: { getUserGroupRates: vi.fn() },
+}))
+
+vi.mock('@/stores/app', () => ({
+  useAppStore: () => ({
+    cachedPublicSettings: null,
+    showError: vi.fn(),
+    showSuccess: vi.fn(),
+  }),
+}))
+
+vi.mock('vue-i18n', async () => {
+  const actual = await vi.importActual<typeof import('vue-i18n')>('vue-i18n')
+  return {
+    ...actual,
+    useI18n: () => ({ t: (key: string) => key }),
+  }
+})
+
+const BaseDialogStub = defineComponent({
+  props: { show: { type: Boolean, default: false } },
+  template: '<div v-if="show"><slot /><slot name="footer" /></div>',
+})
+
+function makeMonitor(overrides: Partial<ChannelMonitor> = {}): ChannelMonitor {
+  return {
+    id: 42,
+    name: 'primary',
+    provider: 'openai',
+    api_mode: 'chat_completions',
+    endpoint: 'https://api.example.com',
+    api_key_masked: 'sk-t***',
+    primary_model: 'gpt-4o-mini',
+    extra_models: [],
+    group_name: '',
+    enabled: true,
+    interval_seconds: 60,
+    jitter_seconds: 0,
+    last_checked_at: null,
+    created_by: 1,
+    created_at: '2026-07-16T00:00:00Z',
+    updated_at: '2026-07-16T00:00:00Z',
+    primary_status: '',
+    primary_latency_ms: null,
+    availability_7d: 0,
+    extra_models_status: [],
+    template_id: null,
+    extra_headers: {},
+    body_override_mode: 'off',
+    body_override: null,
+    check_mode: 'probe',
+    account_id: null,
+    ...overrides,
+  }
+}
+
+let unmountWrapper: (() => void) | undefined
+
+function mountDialog(monitor: ChannelMonitor | null = null) {
+  const wrapper = mount(MonitorFormDialog, {
+    props: { show: true, monitor },
+    global: {
+      stubs: {
+        BaseDialog: BaseDialogStub,
+        Toggle: true,
+        ModelTagInput: true,
+        MonitorKeyPickerDialog: true,
+        MonitorAdvancedRequestConfig: true,
+      },
+    },
+  })
+  unmountWrapper = () => wrapper.unmount()
+  return wrapper
+}
+
+const accountTrigger = (wrapper: VueWrapper) =>
+  wrapper.get('[data-testid="monitor-linked-account"] button')
+
+const openAccountDropdown = async (wrapper: VueWrapper) => {
+  await accountTrigger(wrapper).trigger('click')
+  await nextTick()
+  const dropdown = document.body.querySelector<HTMLElement>('.select-dropdown-portal')
+  expect(dropdown).not.toBeNull()
+  return dropdown as HTMLElement
+}
+
+const clickOption = (dropdown: HTMLElement, text: string) => {
+  const option = [...dropdown.querySelectorAll('.select-option')].find((el) =>
+    el.textContent?.includes(text),
+  )
+  expect(option, `option containing "${text}" not found`).toBeDefined()
+  ;(option as HTMLElement).click()
+}
+
+const typeAccountSearch = async (dropdown: HTMLElement, query: string) => {
+  const input = dropdown.querySelector<HTMLInputElement>('.select-search-input')
+  expect(input).not.toBeNull()
+  input!.value = query
+  input!.dispatchEvent(new Event('input'))
+  await nextTick()
+}
+
+afterEach(() => {
+  unmountWrapper?.()
+  unmountWrapper = undefined
+  document.body.innerHTML = ''
+  vi.useRealTimers()
+  vi.restoreAllMocks()
+})
+
+describe('MonitorFormDialog linked account selector', () => {
+  beforeEach(() => {
+    listTemplates.mockReset().mockResolvedValue({ items: [] })
+    accountsList.mockReset().mockResolvedValue({ items: [] })
+    accountsGetById.mockReset()
+    monitorCreate.mockReset().mockResolvedValue({})
+    monitorUpdate.mockReset().mockResolvedValue({})
+  })
+
+  it('loads the first page of provider accounts when quota mode is enabled', async () => {
+    accountsList.mockResolvedValue({ items: [{ id: 1, name: 'a', platform: 'anthropic' }] })
+    const wrapper = mountDialog()
+    await flushPromises()
+    expect(accountsList).not.toHaveBeenCalled()
+
+    await wrapper.get('[data-testid="monitor-check-mode-quota"]').trigger('click')
+    await flushPromises()
+
+    expect(accountsList).toHaveBeenCalledTimes(1)
+    expect(accountsList).toHaveBeenCalledWith(
+      1,
+      50,
+      { platform: 'anthropic' },
+      { signal: expect.any(AbortSignal) },
+    )
+  })
+
+  it('debounced typing hits server-side search and aborts the previous request', async () => {
+    vi.useFakeTimers()
+    const wrapper = mountDialog()
+    await wrapper.get('[data-testid="monitor-check-mode-quota"]').trigger('click')
+    await vi.advanceTimersByTimeAsync(0)
+
+    const dropdown = await openAccountDropdown(wrapper)
+    await typeAccountSearch(dropdown, 'hidden')
+    expect(accountsList).toHaveBeenCalledTimes(1, 'debounce window must not fire a request')
+
+    await vi.advanceTimersByTimeAsync(300)
+
+    expect(accountsList).toHaveBeenCalledTimes(2)
+    const lastCall = accountsList.mock.calls[accountsList.mock.calls.length - 1]
+    expect(lastCall[2]).toMatchObject({ platform: 'anthropic', search: 'hidden' })
