feat: AI-Radar Cloudflare MVP core

.gitignore

@@ -0,0 +1,6 @@
+node_modules/
+.dev.vars
+.env
+.wrangler/
+dist/
+*.log

README.md

@@ -1,29 +1,105 @@
 # AI-Radar
 
-AI-Radar is a zero-cost-first, Cloudflare-native workflow for discovering AI coding-plan, API, token, free-credit, limited-time offer, and price-change information.
+AI-Radar is a zero-cost-first, Cloudflare-native workflow for discovering AI model launches, API availability, Coding Plan, Token, free-credit, limited-time offer, and price-change information.
 
-## Product direction
-
-- Cloud-first: runs on Cloudflare Workers without a local always-on machine.
-- Zero-cost-first: the default architecture must stay within free tiers; paid APIs are not required.
-- Low-token monitoring: HTTP/RSS/GitHub/page-change detection runs continuously; AI is only used after a meaningful change is found.
-- Telegram-first delivery: P1 items are pushed immediately; P2/P3 are summarized into a daily report at 12:30 Asia/Shanghai.
-- Public-ready: the public web view is separated from the admin/control plane from the beginning.
-
-## MVP pipeline
+## Discovery Radar v2 architecture
 
 ```text
-Sources
-  -> collectors (RSS / Web / GitHub / future X adapter)
-  -> change detection + deduplication
-  -> deterministic rules
-  -> optional Workers AI enrichment
-  -> D1
-  -> P1 Telegram push
-  -> P2/P3 daily report
-  -> public web dashboard
+Known sources (RSS / Web / GitHub)
+            │
+            ├──────────────┐
+            │              │
+            ↓              ↓
+      normal monitoring   discovery feeds
+                         (OpenRouter / Hugging Face /
+                          Artificial Analysis)
+            │              │
+            └──────┬───────┘
+                   ↓
+        change detection + source-entry dedup
+                   ↓
+          deterministic event classifier
+                   ↓
+       optional Workers AI enrichment
+                   ↓
+              Cloudflare D1
+          ↙            ↓             ↘
+ P1 Telegram     dynamic source pool   P2/P3 daily
+ immediate       temporary → candidate 12:30 Asia/Shanghai
+                 → core
 ```
 
-## Status
+The discovery layer is designed to find previously unknown models/vendors instead of relying only on a manually maintained seed list. A newly discovered model page can be added automatically as a temporary source for 30 days. If it produces meaningful follow-up signals it is promoted to candidate, then to core after repeated useful hits. The active temporary/candidate pool is capped to avoid uncontrolled source growth.
+
+## Event types
+
+AI-Radar now treats model intelligence as first-class data rather than requiring a discount or free-credit signal:
+
+- `new_model`
+- `model_api_available`
+- `model_open_source`
+- `model_benchmark`
+- `free_credit`
+- `limited_offer`
+- `price_drop`
+- `price_change`
+- `new_plan`
+
+A clear new-model event has a deterministic P2 floor. New models with API availability, coding/agent relevance, or other high-value signals can be promoted to P1.
+
+## Zero-cost rule
+
+The default system does not require paid APIs or paid infrastructure. Discovery uses public web/API endpoints. Workers AI is enabled only after change detection, source-entry/global deduplication, and deterministic filtering. It is hard-capped at 50 calls per UTC day with at most 256 output tokens per call; quota exhaustion or any AI error falls back to the deterministic result.
+
+## Current collectors
+
+- RSS / Atom
+- Ordinary HTML pages with conditional requests and content hashing
+- GitHub repositories through public Atom feeds
+- OpenRouter public model catalog discovery
+- Hugging Face text-generation discovery set
+- Artificial Analysis model-catalog discovery
+- X adapter reserved; the zero-cost discovery layer is the current fallback for X-only announcements
+
+## Source tiers
+
+- `core`: stable long-term monitored source
+- `discovery`: aggregator/catalog used to find unknown models and vendors
+- `temporary`: auto-added page with a 30-day observation window
+- `candidate`: temporary source that produced a meaningful update; observation window extends to 90 days
+- candidate sources become `core` after three meaningful hits
+
+## Cloudflare bindings
+
+The Worker expects `DB` (D1), `AI` (Workers AI), and `ASSETS` (Workers Static Assets). Secrets: `ADMIN_TOKEN`, the dedicated AI-Radar `TELEGRAM_BOT_TOKEN`, and the dedicated group `TELEGRAM_CHAT_ID`. AI-Radar has no Telegram webhook or inbound command handler: the bot is outbound-only. Non-secret safety controls include `AI_DAILY_CALL_LIMIT` and `SOURCE_BATCH_SIZE`. Never commit secret values.
+
+## Initial setup / upgrade
+
+1. Create D1 database `ai-radar-db` if this is a fresh install.
+2. Replace `REPLACE_WITH_D1_DATABASE_ID` in `wrangler.jsonc` if needed.
+3. `npm install`
+4. `npm run db:migrate:remote`
+5. Configure secrets with `wrangler secret put ...`.
+6. `npm run check`
+7. `npm run deploy`
+
+Migration `0006_discovery_radar.sql` adds source lifecycle fields and seeds OpenRouter, Hugging Face, Artificial Analysis, plus Multiverse Computing's official resources page.
+
+Cloudflare Cron runs in UTC. `30 4 * * *` equals 12:30 Asia/Shanghai. The report window is the preceding 24 hours ending at 12:30 Beijing time.
+
+## API
+
+Public: `GET /api/health`, `GET /api/items`, `GET /daily/YYYY-MM-DD`, `GET /latest`.
+
+Admin (Bearer `ADMIN_TOKEN`): `GET/POST /api/admin/sources`, `POST /api/admin/run-harvest`, `POST /api/admin/run-daily`.
+
+## Delivery policy
+
+- P1: immediate Telegram push.
+- P2/P3: daily report at 12:30 Asia/Shanghai.
+- First fetch establishes a baseline and does not flood Telegram.
+- A new model does not get discarded merely because no discount/free-credit information is present.
+
+## Branch status
 
