From 944c2952c72a01b2bf293fce6b9a0b28a0075ebb Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Fri, 21 Aug 2026 12:57:20 +0800
Subject: [PATCH 01/12] fix: wait for managed input before activating session

---
 app/api/session/dispatch/route.ts      |  9 +++++++++
 hooks/useRoom.ts                       |  4 +++-
 lib/session-dispatch-client.ts         |  6 ++++++
 tests/session-dispatch-client.test.mjs |  6 +++++-
 tests/session-start-dispatch.test.mjs  | 11 +++++++++++
 5 files changed, 34 insertions(+), 2 deletions(-)

diff --git a/app/api/session/dispatch/route.ts b/app/api/session/dispatch/route.ts
index 982817d87..a3f51ae5b 100644
--- a/app/api/session/dispatch/route.ts
+++ b/app/api/session/dispatch/route.ts
@@ -20,6 +20,10 @@ export async function POST(req: Request) {
     agent_name?: string;
     sessionId?: string;
     session_id?: string;
+    requireAgentSessionReady?: boolean;
+    require_agent_session_ready?: boolean;
+    requireRoomInputParticipantsReady?: boolean;
+    require_room_input_participants_ready?: boolean;
     requireRoomVideoInputReady?: boolean;
     require_room_video_input_ready?: boolean;
   };
@@ -57,6 +61,11 @@ export async function POST(req: Request) {
       sessionId,
       agentName,
       readiness: {
+        requireAgentSessionReady:
+          body.requireAgentSessionReady === true || body.require_agent_session_ready === true,
+        requireRoomInputParticipantsReady:
+          body.requireRoomInputParticipantsReady === true ||
+          body.require_room_input_participants_ready === true,
         requireRoomVideoInputReady:
           body.requireRoomVideoInputReady === true || body.require_room_video_input_ready === true,
       },
diff --git a/hooks/useRoom.ts b/hooks/useRoom.ts
index d44922f1d..f79cf2526 100644
--- a/hooks/useRoom.ts
+++ b/hooks/useRoom.ts
@@ -212,7 +212,6 @@ export function useRoom(appConfig: AppConfig) {
       await recoverFromStartError(error);
     };
 
-    setIsSessionActive(true);
     beginFrontendObservabilitySession(room);
 
     const dispatchAgentSession = async () => {
@@ -220,6 +219,8 @@ export function useRoom(appConfig: AppConfig) {
       dispatchSessionId = sessionId;
       const signal = beginAgentSessionStart(room.name, sessionId);
       const dispatchPromise = requestAgentSessionDispatch(appConfig.agentName, sessionId, {
+        requireAgentSessionReady: usesManagedRoomInput,
+        requireRoomInputParticipantsReady: usesManagedRoomInput,
         requireRoomVideoInputReady: requiresRoomVideoInputReady(appConfig),
         signal,
       });
@@ -310,6 +311,7 @@ export function useRoom(appConfig: AppConfig) {
       if (!usesSandboxConcurrentStartup) {
         await dispatchAgentSession();
       }
+      setIsSessionActive(true);
     } catch (error) {
       await handleStartError(error);
     }
diff --git a/lib/session-dispatch-client.ts b/lib/session-dispatch-client.ts
index 0d6b7ef1b..10812ba40 100644
--- a/lib/session-dispatch-client.ts
+++ b/lib/session-dispatch-client.ts
@@ -1,5 +1,7 @@
 type DispatchOptions = {
   signal?: AbortSignal;
+  requireAgentSessionReady?: boolean;
+  requireRoomInputParticipantsReady?: boolean;
   requireRoomVideoInputReady?: boolean;
 };
 
@@ -27,6 +29,10 @@ export async function requestAgentSessionDispatch(
     body: JSON.stringify({
       agentName: normalizedAgentName,
       sessionId: normalizedSessionId,
+      ...(options.requireAgentSessionReady ? { requireAgentSessionReady: true } : {}),
+      ...(options.requireRoomInputParticipantsReady
+        ? { requireRoomInputParticipantsReady: true }
+        : {}),
       ...(options.requireRoomVideoInputReady ? { requireRoomVideoInputReady: true } : {}),
     }),
     signal: options.signal,
diff --git a/tests/session-dispatch-client.test.mjs b/tests/session-dispatch-client.test.mjs
index 09bb1e45a..b57f06077 100644
--- a/tests/session-dispatch-client.test.mjs
+++ b/tests/session-dispatch-client.test.mjs
@@ -43,7 +43,7 @@ test('agent session dispatch sends only canonical session id to Next API', async
   }
 });
 
-test('agent session dispatch can require room video input readiness', async () => {
+test('agent session dispatch can require managed room input readiness', async () => {
   const originalFetch = globalThis.fetch;
   let postedBody;
   globalThis.fetch = async (_url, init) => {
@@ -55,12 +55,16 @@ test('agent session dispatch can require room video input readiness', async () =
     const { requestAgentSessionDispatch } = await loadSessionDispatchClientModule();
 
     await requestAgentSessionDispatch('agent-a', '11111111-2222-4333-8444-555555555555', {
+      requireAgentSessionReady: true,
+      requireRoomInputParticipantsReady: true,
       requireRoomVideoInputReady: true,
     });
 
     assert.deepEqual(postedBody, {
       agentName: 'agent-a',
       sessionId: '11111111-2222-4333-8444-555555555555',
+      requireAgentSessionReady: true,
+      requireRoomInputParticipantsReady: true,
       requireRoomVideoInputReady: true,
     });
   } finally {
diff --git a/tests/session-start-dispatch.test.mjs b/tests/session-start-dispatch.test.mjs
index dca58fa7e..7af9d3f38 100644
--- a/tests/session-start-dispatch.test.mjs
+++ b/tests/session-start-dispatch.test.mjs
@@ -177,6 +177,10 @@ test('session dispatch route only accepts anonymous LiveKit agent fallback after
   );
   assert.match(routeSource, /requireRoomVideoInputReady/);
   assert.match(routeSource, /require_room_video_input_ready/);
+  assert.match(routeSource, /requireAgentSessionReady/);
+  assert.match(routeSource, /require_agent_session_ready/);
+  assert.match(routeSource, /requireRoomInputParticipantsReady/);
+  assert.match(routeSource, /require_room_input_participants_ready/);
   assert.match(readinessSource, /type AgentParticipantMatchOptions/);
   assert.match(readinessSource, /type ReusableAgentParticipantOptions/);
   assert.match(readinessSource, /allowAnonymousLiveKitAgentFallback/);
@@ -211,6 +215,13 @@ test('start call dispatches the agent with a cancellable room session id', async
     useRoomSource,
     /requireRoomVideoInputReady: requiresRoomVideoInputReady\(appConfig\)/
   );
+  assert.match(useRoomSource, /requireAgentSessionReady: usesManagedRoomInput/);
+  assert.match(useRoomSource, /requireRoomInputParticipantsReady: usesManagedRoomInput/);
+  assert.ok(
+    useRoomSource.lastIndexOf('setIsSessionActive(true)') >
+      useRoomSource.lastIndexOf('await dispatchAgentSession()'),
+    'session view must become active only after dispatch readiness completes'
+  );
   assert.doesNotMatch(useRoomSource, /requestAgentSessionDispatch\(\s*room\.name,/);
 });
 

From deda8fa7cec594dbafaa9fe6b824ada064e79bcd Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Fri, 21 Aug 2026 13:18:06 +0800
Subject: [PATCH 02/12] fix: wait for configured session inputs

