ChatGPT in your group chat, for better or worse

What it does

wechatgpt is a Go-based relay: it logs into WeChat (via QR code) or Telegram, listens for messages containing a trigger keyword, and forwards them to OpenAI’s API. Replies come back through the same channel. It also handles image generation via DALL·E and vision analysis through GPT-4’s image model when prompted with the right keywords.

The interesting bit

The project sits at an awkward intersection of two brittle protocols: WeChat’s unofficial web login (which breaks often) and OpenAI’s rate-limited, paid API. The author acknowledges this fragility openly in the Q&A section, which is more honesty than most bridge projects manage. The Docker packaging and environment-variable configuration suggest it’s meant to run headless on cheap VPS boxes.

Key highlights

• Supports both WeChat (via openwechat library) and Telegram bots from the same binary
• Keyword-triggered responses, with optional username whitelisting for Telegram
• Multi-modal: GPT-4 vision for image description, DALL·E for image generation
• Configurable models via environment variables or YAML
• Docker Compose setup included

Caveats

• Telegram support is explicitly unmaintained (“暂不对 telegram 相关原始代码进行维护”)
• WeChat web login is notoriously unstable; QR code expiration and token file corruption are documented failure modes
• Requires a paid OpenAI API key; the README notes billing setup as a common stumbling block

Verdict

Useful if you need a quick, self-hosted ChatGPT proxy for Chinese social platforms and don’t mind babysitting the login flow. Skip it if you need reliability or official API support — the maintenance burden is visible in the README itself.