Akasha-WeChat

Akasha-WeChat is a bridge tool that connects a WeChat personal account to an AI language model, turning a secondary WeChat account into a chatbot that can respond to messages using models like DeepSeek, Kimi, Claude, or any other provider supported by AstrBot. The project is written in Python and runs on Windows, where the desktop WeChat application must be installed and logged in.

The system works through a chain of components. WeFlow, a third-party Windows tool, hooks into the local WeChat client and exposes an API that streams incoming messages in real time without polling. The bridge script in this repository picks up those messages, optionally buffers several messages together before forwarding them, and converts them into the OneBot v11 format, which is a standardized protocol for chat bot communication. AstrBot, a separate AI agent platform, receives the converted messages over a WebSocket connection, passes them through its plugin pipeline to call the chosen AI model, and sends back a reply. The bridge then delivers that reply into WeChat by simulating UI interactions on the desktop application.

A few practical features are included. Image messages are automatically downloaded, described by a vision-capable AI model (such as llava or Kimi), and the description is injected as text before the message reaches the main AI. Group chat behavior can be set to three modes: reply only when the bot is mentioned, reply to all messages, or process messages in batches. A self-reply guard prevents the AI from responding to its own messages in a loop. The message buffer combines multiple rapid messages from the same conversation before sending them to the AI, reducing the number of API calls.

A web control panel is available at a local address and lets you start and stop the bridge, view live logs, check connection status, and edit all configuration options directly in the browser without modifying files by hand.

The author notes in the README that they are a high school student preparing for college entrance exams and may not respond to questions promptly. The project uses the MIT license.