---
 app/api/session/session-dispatch-service.ts |  2 +-
 hooks/useRoom.ts                            | 10 ++++++++--
 lib/input-device-config.ts                  | 10 ++++++++++
 tests/local-dispatch-config.test.mjs        |  9 +++++++++
 tests/session-prewarm.test.mjs              |  2 +-
 tests/session-start-dispatch.test.mjs       |  6 +++++-
 6 files changed, 34 insertions(+), 5 deletions(-)

diff --git a/app/api/session/session-dispatch-service.ts b/app/api/session/session-dispatch-service.ts
index 067468acc..d05b9ea5b 100644
--- a/app/api/session/session-dispatch-service.ts
+++ b/app/api/session/session-dispatch-service.ts
@@ -75,7 +75,7 @@ const globalForInFlightDispatches = globalThis as typeof globalThis & {
 const inFlightDispatches =
   globalForInFlightDispatches.__liveavatarInFlightDispatches ??
   (globalForInFlightDispatches.__liveavatarInFlightDispatches = new Map());
-const DEFAULT_AGENT_DISPATCH_TIMEOUT_MS = 8_000;
+const DEFAULT_AGENT_DISPATCH_TIMEOUT_MS = 30_000;
 const DEFAULT_PREWARM_TOTAL_TIMEOUT_MS = 45_000;
 
 export type PrewarmPhase = 'room' | 'worker_readiness' | 'dispatch_readiness';
diff --git a/hooks/useRoom.ts b/hooks/useRoom.ts
index f79cf2526..9a5aa90fd 100644
--- a/hooks/useRoom.ts
+++ b/hooks/useRoom.ts
@@ -6,7 +6,10 @@ import { useBrowserSourceClient } from '@/hooks/useBrowserSourceClient';
 import { getVoiceSessionId, resetVoiceSessionId } from '@/lib/browser-room-session';
 import { readConnectionDetailsResponse } from '@/lib/connection-details-response';
 import { isValidConnectionRoomId } from '@/lib/connection-room-id';
-import { usesServerRoomInputDevice } from '@/lib/input-device-config';
+import {
+  usesBothServerRoomInputParticipants,
+  usesServerRoomInputDevice,
+} from '@/lib/input-device-config';
 import {
   FRONTEND_EVENTS,
   beginFrontendObservabilitySession,
@@ -220,7 +223,10 @@ export function useRoom(appConfig: AppConfig) {
       const signal = beginAgentSessionStart(room.name, sessionId);
       const dispatchPromise = requestAgentSessionDispatch(appConfig.agentName, sessionId, {
         requireAgentSessionReady: usesManagedRoomInput,
-        requireRoomInputParticipantsReady: usesManagedRoomInput,
+        requireRoomInputParticipantsReady: usesBothServerRoomInputParticipants(
+          appConfig.audioInputDevice,
+          appConfig.visionInputDevice
+        ),
         requireRoomVideoInputReady: requiresRoomVideoInputReady(appConfig),
         signal,
       });
diff --git a/lib/input-device-config.ts b/lib/input-device-config.ts
index 7807150fc..e89d7bd21 100644
--- a/lib/input-device-config.ts
+++ b/lib/input-device-config.ts
@@ -42,6 +42,16 @@ export function usesServerRoomInputDevice(inputDevice: string): boolean {
   return SERVER_ROOM_INPUT_DEVICES.has(inputDevice);
 }
 
+export function usesBothServerRoomInputParticipants(
+  audioInputDevice?: string | null,
+  visionInputDevice?: string | null
+): boolean {
+  return (
+    usesServerRoomInputDevice(audioInputDevice || '') &&
+    usesServerRoomInputDevice(visionInputDevice || '')
+  );
+}
+
 export function resolveRoleInputDevices({
   inputSource,
   audioInputDevice,
diff --git a/tests/local-dispatch-config.test.mjs b/tests/local-dispatch-config.test.mjs
index 15c9a703d..abf44fd18 100644
--- a/tests/local-dispatch-config.test.mjs
+++ b/tests/local-dispatch-config.test.mjs
@@ -2,6 +2,8 @@ import assert from 'node:assert/strict';
 import { readFile } from 'node:fs/promises';
 import { test } from 'node:test';
 
+const { usesBothServerRoomInputParticipants } = await import('../lib/input-device-config.ts');
+
 async function loadAppConfigModule() {
   return import('../app-config.ts');
 }
@@ -136,6 +138,13 @@ test('frontend resolves mixed xunfei audio with browser vision role devices', as
   assert.equal(config.showDefaultCameraPreview, false);
 });
 
+test('room participant readiness follows the configured server-owned inputs', () => {
+  assert.equal(usesBothServerRoomInputParticipants('xunfei', 'generic'), true);
+  assert.equal(usesBothServerRoomInputParticipants('xunfei', 'browser'), false);
+  assert.equal(usesBothServerRoomInputParticipants('browser', 'generic'), false);
+  assert.equal(usesBothServerRoomInputParticipants('browser', 'browser'), false);
+});
+
 test('frontend normalizes invalid mixed output devices to the base role input device', async () => {
   const { resolveInputDeviceConfig } = await loadAppConfigModule();
 
diff --git a/tests/session-prewarm.test.mjs b/tests/session-prewarm.test.mjs
index 9a44fdb32..1eff3fc6b 100644
--- a/tests/session-prewarm.test.mjs
+++ b/tests/session-prewarm.test.mjs
@@ -384,7 +384,7 @@ test('regular dispatch keeps its 8s timeout while prewarm gets the default 45s t
       ),
       /agent dispatch failed/
     );
-    assert.equal(now - regularStartedAt, 8_000);
+    assert.equal(now - regularStartedAt, 30_000);
 
     const prewarmStartedAt = now;
     await assert.rejects(
diff --git a/tests/session-start-dispatch.test.mjs b/tests/session-start-dispatch.test.mjs
index 7af9d3f38..35a87e0b1 100644
--- a/tests/session-start-dispatch.test.mjs
+++ b/tests/session-start-dispatch.test.mjs
@@ -216,7 +216,11 @@ test('start call dispatches the agent with a cancellable room session id', async
     /requireRoomVideoInputReady: requiresRoomVideoInputReady\(appConfig\)/
   );
   assert.match(useRoomSource, /requireAgentSessionReady: usesManagedRoomInput/);
-  assert.match(useRoomSource, /requireRoomInputParticipantsReady: usesManagedRoomInput/);
+  assert.match(
+    useRoomSource,
+    /requireRoomInputParticipantsReady: usesBothServerRoomInputParticipants\(/
+  );
+  assert.match(useRoomSource, /await Promise\.allSettled\(\[/);
   assert.ok(
     useRoomSource.lastIndexOf('setIsSessionActive(true)') >
       useRoomSource.lastIndexOf('await dispatchAgentSession()'),

From 50c624ce5a4f2b545917339d202ee7771f9b59da Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Fri, 21 Aug 2026 13:22:38 +0800
Subject: [PATCH 03/12] fix: use agent session readiness as startup gate