-Initial implementation in progress.
+Discovery Radar v2 is implemented on `feature/discovery-radar-v2`, based on `ai-radar-mvp`. `main` is unchanged.

docs/free-tier-capacity.md

@@ -0,0 +1,39 @@
+# AI-Radar 免费层容量评估（2026-08-31）
+
+## 官方上限
+
+- Workers Free：100,000 个动态请求/日；每次 HTTP/Cron 调用 10 ms CPU；每次调用 50 个外部子请求；每账号最多 5 个 Cron Trigger。静态资源请求免费且不限量。[1][2]
+- D1 Free：5,000,000 rows read/日、100,000 rows written/日、5 GB 账号总存储；单库 500 MB；单次 Worker 调用最多 50 条 D1 查询。[3][4]
+- Workers AI：每天 10,000 Neurons 免费额度；AI-Radar 另设 50 次/UTC 日、每次最多 256 输出 token 的应用层硬上限。[5]
+- 当前静态页面采用 Workers Static Assets 的 `assets.directory` + `ASSETS` binding + `run_worker_first`，符合官方当前全栈 Worker 推荐方式。[6]
+- `ADMIN_TOKEN`、AI-Radar 专用 Telegram Bot Token 与独立群 Chat ID 只使用加密 Secret；不使用 Topic，且不与 Hermes 共享 Bot 凭据。官方推荐 `wrangler secret put`，Secret 值设置后不在 Wrangler 或 Dashboard 回显。[7]
+
+## 估算假设
+
+- 每个信源平均每 60 分钟抓取一次；Cron 每 5 分钟运行，另有 1 次日报 Cron，即 289 次 Worker 调用/日。
+- 外部抓取 `fetch()` 是子请求，不计为入站 Worker 请求；每轮最多处理 10 个源，低于 50 子请求上限。
+- D1 保守按每次源抓取 3 rows read、4 rows written 估算；另加 1,000 次/日公开 API 访问，每次最多读取 100 条。
+- Workers AI 最坏按 50 次/日、约 4,300 Neurons/日估算，仍低于 10,000 免费额。
+- 页面 HTML/CSS/JS 由 Static Assets 直接提供，不计动态请求；`/api/items`、`/latest`、`/daily/*` 才计动态请求。
+
+## 结果
+
+| 信源数 | 源抓取/日 | Worker 基础调用/日 | D1 read/日（含 1,000 次 API 浏览） | D1 write/日 | AI 上限 | 结论 |
+|---:|---:|---:|---:|---:|---:|---|
+| 20 | 480 | 289 + 动态页面/API 访问 | 101,729 | 2,020 | 4,300 Neurons | **SAFE** |
+| 50 | 1,200 | 289 + 动态页面/API 访问 | 103,889 | 4,900 | 4,300 Neurons | **SAFE** |
+| 100 | 2,400 | 289 + 动态页面/API 访问 | 107,489 | 9,700 | 4,300 Neurons | **WARNING** |
+
+100 个源的 D1 与请求额度仍远低于免费上限，但平均每个 5 分钟周期会有约 8.3 个到期源，接近当前 `SOURCE_BATCH_SIZE=10`，并且复杂 HTML 的解析可能触碰 Free Cron 10 ms CPU 限制。因此 100 源不作为默认配置；必须先按源类型拆分频率、监控 `exceededCpu`，再决定是否扩容。
+
+当前 9 个源按实际 60/120/180 分钟混合频率约 160 次抓取/日，明显低于 20 源模型，属于 **SAFE**。默认配置不会自动开通付费服务；额度耗尽时应失败或降级，而不是产生额外调用。
+
+## Sources
+
+[1] https://developers.cloudflare.com/workers/platform/pricing
+[2] https://developers.cloudflare.com/workers/platform/limits
+[3] https://developers.cloudflare.com/d1/platform/pricing
+[4] https://developers.cloudflare.com/d1/platform/limits
+[5] https://developers.cloudflare.com/workers-ai/platform/pricing
+[6] https://developers.cloudflare.com/workers/static-assets
+[7] https://developers.cloudflare.com/workers/configuration/secrets

