WeChat Assistant — Kimi Plugin

A Kimi Desktop plugin that lets the AI operate your WeChat (Weixin) desktop client: read chats and messages, send replies, listen for new messages, and more — all through the official desktop client, no account hacking, no unofficial protocols.

⚠️ Disclaimer: Automating personal WeChat accounts may be subject to WeChat's Terms of Service. Use at your own risk. This plugin only drives the official client's UI like a human would — it does not inject code, hook processes, or use reverse-engineered protocols.

Features

Feature	Windows (WeChat 4.x)	Windows (WeChat 3.9.x)	Linux (WeChat 4.x)
Connection status	✅	✅	✅ experimental
List chats	✅	✅	✅ experimental
Read messages (incl. scroll-back history)	✅	✅	✅ experimental
Send text messages	✅	✅	✅ experimental
Listen for new messages (real-time polling)	✅	✅	✅ experimental
Send files	🔜 planned	✅	🔜 planned
@mentions in groups	🔜 planned	✅	🔜 planned
Friend requests (list/accept)	🔜 planned	✅	🔜 planned
Contact list	🔜 planned	✅	🔜 planned

How it works (architecture)

WeChat 4.x (Qt-based) does not expose its UI to Windows UI Automation — the window is accessibility-dark. This plugin uses three backends, auto-detected:

• win4 (Windows, WeChat 4.x): visual automation — captures the window with PrintWindow (works even when the window is behind others), reads it with OCR (RapidOCR, Chinese + Latin support), and acts with coordinate clicks + clipboard paste. Restores minimized windows automatically. Scrolls the message pane to read conversation history page by page.
• win (Windows, WeChat 3.9.x classic): UI Automation (uiautomation) — full feature set.
• linux (Linux, official WeChat 4.x): AT-SPI accessibility (pyatspi) — experimental.

Force a backend with --backend win|win4|linux or the WX_BACKEND environment variable.

Requirements

• Kimi Desktop
• WeChat desktop, logged in, main window open (it may be in the background):
  • Windows: WeChat 4.x (current version) or 3.9.x classic
  • Linux: official WeChat 4.x from linux.weixin.qq.com (experimental)
• Python dependencies:

  pip install uiautomation pywin32 comtypes rapidocr-onnxruntime pillow numpy opencv-python

  Linux additionally: sudo apt install python3-pyatspi (or your distro's equivalent — not available via pip).

Installation

1. Clone this repository to a permanent location (the plugin registers from this path — don't delete it):

   git clone <your-repo-url> wechat-assistant

2. Register it into Kimi's personal plugin market:

   kimi-daimon kimi-plugin register-personal /path/to/wechat-assistant

   (Or simply ask Kimi: "register the plugin at /path/to/wechat-assistant".)
3. In Kimi Desktop, open the Plugins page → 「个人」 (Personal) tab → click ＋ on WeChat Assistant to install it. No restart needed.
4. Open WeChat and log in (keep the main window open; it can stay in the background).

Usage

Just talk to Kimi naturally:

• "Do I have new WeChat messages?"
• "Read the latest messages from my chat with Anna"
• "Reply to Anna that I'm on my way" — Kimi will show you the exact text and ask for confirmation before sending.
• "Let me know if Carlos writes me in the next 10 minutes"
• "Read the conversation history with File Transfer going back 10 pages" — scroll-back via OCR.

All write operations (send, accept friend, etc.) require explicit confirmation — the CLI enforces --confirm yes.

CLI reference (what the agent calls)

python scripts/wx_cli.py status                     # connection status
python scripts/wx_cli.py chats                      # list chats
python scripts/wx_cli.py messages --chat "Anna" --count 30 [--pages 5]
python scripts/wx_cli.py send --chat "Anna" --text "Hi!" --confirm yes
python scripts/wx_cli.py listen --chat "Anna" --timeout 300
python scripts/wx_cli.py send-file --chat "Anna" --file report.pdf --confirm yes   # win 3.9
python scripts/wx_cli.py at --chat "Group" --who "Anna" --text "look" --confirm yes # win 3.9
python scripts/wx_cli.py contacts / friends-new / friend-accept                     # win 3.9
python scripts/wx_cli.py tree                       # debug: UI tree / raw OCR dump
python scripts/wx_cli.py screenshot --path out.png  # debug: window capture (win4)

All output is JSON on stdout. Errors are actionable: {"ok": false, "error": "...", "hint": "..."}.

Safety & privacy

• The agent will never send a message without your explicit confirmation (enforced by the CLI).
• Read messages stay local — the plugin has no network code at all.
• Windows session must be unlocked while sending (clicks need an active session).
• If WeChat is minimized to the system tray, click its tray icon to restore the main window before use.

Limitations

• WeChat Web is not supported (Tencent blocks web login server-side for most accounts — not a plugin limitation).
• win4 backend relies on OCR: very small text or emoji may be misread; WeChat UI layout changes may require adjustments.
• Images/videos/voice messages are detected as markers (e.g. [图片]), their content can't be read.
• Linux backend is experimental (core messaging only).

Project layout

├── kimi.plugin.json            # plugin manifest
├── skills/wechat-assistant/
│   └── SKILL.md                # agent instructions (Spanish)
└── scripts/
    ├── wx_cli.py               # CLI entry point (backend auto-detection)
    ├── wx_engine_win4.py       # WeChat 4.x visual/OCR backend (Windows)
    ├── wx_engine.py            # WeChat 3.9.x UIA backend (Windows)
    └── wx_engine_linux.py      # WeChat 4.x AT-SPI backend (Linux, experimental)

License

MIT