---
 app/api/session/dispatch/route.ts      |  5 -----
 hooks/useRoom.ts                       |  9 +--------
 lib/input-device-config.ts             | 10 ----------
 lib/session-dispatch-client.ts         |  4 ----
 tests/local-dispatch-config.test.mjs   |  9 ---------
 tests/session-dispatch-client.test.mjs |  4 +---
 tests/session-start-dispatch.test.mjs  |  6 ------
 7 files changed, 2 insertions(+), 45 deletions(-)

diff --git a/app/api/session/dispatch/route.ts b/app/api/session/dispatch/route.ts
index a3f51ae5b..641fb02ad 100644
--- a/app/api/session/dispatch/route.ts
+++ b/app/api/session/dispatch/route.ts
@@ -22,8 +22,6 @@ export async function POST(req: Request) {
     session_id?: string;
     requireAgentSessionReady?: boolean;
     require_agent_session_ready?: boolean;
-    requireRoomInputParticipantsReady?: boolean;
-    require_room_input_participants_ready?: boolean;
     requireRoomVideoInputReady?: boolean;
     require_room_video_input_ready?: boolean;
   };
@@ -63,9 +61,6 @@ export async function POST(req: Request) {
       readiness: {
         requireAgentSessionReady:
           body.requireAgentSessionReady === true || body.require_agent_session_ready === true,
-        requireRoomInputParticipantsReady:
-          body.requireRoomInputParticipantsReady === true ||
-          body.require_room_input_participants_ready === true,
         requireRoomVideoInputReady:
           body.requireRoomVideoInputReady === true || body.require_room_video_input_ready === true,
       },
diff --git a/hooks/useRoom.ts b/hooks/useRoom.ts
index 9a5aa90fd..1206c8c5a 100644
--- a/hooks/useRoom.ts
+++ b/hooks/useRoom.ts
@@ -6,10 +6,7 @@ import { useBrowserSourceClient } from '@/hooks/useBrowserSourceClient';
 import { getVoiceSessionId, resetVoiceSessionId } from '@/lib/browser-room-session';
 import { readConnectionDetailsResponse } from '@/lib/connection-details-response';
 import { isValidConnectionRoomId } from '@/lib/connection-room-id';
-import {
-  usesBothServerRoomInputParticipants,
-  usesServerRoomInputDevice,
-} from '@/lib/input-device-config';
+import { usesServerRoomInputDevice } from '@/lib/input-device-config';
 import {
   FRONTEND_EVENTS,
   beginFrontendObservabilitySession,
@@ -223,10 +220,6 @@ export function useRoom(appConfig: AppConfig) {
       const signal = beginAgentSessionStart(room.name, sessionId);
       const dispatchPromise = requestAgentSessionDispatch(appConfig.agentName, sessionId, {
         requireAgentSessionReady: usesManagedRoomInput,
-        requireRoomInputParticipantsReady: usesBothServerRoomInputParticipants(
-          appConfig.audioInputDevice,
-          appConfig.visionInputDevice
-        ),
         requireRoomVideoInputReady: requiresRoomVideoInputReady(appConfig),
         signal,
       });
diff --git a/lib/input-device-config.ts b/lib/input-device-config.ts
index e89d7bd21..7807150fc 100644
--- a/lib/input-device-config.ts
+++ b/lib/input-device-config.ts
@@ -42,16 +42,6 @@ export function usesServerRoomInputDevice(inputDevice: string): boolean {
   return SERVER_ROOM_INPUT_DEVICES.has(inputDevice);
 }
 
-export function usesBothServerRoomInputParticipants(
-  audioInputDevice?: string | null,
-  visionInputDevice?: string | null
-): boolean {
-  return (
-    usesServerRoomInputDevice(audioInputDevice || '') &&
-    usesServerRoomInputDevice(visionInputDevice || '')
-  );
-}
-
 export function resolveRoleInputDevices({
   inputSource,
   audioInputDevice,
diff --git a/lib/session-dispatch-client.ts b/lib/session-dispatch-client.ts
index 10812ba40..1635966dc 100644
--- a/lib/session-dispatch-client.ts
+++ b/lib/session-dispatch-client.ts
@@ -1,7 +1,6 @@
 type DispatchOptions = {
   signal?: AbortSignal;
   requireAgentSessionReady?: boolean;
-  requireRoomInputParticipantsReady?: boolean;
   requireRoomVideoInputReady?: boolean;
 };
 
@@ -30,9 +29,6 @@ export async function requestAgentSessionDispatch(
       agentName: normalizedAgentName,
       sessionId: normalizedSessionId,
       ...(options.requireAgentSessionReady ? { requireAgentSessionReady: true } : {}),
-      ...(options.requireRoomInputParticipantsReady
-        ? { requireRoomInputParticipantsReady: true }
-        : {}),
       ...(options.requireRoomVideoInputReady ? { requireRoomVideoInputReady: true } : {}),
     }),
     signal: options.signal,
diff --git a/tests/local-dispatch-config.test.mjs b/tests/local-dispatch-config.test.mjs
index abf44fd18..15c9a703d 100644
--- a/tests/local-dispatch-config.test.mjs
+++ b/tests/local-dispatch-config.test.mjs
@@ -2,8 +2,6 @@ import assert from 'node:assert/strict';
 import { readFile } from 'node:fs/promises';
 import { test } from 'node:test';
 
-const { usesBothServerRoomInputParticipants } = await import('../lib/input-device-config.ts');
-
 async function loadAppConfigModule() {
   return import('../app-config.ts');
 }
@@ -138,13 +136,6 @@ test('frontend resolves mixed xunfei audio with browser vision role devices', as
   assert.equal(config.showDefaultCameraPreview, false);
 });
 
-test('room participant readiness follows the configured server-owned inputs', () => {
-  assert.equal(usesBothServerRoomInputParticipants('xunfei', 'generic'), true);
-  assert.equal(usesBothServerRoomInputParticipants('xunfei', 'browser'), false);
-  assert.equal(usesBothServerRoomInputParticipants('browser', 'generic'), false);
-  assert.equal(usesBothServerRoomInputParticipants('browser', 'browser'), false);
-});
-
 test('frontend normalizes invalid mixed output devices to the base role input device', async () => {
   const { resolveInputDeviceConfig } = await loadAppConfigModule();
 
diff --git a/tests/session-dispatch-client.test.mjs b/tests/session-dispatch-client.test.mjs
index b57f06077..375aa80cb 100644
--- a/tests/session-dispatch-client.test.mjs
+++ b/tests/session-dispatch-client.test.mjs
@@ -43,7 +43,7 @@ test('agent session dispatch sends only canonical session id to Next API', async
   }
 });
 
-test('agent session dispatch can require managed room input readiness', async () => {
+test('agent session dispatch can require authoritative session readiness', async () => {
   const originalFetch = globalThis.fetch;
   let postedBody;
   globalThis.fetch = async (_url, init) => {
@@ -56,7 +56,6 @@ test('agent session dispatch can require managed room input readiness', async ()
 
     await requestAgentSessionDispatch('agent-a', '11111111-2222-4333-8444-555555555555', {
       requireAgentSessionReady: true,
-      requireRoomInputParticipantsReady: true,
       requireRoomVideoInputReady: true,
     });
 
@@ -64,7 +63,6 @@ test('agent session dispatch can require managed room input readiness', async ()
       agentName: 'agent-a',
       sessionId: '11111111-2222-4333-8444-555555555555',
       requireAgentSessionReady: true,
-      requireRoomInputParticipantsReady: true,
       requireRoomVideoInputReady: true,
     });
   } finally {
diff --git a/tests/session-start-dispatch.test.mjs b/tests/session-start-dispatch.test.mjs
index 35a87e0b1..404e665bc 100644
--- a/tests/session-start-dispatch.test.mjs
+++ b/tests/session-start-dispatch.test.mjs
@@ -179,8 +179,6 @@ test('session dispatch route only accepts anonymous LiveKit agent fallback after
   assert.match(routeSource, /require_room_video_input_ready/);
   assert.match(routeSource, /requireAgentSessionReady/);
   assert.match(routeSource, /require_agent_session_ready/);
-  assert.match(routeSource, /requireRoomInputParticipantsReady/);
-  assert.match(routeSource, /require_room_input_participants_ready/);
   assert.match(readinessSource, /type AgentParticipantMatchOptions/);
   assert.match(readinessSource, /type ReusableAgentParticipantOptions/);
   assert.match(readinessSource, /allowAnonymousLiveKitAgentFallback/);
@@ -216,10 +214,6 @@ test('start call dispatches the agent with a cancellable room session id', async
     /requireRoomVideoInputReady: requiresRoomVideoInputReady\(appConfig\)/
   );
   assert.match(useRoomSource, /requireAgentSessionReady: usesManagedRoomInput/);
-  assert.match(
-    useRoomSource,
-    /requireRoomInputParticipantsReady: usesBothServerRoomInputParticipants\(/
-  );
   assert.match(useRoomSource, /await Promise\.allSettled\(\[/);
   assert.ok(
     useRoomSource.lastIndexOf('setIsSessionActive(true)') >

