Your WeChat, but with 12 AI interns and a surveillance streak

What it does

This is a WeChaty-based bot that logs into your personal WeChat via QR code, then routes incoming messages through a small army of AI providers—ChatGPT, DeepSeek, Claude, Ollama, Kimi, and roughly half a dozen Chinese models. It can auto-reply in whitelisted chats, dump your local WeChat cache (contacts, history, Moments) for analysis, and even pipe everything through the Pi coding agent so your WeChat becomes a voice interface for a terminal AI.

The interesting bit

The project doesn’t just call APIs; it layers a local data extraction engine (wb wx ... commands via OpenCLI’s wx-cli) underneath the chatbot, letting you run sentiment analysis or zombie-friend detection on your own social graph. There’s also a nascent Lark (Feishu) bridge, though it only does manual CLI operations—no real-time bot mode yet.

Key highlights

• Supports 12+ AI backends through a unified provider mechanism, including local Ollama and self-hosted Dify
• Can operate purely on local WeChat data without any LLM for privacy-conscious scraping
• Whitelist-based gating: only named friends and group chats can trigger auto-replies, with @bot required in groups
• Stores all intercepted messages to local JSONL for audit trail or downstream analysis
• Pi agent mode turns the bot into a coding assistant relay, not just a chat mirror

Caveats

• The README warns repeatedly that WeChat’s web protocol carries “风控和封号风险”—account bans are a real, documented risk
• padlocal protocol is unmaintained; users must hunt for “more stable” alternatives themselves
• Kimi integration hits rate limits hard; the maintainer notes users keep misconfiguring iFlytek keys too
• Lark IM integration is read/search/send only—no automatic event loop

Verdict

Worth a look if you need to automate WeChat operations and accept the封号 roulette. Skip it if you want production reliability or a clean API; this is duct tape over an unofficial protocol, not infrastructure.