docs/x-zero-cost-options.md

@@ -0,0 +1,37 @@
+# X 零成本抓取候选方案（2026-08-31）
+
+## 结论
+
+本阶段不正式上线 X Collector。2026 年官方 X API 已转为按资源计费：读取 Post 为每条 $0.005，需购买 credits；“免费 xAI credits”是购买 X API credits 后的返利，不是免费读取额度。[8] 官方 changelog 仅说明经批准的 Public Utility Apps 可继续免费规模化访问、近期 Legacy Free 用户有一次性 $10 voucher，这两类都不是 AI-Radar 可长期默认依赖的公开免费层。[9]
+
+因此继续保留 `XCollector` 接口，首选未来出现的官方免费 read-only 计划；当前不购买 X API、不使用付费代理、不部署自动登录脚本。
+
+## 候选矩阵
+
+| 方案 | 免费 | 需要登录 | 稳定性 | Cloudflare 可直接运行 | 主要风险 | 推荐级别 |
+|---|---|---|---|---|---|---|
+| 官方 X API Pay-per-use | 否 | 开发者账号 | 高 | 是 | 明确按读取资源收费；违反长期 0 元目标 | **不采用** |
+| Public Utility Apps 特批 | 可能 | 需要申请与审核 | 高 | 是 | 非通用免费层，资格不确定 | **C：仅在官方批准后重评** |
+| RSSHub 自托管 Twitter 路由 | 软件免费 | 通常需要 X Cookie/账号 | 低至中 | 不能直接作为纯 Worker 路由运行 | Cookie 失效、接口字段轮换、账号风控；2026 仍有时间线空结果问题。[10] | **D** |
+| Nitter/公共实例 RSS | 表面免费 | 否 | 低 | 技术上可抓公开实例 | 上游 guest access、实例存活、法律与封禁风险；不能作为长期 SLA。[11] | **D** |
+| X Syndication 单 Tweet endpoint | 免费且免登录 | 否 | 中 | 是 | 只能在已知 Tweet ID 后读取，不能发现新内容 | **C：只作详情补全** |
+| 搜索引擎 `site:x.com` 线索 | 取决于搜索服务免费额度 | 否 | 低至中 | 取决于供应商 | 索引延迟、漏检、搜索 API 配额，不是完整时间线 | **C：只作发现补充** |
+| OpenCLI/浏览器登录态 | 工具本身免费 | 是 | 中 | 否（依赖本机浏览器会话） | 电脑关闭即停止，不符合脱离本机要求 | **不用于生产** |
+| 用户手工提交 X 链接 | 免费 | Worker 不需要登录 | 高 | 是 | 不是自动发现源 | **B：可作为补充入口** |
+
+## 后续触发条件
+
+仅在以下任一条件满足时重新评估并实现：
+
+1. X 官方提供明确、可长期使用的免费只读/search 配额；
+2. 获得 Public Utility Apps 免费资格且条款允许该监控场景；
+3. 出现无需账户 Cookie、可在 Cloudflare 运行、连续验证至少 30 天的合法稳定公共源。
+
+在此之前，AI-Radar 的 RSS/Web/GitHub 主链独立运行，X 不影响上线与日报。
+
+## Sources
+
+[8] https://docs.x.com/x-api/getting-started/pricing
+[9] https://docs.x.com/changelog
+[10] https://github.com/DIYgod/RSSHub/issues/22938
+[11] https://github.com/zedeus/nitter