From 18db439cd6569fa4bf7a7e6571eefe8fe89aa9d5 Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Fri, 21 Aug 2026 13:24:57 +0800
Subject: [PATCH 04/12] fix: keep video readiness out of voice startup

---
 app/api/session/dispatch/route.ts      | 4 ----
 hooks/useRoom.ts                       | 8 --------
 lib/session-dispatch-client.ts         | 2 --
 tests/session-dispatch-client.test.mjs | 2 --
 tests/session-start-dispatch.test.mjs  | 6 ------
 5 files changed, 22 deletions(-)

diff --git a/app/api/session/dispatch/route.ts b/app/api/session/dispatch/route.ts
index 641fb02ad..59661c172 100644
--- a/app/api/session/dispatch/route.ts
+++ b/app/api/session/dispatch/route.ts
@@ -22,8 +22,6 @@ export async function POST(req: Request) {
     session_id?: string;
     requireAgentSessionReady?: boolean;
     require_agent_session_ready?: boolean;
-    requireRoomVideoInputReady?: boolean;
-    require_room_video_input_ready?: boolean;
   };
   try {
     body = await req.json();
@@ -61,8 +59,6 @@ export async function POST(req: Request) {
       readiness: {
         requireAgentSessionReady:
           body.requireAgentSessionReady === true || body.require_agent_session_ready === true,
-        requireRoomVideoInputReady:
-          body.requireRoomVideoInputReady === true || body.require_room_video_input_ready === true,
       },
     });
     return NextResponse.json({ status: 'dispatched', roomName, agentName, sessionId, dispatch });
diff --git a/hooks/useRoom.ts b/hooks/useRoom.ts
index 1206c8c5a..25e023a88 100644
--- a/hooks/useRoom.ts
+++ b/hooks/useRoom.ts
@@ -6,7 +6,6 @@ import { useBrowserSourceClient } from '@/hooks/useBrowserSourceClient';
 import { getVoiceSessionId, resetVoiceSessionId } from '@/lib/browser-room-session';
 import { readConnectionDetailsResponse } from '@/lib/connection-details-response';
 import { isValidConnectionRoomId } from '@/lib/connection-room-id';
-import { usesServerRoomInputDevice } from '@/lib/input-device-config';
 import {
   FRONTEND_EVENTS,
   beginFrontendObservabilitySession,
@@ -27,12 +26,6 @@ import {
   waitForAgentSessionStop,
 } from '@/lib/session-stop-client';
 
-function requiresRoomVideoInputReady(appConfig: AppConfig) {
-  return appConfig.visionInputDevice
-    ? usesServerRoomInputDevice(appConfig.visionInputDevice)
-    : false;
-}
-
 export function useRoom(appConfig: AppConfig) {
   const aborted = useRef(false);
   const sessionIdRef = useRef<string | null>(null);
@@ -220,7 +213,6 @@ export function useRoom(appConfig: AppConfig) {
       const signal = beginAgentSessionStart(room.name, sessionId);
       const dispatchPromise = requestAgentSessionDispatch(appConfig.agentName, sessionId, {
         requireAgentSessionReady: usesManagedRoomInput,
-        requireRoomVideoInputReady: requiresRoomVideoInputReady(appConfig),
         signal,
       });
       registerAgentSessionDispatch(room.name, sessionId, dispatchPromise);
diff --git a/lib/session-dispatch-client.ts b/lib/session-dispatch-client.ts
index 1635966dc..06fa7c93d 100644
--- a/lib/session-dispatch-client.ts
+++ b/lib/session-dispatch-client.ts
@@ -1,7 +1,6 @@
 type DispatchOptions = {
   signal?: AbortSignal;
   requireAgentSessionReady?: boolean;
-  requireRoomVideoInputReady?: boolean;
 };
 
 export class AgentSessionDispatchCancelledError extends Error {
@@ -29,7 +28,6 @@ export async function requestAgentSessionDispatch(
       agentName: normalizedAgentName,
       sessionId: normalizedSessionId,
       ...(options.requireAgentSessionReady ? { requireAgentSessionReady: true } : {}),
-      ...(options.requireRoomVideoInputReady ? { requireRoomVideoInputReady: true } : {}),
     }),
     signal: options.signal,
   });
diff --git a/tests/session-dispatch-client.test.mjs b/tests/session-dispatch-client.test.mjs
index 375aa80cb..3e0c9a8c2 100644
--- a/tests/session-dispatch-client.test.mjs
+++ b/tests/session-dispatch-client.test.mjs
@@ -56,14 +56,12 @@ test('agent session dispatch can require authoritative session readiness', async
 
     await requestAgentSessionDispatch('agent-a', '11111111-2222-4333-8444-555555555555', {
       requireAgentSessionReady: true,
-      requireRoomVideoInputReady: true,
     });
 
     assert.deepEqual(postedBody, {
       agentName: 'agent-a',
       sessionId: '11111111-2222-4333-8444-555555555555',
       requireAgentSessionReady: true,
-      requireRoomVideoInputReady: true,
     });
   } finally {
     globalThis.fetch = originalFetch;
diff --git a/tests/session-start-dispatch.test.mjs b/tests/session-start-dispatch.test.mjs
index 404e665bc..eba6d783f 100644
--- a/tests/session-start-dispatch.test.mjs
+++ b/tests/session-start-dispatch.test.mjs
@@ -175,8 +175,6 @@ test('session dispatch route only accepts anonymous LiveKit agent fallback after
     serviceSource,
     /const alreadyJoined = await findReusableAgentParticipant\(\s*roomClient,\s*roomName,\s*agentName,\s*reusableAgentOptions\s*\);/
   );
-  assert.match(routeSource, /requireRoomVideoInputReady/);
-  assert.match(routeSource, /require_room_video_input_ready/);
   assert.match(routeSource, /requireAgentSessionReady/);
   assert.match(routeSource, /require_agent_session_ready/);
   assert.match(readinessSource, /type AgentParticipantMatchOptions/);
@@ -209,10 +207,6 @@ test('start call dispatches the agent with a cancellable room session id', async
   assert.match(useRoomSource, /isExpectedStartCancellation/);
   assert.match(useRoomSource, /waitForAgentSessionStop/);
   assert.match(useRoomSource, /requestAgentSessionDispatch\(\s*appConfig\.agentName,\s*sessionId,/);
-  assert.match(
-    useRoomSource,
-    /requireRoomVideoInputReady: requiresRoomVideoInputReady\(appConfig\)/
-  );
   assert.match(useRoomSource, /requireAgentSessionReady: usesManagedRoomInput/);
   assert.match(useRoomSource, /await Promise\.allSettled\(\[/);
   assert.ok(

From 5ac7bad3de9d8bee0abcf35de4e238ba76a8e2e2 Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Fri, 21 Aug 2026 13:29:21 +0800
Subject: [PATCH 05/12] fix: recognize server serialized readiness attributes

