Nine AI agents debate your holdings to deliver a decision memo in five minutes
Select your held stocks, and technical, news, and fundamental analysts debate their findings, review risks, and send a final judgment from a portfolio manager to your messenger. This self-hosted AI stock monitoring system covers A-shares, Hong Kong, and US markets. Deploy it with Docker and connect your preferred LLM.
The analysis flow created by the agent team
PanWatch operates on a team of nine agents based on TradingAgents. When you select a specific held stock, analysts in each role collect data and exchange opinions. After a risk review, the portfolio manager writes the final decision memo. This process completes within 3 to 5 minutes, and you can view the entire reasoning process in the messenger connected to the app. You can also set schedules for automatic execution after the previous day's close, during trading hours, or at market close.
Simulation and alerts for practical trading
It includes a paper trading feature that lets you practice signal-based trading without real funds.
- Multi-market support Search and manage Chinese A-shares, Hong Kong, and US stocks and ETFs within the same workflow.
- Technical indicator integration View MACD, RSI, KDJ, Bollinger Bands, and candle patterns at a glance, with support and resistance lines calculated automatically.
- Mobile optimization Installed as a PWA, it functions like a native app from the mobile home screen.
Self-hosting and model selection freedom
PanWatch is deployed as a Docker image and allows you to freely connect OpenAI-compatible APIs or local models via Ollama. This means you can experiment with various LLMs while maintaining data sovereignty. During initial setup, you must explicitly specify the browser time zone, which affects background scheduling and notification quiet hours. If you do not need the Playwright-based screenshot feature, you can skip its installation via an environment variable to save resources.
Things to check before adoption
This tool requires Python 3.11 or higher, Node.js 24.14.0, and pnpm 9.15.9. When deploying with Docker, network access is required because it downloads the Chromium headless shell on first startup. For ETFs, fund holdings, net asset value (NAV), fees, and tracking error are not provided as structured data, so there are different limitations compared to stock analysis. Additionally, unpublished holiday data blocks automatic execution, so you must update the bundled calendar data when a new year begins. It is released under the MIT license, allowing free use.
