When Your Stock Watchlist Hires a Committee Instead of an Intern

What it does

PanWatch is a self-hosted web dashboard for tracking A-share, Hong Kong, and US equities. It aggregates real-time quotes, portfolio positions across multiple accounts, and standard technical indicators such as MACD, RSI, and KDJ. The standout feature is an integration with the TradingAgents framework, which chains technical, sentiment, news, and fundamental analysts into a structured bull-versus-bear debate, followed by risk-control review and a final decision pushed to Telegram, WeChat, DingTalk, or other channels.

The interesting bit

Instead of a single LLM summary, PanWatch orchestrates a full research pipeline: four analyst personas argue, a risk manager vets the consensus, and a portfolio manager drafts the brief. The README says this takes three to five minutes and costs roughly $0.05 per run on the default DeepSeek model, making it either admirably transparent or slightly alarming depending on your patience.

Key highlights

• Self-hosted FastAPI backend with React frontend; the app runs locally, though AI analysis relies on external OpenAI-compatible APIs.
• Covers A-shares, Hong Kong, and US markets with multi-account portfolio aggregation and a paper-trading simulator.
• PWA support means it installs like a native app on mobile, with dedicated mobile layouts shown in the repo screenshots.
• Alerting engine supports compound conditions—price, volume, RSI, and more—with AND/OR logic, cooldown timers, and per-rule notification routing.
• Scheduled agents deliver pre-market briefs, intraday monitoring, and post-market daily reports via APScheduler.

Caveats

• The Docker image bundles Playwright, so the container downloads Chromium on first boot; expect a slow, network-dependent startup unless you disable browser features with an environment variable.
• Each deep-analysis run takes three to five minutes and consumes LLM tokens; the README cites roughly $0.05 per run with DeepSeek, so frequent checks are not free.
• AI features require configuring an external API key; there is no bundled local model for offline inference.

Verdict

Worth a look if you are a retail investor in Chinese or US markets who wants algorithmic commentary without subscribing to a cloud dashboard. Skip it if you need sub-second trading signals or a fully offline, zero-cost stack.