---
 lib/session-dispatch-readiness.ts         |  7 ++++++-
 tests/session-dispatch-readiness.test.mjs | 18 ++++++++++++++++++
 2 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/lib/session-dispatch-readiness.ts b/lib/session-dispatch-readiness.ts
index 2c6026cc3..0b55e95b1 100644
--- a/lib/session-dispatch-readiness.ts
+++ b/lib/session-dispatch-readiness.ts
@@ -16,6 +16,7 @@ export type ReusableAgentParticipantOptions = AgentParticipantMatchOptions & {
 };
 
 export const AGENT_SESSION_READY_ATTRIBUTE = 'liveavatar.agent.session_ready';
+const AGENT_SESSION_READY_ATTRIBUTE_CAMEL = 'liveavatarAgentSessionReady';
 
 const ROOM_AUDIO_INPUT_IDENTITY = 'room_audio_input';
 const ROOM_VIDEO_INPUT_IDENTITY = 'room_video_input';
@@ -127,7 +128,11 @@ function isExpectedAgentParticipant(participant: ParticipantInfo, agentName: str
 }
 
 function isAgentSessionReady(participant: ParticipantInfo) {
-  return participant.attributes?.[AGENT_SESSION_READY_ATTRIBUTE] === 'true';
+  const attributes = participant.attributes ?? {};
+  return (
+    attributes[AGENT_SESSION_READY_ATTRIBUTE] === 'true' ||
+    attributes[AGENT_SESSION_READY_ATTRIBUTE_CAMEL] === 'true'
+  );
 }
 
 function isAnonymousLiveKitAgentParticipant(participant: ParticipantInfo) {
diff --git a/tests/session-dispatch-readiness.test.mjs b/tests/session-dispatch-readiness.test.mjs
index 7977a5d74..93567c1d6 100644
--- a/tests/session-dispatch-readiness.test.mjs
+++ b/tests/session-dispatch-readiness.test.mjs
@@ -62,6 +62,24 @@ test('dispatch can require room video input readiness before reusing an agent',
   );
 });
 
+test('dispatch accepts the server SDK camel-cased agent ready attribute', () => {
+  const agent = participant({
+    identity: 'agent-AJ_ready',
+    kind: ParticipantInfo_Kind.AGENT,
+    attributes: {
+      lkAgentName: 'frontdesk-agent',
+      liveavatarAgentSessionReady: 'true',
+    },
+  });
+
+  assert.equal(
+    findReusableAgentParticipant([agent], 'frontdesk-agent', {
+      requireAgentSessionReady: true,
+    }),
+    agent
+  );
+});
+
 test('dispatch can reuse an active agent once room video input is publishing', () => {
   const agent = participant({
     identity: 'agent-AJ_running',

From 5e4c87c1c4cd3137325e1f5b73f04b7e57d13ef4 Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Fri, 21 Aug 2026 13:57:28 +0800
Subject: [PATCH 06/12] fix: gate prewarm on authoritative session readiness

---
 app/api/session/session-dispatch-service.ts | 1 -
 lib/session-dispatch-readiness.ts           | 1 +
 tests/session-prewarm.test.mjs              | 6 +++---
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/app/api/session/session-dispatch-service.ts b/app/api/session/session-dispatch-service.ts
index d05b9ea5b..44075ff1d 100644
--- a/app/api/session/session-dispatch-service.ts
+++ b/app/api/session/session-dispatch-service.ts
@@ -302,7 +302,6 @@ export async function prewarmRoomSession(
             ...request,
             readiness: {
               requireAgentSessionReady: true,
-              requireRoomInputParticipantsReady: true,
             },
           },
           {
diff --git a/lib/session-dispatch-readiness.ts b/lib/session-dispatch-readiness.ts
index 0b55e95b1..51cb58e47 100644
--- a/lib/session-dispatch-readiness.ts
+++ b/lib/session-dispatch-readiness.ts
@@ -16,6 +16,7 @@ export type ReusableAgentParticipantOptions = AgentParticipantMatchOptions & {
 };
 
 export const AGENT_SESSION_READY_ATTRIBUTE = 'liveavatar.agent.session_ready';
+// livekit-server-sdk maps protobuf attribute keys to camelCase object keys.
 const AGENT_SESSION_READY_ATTRIBUTE_CAMEL = 'liveavatarAgentSessionReady';
 
 const ROOM_AUDIO_INPUT_IDENTITY = 'room_audio_input';
diff --git a/tests/session-prewarm.test.mjs b/tests/session-prewarm.test.mjs
index 1eff3fc6b..243ac835e 100644
--- a/tests/session-prewarm.test.mjs
+++ b/tests/session-prewarm.test.mjs
@@ -331,7 +331,7 @@ test('missing LiveKit configuration fails before registering a room session', as
   }
 });
 
-test('regular dispatch keeps its 8s timeout while prewarm gets the default 45s total budget', async () => {
+test('regular dispatch keeps its 30s timeout while prewarm gets the default 45s total budget', async () => {
   const originalNow = Date.now;
   const originalTimeout = process.env.AGENT_DISPATCH_TIMEOUT_MS;
   const originalPrewarmTimeout = process.env.LIVEAVATAR_PREWARM_TOTAL_TIMEOUT_MS;
@@ -1322,7 +1322,7 @@ test('shared dispatch token stays active through per-caller readiness waits', as
   assert.doesNotMatch(readinessSource, /beginRoomSessionDispatch|finishRoomSessionDispatch/);
 });
 
-test('prewarm waits for the agent session and both room input participants', async () => {
+test('prewarm completes when the agent session is ready without waiting for optional video input', async () => {
   const agentName = 'frontdesk-browser-agent-readiness';
   let roomCreated = false;
   let workerReady = false;
@@ -1395,7 +1395,7 @@ test('prewarm waits for the agent session and both room input participants', asy
   assert.deepEqual(result.readiness, {
     agentSessionReady: true,
     audioParticipantReady: true,
-    visionParticipantReady: true,
+    visionParticipantReady: false,
   });
 });
 

From 8f64cf6cf98560e8c4f41ad0d7acb13a53cecb87 Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Wed, 26 Aug 2026 19:17:01 +0800
Subject: [PATCH 07/12] fix: preserve tool preambles in chat history

---
 hooks/useChatMessages.ts           | 12 +++++++---
 lib/transcription-history.ts       | 23 ++++++++++++++++++++
 tests/chat-message-filter.test.mjs | 35 ++++++++++++++++++++++++++++++
 3 files changed, 67 insertions(+), 3 deletions(-)
 create mode 100644 lib/transcription-history.ts

diff --git a/hooks/useChatMessages.ts b/hooks/useChatMessages.ts
index e9b75f5d0..ed6be0d0a 100644
--- a/hooks/useChatMessages.ts
+++ b/hooks/useChatMessages.ts
@@ -1,4 +1,4 @@
-import { useMemo } from 'react';
+import { useEffect, useMemo, useState } from 'react';
 import { Room } from 'livekit-client';
 import {
   type ReceivedChatMessage,
@@ -9,6 +9,7 @@ import {
 } from '@livekit/components-react';
 import { type AppConfig } from '@/app-config';
 import { isRenderableChatMessage } from '@/lib/chat-message-filter';
+import { mergeTranscriptionHistory } from '@/lib/transcription-history';
 
 function transcriptionToChatMessage(
   textStream: TextStreamData,
@@ -76,10 +77,15 @@ export function useChatMessages(
   const chat = useChat();
   const room = useRoomContext();
   const transcriptions: TextStreamData[] = useTranscriptions();
+  const [transcriptionHistory, setTranscriptionHistory] = useState<TextStreamData[]>([]);
+
+  useEffect(() => {
+    setTranscriptionHistory((previous) => mergeTranscriptionHistory(previous, transcriptions));
+  }, [transcriptions]);
 
   const mergedTranscriptions = useMemo(() => {
     // 处理转录消息
-    const transcriptionMessages = transcriptions.map((transcription) =>
+    const transcriptionMessages = transcriptionHistory.map((transcription) =>
       transcriptionToChatMessage(transcription, room, config)
     );
 
