Automated AI news aggregation, analysis & knowledge management. RSS → AI Pipeline → Daily Digest → Web Dashboard.
📖 Product Guide (Chinese): docs/README_zh.md
| Dimension | Detail |
|---|---|
| Positioning | AI Intelligence Platform — News / Knowledge / Graph / Research |
| Pipeline | RSS → Classify → Summarize → Score → Daily Report + Web UI |
| SSOT | Knowledge Card (YAML) — all modules read from cards |
| Web | FastAPI on :8765, 8 pages, API-driven UI |
| Database | SQLite (WAL mode), 4 tables: entities / relationships / articles / reports |
| Entities | 123 knowledge cards, 8 types (model, company, tech, concept, product, person, methodology, event) |
| Relations | 592 edges in the knowledge graph |
| RSS Sources | 16 sources (14 active), RSS + HTML parsing |
| Tests | 249 passed, 0 failures |
# 1. Install dependencies
uv sync --frozen --all-groups
npm ci
# 2. Configure API keys
cp .env.example .env
# Edit .env: set DEEPSEEK_API_KEY=sk-xxx (DeepSeek is the default AI provider)
# 3. Start the web server
uv run uvicorn src.api.api:app --reload --port 8765
# 4. Run a one-off daily pipeline
uv run python pipeline.py --fetch-direct
# 5. Run tests
uv run pytest tests/ -q # 249 passedOpen http://127.0.0.1:8765 for the Dashboard.
See docs/TOOLCHAIN.md for pinned runtime versions and the upgrade policy.
ai-news/
├── pipeline.py # CLI entry point — 9-stage pipeline
├── collector.py # RSS collector — runs hourly
├── config.yaml # RSS sources, categories, pipeline config
├── prompts/ # AI prompt templates
├── templates/ # Report templates (daily/weekly/monthly)
├── data/
│ ├── ai_news.db # SQLite database
│ ├── inbox.jsonl # Raw fetched articles
│ └── knowledge/ # Knowledge cards (YAML), organized by type
├── src/
│ ├── interfaces/ # Shared layer — i18n
│ ├── engine/ # AI engine + data processing (18 modules)
│ │ ├── ai_client.py # Multi-provider AI (DeepSeek/Kimi) + Embedding registry
│ │ ├── database.py # SQLite CRUD + search + stats
│ │ ├── fetcher.py # RSS + HTML fetching (16 sources)
│ │ ├── classifier.py # AI article classification
│ │ ├── summarizer.py # AI summarization with context
│ │ ├── scorer.py # AI 1-5★ scoring
│ │ ├── reporter.py # Markdown daily report generation
│ │ ├── knowledge.py # Card loading / Jaccard + semantic matching
│ │ ├── embeddings.py # Semantic embeddings (SiliconFlow BGE, 1024d)
│ │ ├── concept_miner.py # Automated concept discovery
│ │ ├── trend_reporter.py # Weekly/monthly trend analysis
│ │ ├── research_engine.py # Deep research engine
│ │ └── ...
│ ├── frontend/ # HTML page generators (14 modules)
│ │ ├── dashboard.py # Dashboard — stats + top stories
│ │ ├── library.py # Knowledge library with category navigation
│ │ ├── entity_page.py # Entity detail view
│ │ ├── kg_d3.py # D3.js interactive knowledge graph
│ │ ├── kg_3d.py # Three.js 3D knowledge graph
│ │ ├── timeline_renderer.py # AI history timeline
│ │ ├── research_page.py # Research workbench
│ │ └── ...
│ └── api/ # FastAPI application
│ └── api.py # 15+ REST endpoints + static file serving
└── tests/ # 249 tests across 14 test modules
inbox.jsonl
→ Classify (AI categorization)
→ Concept Miner (discover new entities)
→ Knowledge Match (card matching, semantic + Jaccard)
→ Summarize (AI summary with historical context)
→ Score (AI 1-5★ rating)
→ Report (Markdown daily digest)
→ DB Sync (SQLite)
→ Dashboard + Library + Graph + Timeline refresh
| Route | Content |
|---|---|
/ |
Dashboard — stats, top stories, health |
/library |
Knowledge library — 123 cards, 8 types, semantic search |
/graph |
D3.js interactive knowledge graph (2D) |
/graph3d |
Three.js 3D knowledge graph |
/timeline |
AI industry timeline |
/events |
Milestone events timeline |
/reports |
Report browser (daily/weekly/monthly) |
/research |
Deep research assistant |
/entity/{id} |
Entity detail page |
/api/entities |
Entity list with type filter |
/api/articles |
Article list with score filter |
/api/search?q=&semantic=true |
Full-text + semantic hybrid search |
/api/stats |
Database statistics |
/api/health |
System health check |
/api/research |
POST — AI-powered deep research |
/api/embeddings/status |
Embedding readiness status |
/api/embeddings/rebuild |
POST — rebuild all embeddings |
| Task | Frequency | Time |
|---|---|---|
| Collector | Hourly | from 8:00 |
| Daily Pipeline | Daily | 9:00 |
| Weekly Report | Sunday | 10:00 |
| Monthly Report | 1st of month | 10:00 |
Windows Task Scheduler .bat / .vbs files included.
| Layer | Choice | Notes |
|---|---|---|
| Language | Python 3.13 | |
| AI Engine | DeepSeek + Kimi | OpenAI-compatible dual backends |
| Embeddings | SiliconFlow BGE (1024d) | Pluggable: openai/kimi/local |
| Web | FastAPI + Uvicorn | API-driven HTML shells |
| Database | SQLite (WAL mode) | Zero-config, foreign keys |
| Frontend | Vanilla JS + fetch API | No framework dependency |
| Package | uv | Fast Python package manager |
| Scheduling | Windows Task Scheduler |
- SSOT: Knowledge Cards (YAML) are the single source of truth
- API-driven UI: HTML shells + vanilla JS
fetch()— no SSR, no React/Vue - File-driven: No ORM — direct SQLite + YAML + JSONL
- Context budget: Files <300 lines, one feature per conversation
- importance ≠ score: Manual curation vs. AI real-time scoring
- i18n: Full Chinese/English toggle, 100+ translation keys
All documentation is organized under docs/:
| Document | Audience | Description |
|---|---|---|
docs/ARCHITECTURE.md |
Developers | System architecture deep-dive |
docs/ROADMAP.md |
Developers | Development roadmap |
docs/DECISIONS.md |
Developers | Architecture Decision Records (ADR) |
docs/DESIGN_SYSTEM.md |
Frontend developers | Frontend design constitution |
docs/ENGINEERING.md |
Developers | Engineering principles |
docs/KNOWLEDGE_CARD_SCHEMA.md |
Developers | Knowledge card format specification |
docs/OPENSPEC.md |
Developers | Original design specification |
docs/INFORMATION_ARCHITECTURE.md |
Developers | Information architecture |
docs/CHANGELOG.md |
Developers | Release changelog |
docs/README_zh.md |
Users | Product guide (Chinese) |
README.md |
Everyone | Project overview & quick start |
make verify该命令会依次重建静态页面、运行完整测试、检查 FastAPI 路由，并检查或临时启动开发服务器。未安装 Make 时可直接运行等价命令：
uv run python tools/verify_frontend.py视觉检查截图统一输出到 output/playwright/：
make visual-check
# 未安装 Make 时：
uv run python tools/capture_frontend.py脚本会在配置 DOUBAO_API_KEY、DASHSCOPE_API_KEY 或 OPENAI_API_KEY 后自动调用 vision skill，并将分析结果保存为同目录下的 .analysis.txt 文件。
MIT — see LICENSE