migrations/0001_init.sql

@@ -0,0 +1,91 @@
+PRAGMA foreign_keys = ON;
+
+CREATE TABLE IF NOT EXISTS sources (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  name TEXT NOT NULL,
+  url TEXT NOT NULL,
+  type TEXT NOT NULL CHECK(type IN ('rss','web','github','x')),
+  trust_level TEXT NOT NULL DEFAULT 'C' CHECK(trust_level IN ('A','B','C','D')),
+  enabled INTEGER NOT NULL DEFAULT 1,
+  interval_minutes INTEGER NOT NULL DEFAULT 15,
+  config_json TEXT,
+  etag TEXT,
+  last_modified TEXT,
+  content_hash TEXT,
+  next_fetch_at TEXT,
+  last_fetch_at TEXT,
+  last_success_at TEXT,
+  failure_count INTEGER NOT NULL DEFAULT 0,
+  status TEXT NOT NULL DEFAULT 'new',
+  created_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE INDEX IF NOT EXISTS idx_sources_due ON sources(enabled, next_fetch_at);
+
+CREATE TABLE IF NOT EXISTS items (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  source_id INTEGER NOT NULL,
+  external_id TEXT,
+  fingerprint TEXT NOT NULL UNIQUE,
+  title TEXT NOT NULL,
+  summary TEXT,
+  url TEXT,
+  published_at TEXT,
+  discovered_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  kind TEXT NOT NULL DEFAULT 'other',
+  priority TEXT NOT NULL DEFAULT 'P3' CHECK(priority IN ('P1','P2','P3')),
+  score INTEGER NOT NULL DEFAULT 0,
+  source_confidence TEXT NOT NULL DEFAULT 'low',
+  verification_status TEXT NOT NULL DEFAULT 'unverified',
+  vendor TEXT,
+  product TEXT,
+  previous_price REAL,
+  current_price REAL,
+  currency TEXT,
+  expires_at TEXT,
+  raw_excerpt TEXT,
+  ai_enriched INTEGER NOT NULL DEFAULT 0,
+  pushed_at TEXT,
+  daily_report_date TEXT,
+  created_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  FOREIGN KEY(source_id) REFERENCES sources(id) ON DELETE CASCADE
+);
+
+CREATE INDEX IF NOT EXISTS idx_items_discovered ON items(discovered_at DESC);
+CREATE INDEX IF NOT EXISTS idx_items_priority ON items(priority, discovered_at DESC);
+CREATE INDEX IF NOT EXISTS idx_items_report ON items(daily_report_date, priority);
+
+CREATE TABLE IF NOT EXISTS price_history (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  item_id INTEGER,
+  vendor TEXT,
+  product TEXT,
+  price REAL NOT NULL,
+  currency TEXT,
+  observed_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  FOREIGN KEY(item_id) REFERENCES items(id) ON DELETE SET NULL
+);
+
+CREATE TABLE IF NOT EXISTS daily_reports (
+  report_date TEXT PRIMARY KEY,
+  window_start TEXT NOT NULL,
+  window_end TEXT NOT NULL,
+  item_count INTEGER NOT NULL DEFAULT 0,
+  html TEXT NOT NULL,
+  generated_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  telegram_pushed_at TEXT
+);
+
+CREATE TABLE IF NOT EXISTS fetch_logs (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  source_id INTEGER NOT NULL,
+  fetched_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  status_code INTEGER,
+  changed INTEGER NOT NULL DEFAULT 0,
+  duration_ms INTEGER,
+  error TEXT,
+  FOREIGN KEY(source_id) REFERENCES sources(id) ON DELETE CASCADE
+);
+
+CREATE INDEX IF NOT EXISTS idx_fetch_logs_source ON fetch_logs(source_id, fetched_at DESC);