@@ -112,7 +118,7 @@ export function useChatMessages(
     ].filter(isRenderableChatMessage);
 
     return merged.sort((a, b) => a.timestamp - b.timestamp);
-  }, [transcriptions, chat.chatMessages, room, config]);
+  }, [transcriptionHistory, chat.chatMessages, room, config]);
 
   return mergedTranscriptions;
 }
diff --git a/lib/transcription-history.ts b/lib/transcription-history.ts
new file mode 100644
index 000000000..fce673cf9
--- /dev/null
+++ b/lib/transcription-history.ts
@@ -0,0 +1,23 @@
+import { type TextStreamData } from '@livekit/components-react';
+
+const DEFAULT_TRANSCRIPTION_HISTORY_SIZE = 100;
+
+/**
+ * Keep completed text streams even when LiveKit replaces the current entry for
+ * the same speech segment. A tool preamble and its final answer can share one
+ * segment id while still arriving as distinct text streams.
+ */
+export function mergeTranscriptionHistory(
+  previous: TextStreamData[],
+  current: TextStreamData[],
+  maxEntries = DEFAULT_TRANSCRIPTION_HISTORY_SIZE
+): TextStreamData[] {
+  if (current.length === 0) return [];
+
+  const byStreamId = new Map(previous.map((entry) => [entry.streamInfo.id, entry]));
+  current.forEach((entry) => byStreamId.set(entry.streamInfo.id, entry));
+
+  return Array.from(byStreamId.values())
+    .sort((a, b) => a.streamInfo.timestamp - b.streamInfo.timestamp)
+    .slice(-maxEntries);
+}
diff --git a/tests/chat-message-filter.test.mjs b/tests/chat-message-filter.test.mjs
index 9f3074b99..35dd6b4fa 100644
--- a/tests/chat-message-filter.test.mjs
+++ b/tests/chat-message-filter.test.mjs
@@ -3,6 +3,19 @@ import { readFile } from 'node:fs/promises';
 import { test } from 'node:test';
 
 const { isRenderableChatMessage } = await import('../lib/chat-message-filter.ts');
+const { mergeTranscriptionHistory } = await import('../lib/transcription-history.ts');
+
+function transcription(id, segmentId, timestamp, text) {
+  return {
+    text,
+    participantInfo: { identity: 'frontdesk-agent' },
+    streamInfo: {
+      id,
+      timestamp,
+      attributes: { 'lk.segment_id': segmentId },
+    },
+  };
+}
 
 test('chat message filter hides empty transcription messages', () => {
   assert.equal(isRenderableChatMessage({ message: '' }), false);
@@ -19,3 +32,25 @@ test('chat message hook filters empty merged messages before sorting', async ()
   assert.match(source, /isRenderableChatMessage/);
   assert.match(source, /\.filter\(isRenderableChatMessage\)/);
 });
+
+test('transcription history preserves a tool preamble replaced by the final stream', () => {
+  const preamble = transcription('stream-preamble', 'speech-1', 100, '我查一下。');
+  const final = transcription('stream-final', 'speech-1', 200, '已经设置好了。');
+
+  const history = mergeTranscriptionHistory([preamble], [final]);
+
+  assert.deepEqual(
+    history.map(({ text }) => text),
+    ['我查一下。', '已经设置好了。']
+  );
+});
+
+test('transcription history updates partial text without duplicating one stream', () => {
+  const partial = transcription('stream-1', 'speech-1', 100, '我查');
+  const completed = transcription('stream-1', 'speech-1', 100, '我查一下。');
+
+  const history = mergeTranscriptionHistory([partial], [completed]);
+
+  assert.equal(history.length, 1);
+  assert.equal(history[0].text, '我查一下。');
+});

From 1197d4a58aaf2dfa1f7fc3076fd066ef45f6724b Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Wed, 26 Aug 2026 19:23:16 +0800
Subject: [PATCH 08/12] fix: retain transcription history across reconnects

---
 hooks/useChatMessages.ts           | 4 ++++
 lib/transcription-history.ts       | 2 +-
 tests/chat-message-filter.test.mjs | 8 ++++++++
 3 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/hooks/useChatMessages.ts b/hooks/useChatMessages.ts
index ed6be0d0a..7125e826d 100644
--- a/hooks/useChatMessages.ts
+++ b/hooks/useChatMessages.ts
@@ -79,6 +79,10 @@ export function useChatMessages(
   const transcriptions: TextStreamData[] = useTranscriptions();
   const [transcriptionHistory, setTranscriptionHistory] = useState<TextStreamData[]>([]);
 
+  useEffect(() => {
+    setTranscriptionHistory([]);
+  }, [room.name]);
+
   useEffect(() => {
     setTranscriptionHistory((previous) => mergeTranscriptionHistory(previous, transcriptions));
   }, [transcriptions]);
diff --git a/lib/transcription-history.ts b/lib/transcription-history.ts
index fce673cf9..2a16a8da8 100644
--- a/lib/transcription-history.ts
+++ b/lib/transcription-history.ts
@@ -12,7 +12,7 @@ export function mergeTranscriptionHistory(
   current: TextStreamData[],
   maxEntries = DEFAULT_TRANSCRIPTION_HISTORY_SIZE
 ): TextStreamData[] {
-  if (current.length === 0) return [];
+  if (current.length === 0) return previous;
 
   const byStreamId = new Map(previous.map((entry) => [entry.streamInfo.id, entry]));
   current.forEach((entry) => byStreamId.set(entry.streamInfo.id, entry));
diff --git a/tests/chat-message-filter.test.mjs b/tests/chat-message-filter.test.mjs
index 35dd6b4fa..632df3fcd 100644
--- a/tests/chat-message-filter.test.mjs
+++ b/tests/chat-message-filter.test.mjs
@@ -54,3 +54,11 @@ test('transcription history updates partial text without duplicating one stream'
   assert.equal(history.length, 1);
   assert.equal(history[0].text, '我查一下。');
 });
+
+test('transcription history survives a transient empty snapshot', () => {
+  const preamble = transcription('stream-1', 'speech-1', 100, '我查一下。');
+
+  const history = mergeTranscriptionHistory([preamble], []);
+
+  assert.deepEqual(history, [preamble]);
+});

From 40a51d5c514d11827311dc48eca86d05624d3708 Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Wed, 26 Aug 2026 19:53:39 +0800
Subject: [PATCH 09/12] Revert "fix: retain transcription history across
 reconnects"

This reverts commit 1197d4a58aaf2dfa1f7fc3076fd066ef45f6724b.
---
 hooks/useChatMessages.ts           | 4 ----
 lib/transcription-history.ts       | 2 +-
 tests/chat-message-filter.test.mjs | 8 --------
 3 files changed, 1 insertion(+), 13 deletions(-)

diff --git a/hooks/useChatMessages.ts b/hooks/useChatMessages.ts
index 7125e826d..ed6be0d0a 100644
--- a/hooks/useChatMessages.ts
+++ b/hooks/useChatMessages.ts
@@ -79,10 +79,6 @@ export function useChatMessages(
   const transcriptions: TextStreamData[] = useTranscriptions();
   const [transcriptionHistory, setTranscriptionHistory] = useState<TextStreamData[]>([]);
 
