Prying open WeChat’s local vault for terminal-dwelling AI agents

What it does WeChat CLI reads your locally stored WeChat data—chats, contacts, favorites, group members—without ever touching a remote server. It extracts the SQLCipher encryption key by scanning the running WeChat process’s memory, then decrypts the SQLite database on-the-fly to answer queries. Output is JSON by default, though you can ask for Markdown or plain text if you’re still human.

The interesting bit The project is essentially a polished wrapper around wechat-decrypt, but its real trick is positioning itself as an AI agent tool rather than a manual forensics utility. It expects to be driven by Claude Code or OpenClaw, offering commands like new-messages that maintain incremental state for automation loops.

Key highlights

• Scans process memory and re-signs the WeChat macOS app with required entitlements if needed to grab decryption keys.
• All decryption happens locally with AES-256-CBC; no data leaves the machine.
• Eleven query commands including search by message type (text, image, file, call), chat statistics, and time-range exports.
• Bundles a native macOS arm64 binary via npm; Windows and Linux rely on pip or source installs.
• Explicitly read-only: it queries the local SQLite files but does not send, modify, or delete messages.

Caveats

• Version compatibility is narrow: the README specifies WeChat for Mac ≤ 4.1.8.100, and newer versions may break key extraction.
• macOS setup can require sudo, Full Disk Access for your terminal, and potentially re-signing WeChat’s entitlements, which the authors note may disrupt WeChat’s auto-updater.
• The core decryption and parsing logic comes from the upstream wechat-decrypt project; this repo adds the CLI interface and AI-agent packaging.

Verdict Worth a look if you want your local AI coding agent to have context from your actual WeChat conversations without cloud-privacy gymnastics. Skip it if you’re on an unsupported WeChat version or unwilling to tamper with app entitlements.