migrations/0002_seed_sources.sql

@@ -0,0 +1,19 @@
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'AI Coding Deals','https://github.com/codertesla/ai-coding-deals','github','B',10,
+       '{"githubOwner":"codertesla","githubRepo":"ai-coding-deals","githubMode":"commits","githubBranch":"main"}',CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://github.com/codertesla/ai-coding-deals');
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'LLM Price Tracker','https://github.com/llerandi/llm-price-tracker','github','B',15,
+       '{"githubOwner":"llerandi","githubRepo":"llm-price-tracker","githubMode":"commits","githubBranch":"main"}',CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://github.com/llerandi/llm-price-tracker');
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'Free LLM API Resources','https://github.com/cheahjs/free-llm-api-resources','github','B',15,
+       '{"githubOwner":"cheahjs","githubRepo":"free-llm-api-resources","githubMode":"commits","githubBranch":"main"}',CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://github.com/cheahjs/free-llm-api-resources');
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'Coding Plan CN','https://github.com/xiaotiewinner/coding-plan','github','B',15,
+       '{"githubOwner":"xiaotiewinner","githubRepo":"coding-plan","githubMode":"commits","githubBranch":"main"}',CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://github.com/xiaotiewinner/coding-plan');

migrations/0003_track_source_entries.sql

@@ -0,0 +1,15 @@
+CREATE TABLE IF NOT EXISTS source_entries (
+  source_id INTEGER NOT NULL,
+  external_id TEXT NOT NULL,
+  first_seen_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  PRIMARY KEY (source_id, external_id),
+  FOREIGN KEY(source_id) REFERENCES sources(id) ON DELETE CASCADE
+);
+
+CREATE INDEX IF NOT EXISTS idx_source_entries_seen ON source_entries(first_seen_at);
+
+UPDATE sources
+SET enabled = 0,
+    status = 'disabled',
+    updated_at = CURRENT_TIMESTAMP
+WHERE url = 'https://github.com/cheahjs/free-llm-api-resources';

migrations/0004_production_sources.sql

@@ -0,0 +1,33 @@
+UPDATE sources
+SET interval_minutes = 60,
+    updated_at = CURRENT_TIMESTAMP
+WHERE url IN (
+  'https://github.com/codertesla/ai-coding-deals',
+  'https://github.com/llerandi/llm-price-tracker',
+  'https://github.com/xiaotiewinner/coding-plan'
+);
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'DeepSeek API Pricing','https://api-docs.deepseek.com/quick_start/pricing/','web','A',120,NULL,CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://api-docs.deepseek.com/quick_start/pricing/');
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'Zhipu BigModel Pricing','https://open.bigmodel.cn/pricing','web','A',120,NULL,CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://open.bigmodel.cn/pricing');
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'MiniMax API Pricing','https://platform.minimaxi.com/docs/guides/pricing-paygo','web','A',180,NULL,CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://platform.minimaxi.com/docs/guides/pricing-paygo');
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'Kimi API Pricing','https://platform.kimi.ai/docs/pricing/chat','web','A',180,NULL,CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://platform.kimi.ai/docs/pricing/chat');
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'OpenCode Releases','https://github.com/opencode-ai/opencode','github','A',60,
+       '{"githubOwner":"opencode-ai","githubRepo":"opencode","githubMode":"releases","githubBranch":"main"}',CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://github.com/opencode-ai/opencode');
+
+INSERT INTO sources(name,url,type,trust_level,interval_minutes,config_json,next_fetch_at)
+SELECT 'HN AI Free Credits','https://hnrss.org/newest?q=AI%20API%20free%20credits','rss','C',60,NULL,CURRENT_TIMESTAMP
+WHERE NOT EXISTS (SELECT 1 FROM sources WHERE url='https://hnrss.org/newest?q=AI%20API%20free%20credits');

migrations/0005_ai_daily_limit.sql

@@ -0,0 +1,5 @@
+CREATE TABLE IF NOT EXISTS ai_daily_usage (
+  usage_date TEXT PRIMARY KEY,
+  calls INTEGER NOT NULL DEFAULT 0,
+  updated_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP
+);