-  useEffect(() => {
-    setTranscriptionHistory([]);
-  }, [room.name]);
-
   useEffect(() => {
     setTranscriptionHistory((previous) => mergeTranscriptionHistory(previous, transcriptions));
   }, [transcriptions]);
diff --git a/lib/transcription-history.ts b/lib/transcription-history.ts
index 2a16a8da8..fce673cf9 100644
--- a/lib/transcription-history.ts
+++ b/lib/transcription-history.ts
@@ -12,7 +12,7 @@ export function mergeTranscriptionHistory(
   current: TextStreamData[],
   maxEntries = DEFAULT_TRANSCRIPTION_HISTORY_SIZE
 ): TextStreamData[] {
-  if (current.length === 0) return previous;
+  if (current.length === 0) return [];
 
   const byStreamId = new Map(previous.map((entry) => [entry.streamInfo.id, entry]));
   current.forEach((entry) => byStreamId.set(entry.streamInfo.id, entry));
diff --git a/tests/chat-message-filter.test.mjs b/tests/chat-message-filter.test.mjs
index 632df3fcd..35dd6b4fa 100644
--- a/tests/chat-message-filter.test.mjs
+++ b/tests/chat-message-filter.test.mjs
@@ -54,11 +54,3 @@ test('transcription history updates partial text without duplicating one stream'
   assert.equal(history.length, 1);
   assert.equal(history[0].text, '我查一下。');
 });
-
-test('transcription history survives a transient empty snapshot', () => {
-  const preamble = transcription('stream-1', 'speech-1', 100, '我查一下。');
-
-  const history = mergeTranscriptionHistory([preamble], []);
-
-  assert.deepEqual(history, [preamble]);
-});

From aec6ac6d7719ce023c6b3c5e590f69b714db7bb9 Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Wed, 26 Aug 2026 19:53:39 +0800
Subject: [PATCH 10/12] Revert "fix: preserve tool preambles in chat history"

This reverts commit 8f64cf6cf98560e8c4f41ad0d7acb13a53cecb87.
---
 hooks/useChatMessages.ts           | 12 +++-------
 lib/transcription-history.ts       | 23 --------------------
 tests/chat-message-filter.test.mjs | 35 ------------------------------
 3 files changed, 3 insertions(+), 67 deletions(-)
 delete mode 100644 lib/transcription-history.ts

diff --git a/hooks/useChatMessages.ts b/hooks/useChatMessages.ts
index ed6be0d0a..e9b75f5d0 100644
--- a/hooks/useChatMessages.ts
+++ b/hooks/useChatMessages.ts
@@ -1,4 +1,4 @@
-import { useEffect, useMemo, useState } from 'react';
+import { useMemo } from 'react';
 import { Room } from 'livekit-client';
 import {
   type ReceivedChatMessage,
@@ -9,7 +9,6 @@ import {
 } from '@livekit/components-react';
 import { type AppConfig } from '@/app-config';
 import { isRenderableChatMessage } from '@/lib/chat-message-filter';
-import { mergeTranscriptionHistory } from '@/lib/transcription-history';
 
 function transcriptionToChatMessage(
   textStream: TextStreamData,
@@ -77,15 +76,10 @@ export function useChatMessages(
   const chat = useChat();
   const room = useRoomContext();
   const transcriptions: TextStreamData[] = useTranscriptions();
-  const [transcriptionHistory, setTranscriptionHistory] = useState<TextStreamData[]>([]);
-
-  useEffect(() => {
-    setTranscriptionHistory((previous) => mergeTranscriptionHistory(previous, transcriptions));
-  }, [transcriptions]);
 
   const mergedTranscriptions = useMemo(() => {
     // 处理转录消息
-    const transcriptionMessages = transcriptionHistory.map((transcription) =>
+    const transcriptionMessages = transcriptions.map((transcription) =>
       transcriptionToChatMessage(transcription, room, config)
     );
 
@@ -118,7 +112,7 @@ export function useChatMessages(
     ].filter(isRenderableChatMessage);
 
     return merged.sort((a, b) => a.timestamp - b.timestamp);
-  }, [transcriptionHistory, chat.chatMessages, room, config]);
+  }, [transcriptions, chat.chatMessages, room, config]);
 
   return mergedTranscriptions;
 }
diff --git a/lib/transcription-history.ts b/lib/transcription-history.ts
deleted file mode 100644
index fce673cf9..000000000
--- a/lib/transcription-history.ts
+++ /dev/null
@@ -1,23 +0,0 @@
-import { type TextStreamData } from '@livekit/components-react';
-
-const DEFAULT_TRANSCRIPTION_HISTORY_SIZE = 100;
-
-/**
- * Keep completed text streams even when LiveKit replaces the current entry for
- * the same speech segment. A tool preamble and its final answer can share one
- * segment id while still arriving as distinct text streams.
- */
-export function mergeTranscriptionHistory(
-  previous: TextStreamData[],
-  current: TextStreamData[],
-  maxEntries = DEFAULT_TRANSCRIPTION_HISTORY_SIZE
-): TextStreamData[] {
-  if (current.length === 0) return [];
-
-  const byStreamId = new Map(previous.map((entry) => [entry.streamInfo.id, entry]));
-  current.forEach((entry) => byStreamId.set(entry.streamInfo.id, entry));
-
-  return Array.from(byStreamId.values())
-    .sort((a, b) => a.streamInfo.timestamp - b.streamInfo.timestamp)
-    .slice(-maxEntries);
-}
diff --git a/tests/chat-message-filter.test.mjs b/tests/chat-message-filter.test.mjs
index 35dd6b4fa..9f3074b99 100644
--- a/tests/chat-message-filter.test.mjs
+++ b/tests/chat-message-filter.test.mjs
@@ -3,19 +3,6 @@ import { readFile } from 'node:fs/promises';
 import { test } from 'node:test';
 
 const { isRenderableChatMessage } = await import('../lib/chat-message-filter.ts');
-const { mergeTranscriptionHistory } = await import('../lib/transcription-history.ts');
-
-function transcription(id, segmentId, timestamp, text) {
-  return {
-    text,
-    participantInfo: { identity: 'frontdesk-agent' },
-    streamInfo: {
-      id,
-      timestamp,
-      attributes: { 'lk.segment_id': segmentId },
-    },
-  };
-}
 
 test('chat message filter hides empty transcription messages', () => {
   assert.equal(isRenderableChatMessage({ message: '' }), false);
@@ -32,25 +19,3 @@ test('chat message hook filters empty merged messages before sorting', async ()
   assert.match(source, /isRenderableChatMessage/);
   assert.match(source, /\.filter\(isRenderableChatMessage\)/);
 });
-
-test('transcription history preserves a tool preamble replaced by the final stream', () => {
-  const preamble = transcription('stream-preamble', 'speech-1', 100, '我查一下。');
-  const final = transcription('stream-final', 'speech-1', 200, '已经设置好了。');
-
-  const history = mergeTranscriptionHistory([preamble], [final]);
-
-  assert.deepEqual(
-    history.map(({ text }) => text),
-    ['我查一下。', '已经设置好了。']
-  );
-});
-
-test('transcription history updates partial text without duplicating one stream', () => {
-  const partial = transcription('stream-1', 'speech-1', 100, '我查');
-  const completed = transcription('stream-1', 'speech-1', 100, '我查一下。');
-
-  const history = mergeTranscriptionHistory([partial], [completed]);
-
-  assert.equal(history.length, 1);
-  assert.equal(history[0].text, '我查一下。');
-});

From 5fe07cf36aec2b82e16cc8aab43560bf6962ddb6 Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Wed, 26 Aug 2026 19:56:53 +0800
Subject: [PATCH 11/12] Revert "Revert "fix: preserve tool preambles in chat
 history""