+    const firstSignal = (accountsList.mock.calls[0][3] as { signal: AbortSignal }).signal
+    expect(firstSignal.aborted).toBe(true)
+  })
+
+  it('hydrates a bound account missing from the first page and keeps the binding', async () => {
+    accountsList.mockResolvedValue({ items: [{ id: 1, name: 'first page account', platform: 'anthropic' }] })
+    accountsGetById.mockResolvedValue({ id: 999, name: 'hidden gem', platform: 'anthropic' })
+    const wrapper = mountDialog(makeMonitor({
+      provider: 'anthropic',
+      check_mode: 'quota',
+      account_id: 999,
+      endpoint: '',
+      primary_model: 'quota',
+    }))
+    await flushPromises()
+
+    expect(accountsGetById).toHaveBeenCalledWith(999)
+    expect(accountTrigger(wrapper).text()).toContain('hidden gem (#999)')
+
+    await wrapper.get('#channel-monitor-form').trigger('submit')
+    await flushPromises()
+    expect(monitorUpdate).toHaveBeenCalledWith(42, expect.objectContaining({ account_id: 999 }))
+  })
+
+  it('clears the binding with a visible hint when the bound account cannot be loaded', async () => {
+    accountsGetById.mockRejectedValue(new Error('gone'))
+    const wrapper = mountDialog(makeMonitor({
+      provider: 'anthropic',
+      check_mode: 'quota',
+      account_id: 999,
+      endpoint: '',
+      primary_model: 'quota',
+    }))
+    await flushPromises()
+
+    expect(wrapper.text()).toContain('admin.channelMonitor.form.linkedAccountMissing')
+    expect(accountTrigger(wrapper).text()).not.toContain('hidden gem')
+
+    await wrapper.get('#channel-monitor-form').trigger('submit')
+    await flushPromises()
+    expect(monitorUpdate).not.toHaveBeenCalled()
+  })
+
+  it('shows the OpenAI codex probe hint only in quota mode on openai', async () => {
+    const wrapper = mountDialog()
+    await flushPromises()
+    expect(wrapper.text()).not.toContain('openAIQuotaProbeHint')
+
+    await wrapper.get('[data-testid="monitor-check-mode-quota"]').trigger('click')
+    await flushPromises()
+    // anthropic 平台不显示。
+    expect(wrapper.text()).not.toContain('openAIQuotaProbeHint')
+
+    await wrapper.get('[data-testid="monitor-provider-openai"]').trigger('click')
+    await flushPromises()
+    expect(wrapper.text()).toContain('openAIQuotaProbeHint')
+  })
+
+  it('keeps a picked account visible when a later search hides it', async () => {
+    vi.useFakeTimers()
+    accountsList
+      .mockResolvedValueOnce({ items: [
+        { id: 1, name: 'alpha', platform: 'anthropic' },
+        { id: 2, name: 'beta', platform: 'anthropic' },
+      ] })
+      .mockResolvedValueOnce({ items: [{ id: 2, name: 'beta', platform: 'anthropic' }] })
+    const wrapper = mountDialog()
+    await wrapper.get('[data-testid="monitor-check-mode-quota"]').trigger('click')
+    await vi.advanceTimersByTimeAsync(0)
+
+    const dropdown = await openAccountDropdown(wrapper)
+    clickOption(dropdown, 'alpha (#1)')
+    await nextTick()
+    expect(accountTrigger(wrapper).text()).toContain('alpha (#1)')
+
+    // 搜索 beta：结果页只剩 beta，但已选中的 alpha 仍固定展示、提交仍保留绑定。
+    const reopened = await openAccountDropdown(wrapper)
+    await typeAccountSearch(reopened, 'beta')
+    await vi.advanceTimersByTimeAsync(300)
+
+    expect(accountTrigger(wrapper).text()).toContain('alpha (#1)')
+
+    await wrapper.findAll('input[type="text"]')[0].setValue('my monitor')
+    await wrapper.get('#channel-monitor-form').trigger('submit')
+    await vi.advanceTimersByTimeAsync(0)
+    expect(monitorCreate).toHaveBeenCalledWith(expect.objectContaining({ account_id: 1 }))
+  })
+
+  it('clears the account binding when the provider switches', async () => {
+    accountsList.mockResolvedValue({ items: [{ id: 1, name: 'alpha', platform: 'anthropic' }] })
+    const wrapper = mountDialog()
+    await wrapper.get('[data-testid="monitor-check-mode-quota"]').trigger('click')
+    await flushPromises()
+
+    const dropdown = await openAccountDropdown(wrapper)
+    clickOption(dropdown, 'alpha (#1)')
+    await nextTick()
+    expect(accountTrigger(wrapper).text()).toContain('alpha (#1)')
+
+    await wrapper.get('[data-testid="monitor-provider-openai"]').trigger('click')
+    await flushPromises()
+
+    expect(accountTrigger(wrapper).text()).not.toContain('alpha')
+    expect(accountTrigger(wrapper).text()).toContain('linkedAccountPlaceholder')
+  })
+})

From 49504adc98d2b6d539491e865a340e644548979e Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Tue, 18 Aug 2026 10:03:19 +0000
Subject: [PATCH 68/68] chore: sync VERSION to 0.1.178 [skip ci]

---
 backend/cmd/server/VERSION | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/backend/cmd/server/VERSION b/backend/cmd/server/VERSION
index dfb8b6438c1a..e5c7f62e4e7a 100644
--- a/backend/cmd/server/VERSION
+++ b/backend/cmd/server/VERSION
@@ -1 +1 @@
-0.1.177
+0.1.178