PanWatch is a self‑hosted AI‑driven stock‑monitoring assistant that couples a multi‑agent investment framework (TradingAgents) with real‑time data for A‑share, Hong Kong and US markets. When the user clicks the “🧠” icon on the holdings page, a pipeline of nine agents—technical, sentiment, news and fundamental analysts—engages in a bull‑bear debate, passes the output through risk review and produces a portfolio‑manager decision memo in 3‑5 minutes, which is then pushed to Telegram, WeChat, DingTalk or other configured channels. The default LLM is deepseek‑chat, costing roughly $0.05 per invocation, with budget controls. All portfolio data stays local; the app is deployed via Docker (docker run -d --name panwatch -p 8000:8000 -v panwatch_data:/app/data sunxiao0721/panwatch:latest) or Docker‑Compose, with optional Playwright‑based Chromium for UI screenshots. Technical analysis includes MA crossovers, MACD, RSI, KDJ, Bollinger Bands, volume‑price patterns and automatic support/resistance levels. Price alerts support AND/OR conditions, cooldowns and per‑rule notification routing. Optional OpenTelemetry tracing can be enabled by installing requirements‑otel.txt and setting OTEL_EXPORTER_OTLP_ENDPOINT. The backend uses FastAPI, SQLAlchemy, APScheduler and the OpenAI SDK; the frontend is React 18 with TypeScript, Tailwind CSS and shadcn/ui. Development requires Python 3.10+, Node 18+ and pnpm, with make dev‑api and make dev‑web commands.