This reverts commit aec6ac6d7719ce023c6b3c5e590f69b714db7bb9.
---
 hooks/useChatMessages.ts           | 12 +++++++---
 lib/transcription-history.ts       | 23 ++++++++++++++++++++
 tests/chat-message-filter.test.mjs | 35 ++++++++++++++++++++++++++++++
 3 files changed, 67 insertions(+), 3 deletions(-)
 create mode 100644 lib/transcription-history.ts

diff --git a/hooks/useChatMessages.ts b/hooks/useChatMessages.ts
index e9b75f5d0..ed6be0d0a 100644
--- a/hooks/useChatMessages.ts
+++ b/hooks/useChatMessages.ts
@@ -1,4 +1,4 @@
-import { useMemo } from 'react';
+import { useEffect, useMemo, useState } from 'react';
 import { Room } from 'livekit-client';
 import {
   type ReceivedChatMessage,
@@ -9,6 +9,7 @@ import {
 } from '@livekit/components-react';
 import { type AppConfig } from '@/app-config';
 import { isRenderableChatMessage } from '@/lib/chat-message-filter';
+import { mergeTranscriptionHistory } from '@/lib/transcription-history';
 
 function transcriptionToChatMessage(
   textStream: TextStreamData,
@@ -76,10 +77,15 @@ export function useChatMessages(
   const chat = useChat();
   const room = useRoomContext();
   const transcriptions: TextStreamData[] = useTranscriptions();
+  const [transcriptionHistory, setTranscriptionHistory] = useState<TextStreamData[]>([]);
+
+  useEffect(() => {
+    setTranscriptionHistory((previous) => mergeTranscriptionHistory(previous, transcriptions));
+  }, [transcriptions]);
 
   const mergedTranscriptions = useMemo(() => {
     // 处理转录消息
-    const transcriptionMessages = transcriptions.map((transcription) =>
+    const transcriptionMessages = transcriptionHistory.map((transcription) =>
       transcriptionToChatMessage(transcription, room, config)
     );
 
@@ -112,7 +118,7 @@ export function useChatMessages(
     ].filter(isRenderableChatMessage);
 
     return merged.sort((a, b) => a.timestamp - b.timestamp);
-  }, [transcriptions, chat.chatMessages, room, config]);
+  }, [transcriptionHistory, chat.chatMessages, room, config]);
 
   return mergedTranscriptions;
 }
diff --git a/lib/transcription-history.ts b/lib/transcription-history.ts
new file mode 100644
index 000000000..fce673cf9
--- /dev/null
+++ b/lib/transcription-history.ts
@@ -0,0 +1,23 @@
+import { type TextStreamData } from '@livekit/components-react';
+
+const DEFAULT_TRANSCRIPTION_HISTORY_SIZE = 100;
+
+/**
+ * Keep completed text streams even when LiveKit replaces the current entry for
+ * the same speech segment. A tool preamble and its final answer can share one
+ * segment id while still arriving as distinct text streams.
+ */
+export function mergeTranscriptionHistory(
+  previous: TextStreamData[],
+  current: TextStreamData[],
+  maxEntries = DEFAULT_TRANSCRIPTION_HISTORY_SIZE
+): TextStreamData[] {
+  if (current.length === 0) return [];
+
+  const byStreamId = new Map(previous.map((entry) => [entry.streamInfo.id, entry]));
+  current.forEach((entry) => byStreamId.set(entry.streamInfo.id, entry));
+
+  return Array.from(byStreamId.values())
+    .sort((a, b) => a.streamInfo.timestamp - b.streamInfo.timestamp)
+    .slice(-maxEntries);
+}
diff --git a/tests/chat-message-filter.test.mjs b/tests/chat-message-filter.test.mjs
index 9f3074b99..35dd6b4fa 100644
--- a/tests/chat-message-filter.test.mjs
+++ b/tests/chat-message-filter.test.mjs
@@ -3,6 +3,19 @@ import { readFile } from 'node:fs/promises';
 import { test } from 'node:test';
 
 const { isRenderableChatMessage } = await import('../lib/chat-message-filter.ts');
+const { mergeTranscriptionHistory } = await import('../lib/transcription-history.ts');
+
+function transcription(id, segmentId, timestamp, text) {
+  return {
+    text,
+    participantInfo: { identity: 'frontdesk-agent' },
+    streamInfo: {
+      id,
+      timestamp,
+      attributes: { 'lk.segment_id': segmentId },
+    },
+  };
+}
 
 test('chat message filter hides empty transcription messages', () => {
   assert.equal(isRenderableChatMessage({ message: '' }), false);
@@ -19,3 +32,25 @@ test('chat message hook filters empty merged messages before sorting', async ()
   assert.match(source, /isRenderableChatMessage/);
   assert.match(source, /\.filter\(isRenderableChatMessage\)/);
 });
+
+test('transcription history preserves a tool preamble replaced by the final stream', () => {
+  const preamble = transcription('stream-preamble', 'speech-1', 100, '我查一下。');
+  const final = transcription('stream-final', 'speech-1', 200, '已经设置好了。');
+
+  const history = mergeTranscriptionHistory([preamble], [final]);
+
+  assert.deepEqual(
+    history.map(({ text }) => text),
+    ['我查一下。', '已经设置好了。']
+  );
+});
+
+test('transcription history updates partial text without duplicating one stream', () => {
+  const partial = transcription('stream-1', 'speech-1', 100, '我查');
+  const completed = transcription('stream-1', 'speech-1', 100, '我查一下。');
+
+  const history = mergeTranscriptionHistory([partial], [completed]);
+
+  assert.equal(history.length, 1);
+  assert.equal(history[0].text, '我查一下。');
+});

From fed9950ed8d732924c2deffd001106e9b864b4f2 Mon Sep 17 00:00:00 2001
From: Codex <codex@local>
Date: Wed, 26 Aug 2026 19:56:53 +0800
Subject: [PATCH 12/12] Revert "Revert "fix: retain transcription history
 across reconnects""

This reverts commit 40a51d5c514d11827311dc48eca86d05624d3708.
---
 hooks/useChatMessages.ts           | 4 ++++
 lib/transcription-history.ts       | 2 +-
 tests/chat-message-filter.test.mjs | 8 ++++++++
 3 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/hooks/useChatMessages.ts b/hooks/useChatMessages.ts
index ed6be0d0a..7125e826d 100644
--- a/hooks/useChatMessages.ts
+++ b/hooks/useChatMessages.ts
@@ -79,6 +79,10 @@ export function useChatMessages(
   const transcriptions: TextStreamData[] = useTranscriptions();
   const [transcriptionHistory, setTranscriptionHistory] = useState<TextStreamData[]>([]);
 
+  useEffect(() => {
+    setTranscriptionHistory([]);
+  }, [room.name]);
+
   useEffect(() => {
     setTranscriptionHistory((previous) => mergeTranscriptionHistory(previous, transcriptions));
   }, [transcriptions]);
diff --git a/lib/transcription-history.ts b/lib/transcription-history.ts
index fce673cf9..2a16a8da8 100644
--- a/lib/transcription-history.ts
+++ b/lib/transcription-history.ts
@@ -12,7 +12,7 @@ export function mergeTranscriptionHistory(
   current: TextStreamData[],
   maxEntries = DEFAULT_TRANSCRIPTION_HISTORY_SIZE
 ): TextStreamData[] {
-  if (current.length === 0) return [];
+  if (current.length === 0) return previous;
 
   const byStreamId = new Map(previous.map((entry) => [entry.streamInfo.id, entry]));
   current.forEach((entry) => byStreamId.set(entry.streamInfo.id, entry));
diff --git a/tests/chat-message-filter.test.mjs b/tests/chat-message-filter.test.mjs
index 35dd6b4fa..632df3fcd 100644
--- a/tests/chat-message-filter.test.mjs
+++ b/tests/chat-message-filter.test.mjs
@@ -54,3 +54,11 @@ test('transcription history updates partial text without duplicating one stream'
   assert.equal(history.length, 1);
   assert.equal(history[0].text, '我查一下。');
 });
+
+test('transcription history survives a transient empty snapshot', () => {
+  const preamble = transcription('stream-1', 'speech-1', 100, '我查一下。');
+
+  const history = mergeTranscriptionHistory([preamble], []);
+
+  assert.deepEqual(history, [preamble]);
+});