An AI scout that scrolls Xianyu so you don't have to

What it does

ai-goofish-monitor runs persistent browser sessions via Playwright to crawl Xianyu (闲鱼), a second-hand marketplace. Users describe what they want in plain language, and the system spawns background jobs that scrape listings, download product images, and feed them to a multimodal LLM for judgment. Matches get piped to Telegram, WeChat Work, Bark, ntfy, or a generic webhook; everything else lands in a local SQLite database.

The interesting bit

The project treats LLM inference as a filtering layer in a scraping pipeline: the model looks at actual product photos and seller info to decide whether to alert you. It also handles the cat-and-mouse game of marketplace anti-bot measures by rotating accounts and proxies automatically when risk controls trigger.

Key highlights

• Natural-language task creation: describe the item, and an async job generates the AI evaluation criteria for you.
• Full web UI for task management, account binding, and log inspection; FastAPI backend with Vite frontend.
• Multi-account support using exported Chrome login-state JSON, with automatic account/proxy failover.
• Cron-based scheduling and granular filters (price range, free-shipping, province/city/district).
• Docker image ships with Chromium baked in; SQLite is the default store, though images and cookies remain on disk.

Caveats

• Requires an OpenAI-compatible API key and a vision-capable model, so it is not free to run at scale.
• You must supply valid Xianyu login cookies via a Chrome extension; without them, the scraper cannot authenticate.
• The README repeatedly warns that aggressive scraping violates Xianyu’s terms and may get accounts restricted.

Verdict

Worth a look if you are building marketplace automation or want to study how Playwright and LLMs can be glued into a stateful crawling system. Skip it if you are after a simple, no-API-cost price tracker.