// the find
yincongcyincong/MuseBot
supports Telegram, Discord, Slack, Lark（飞书），钉钉, 企业微信, QQ, 微信, compatible with various LLMs including OpenAI, Gemini, DeepSeek, Doubao, and OpenRouter. It offers intelligent conversation, image generation, video creation, and more. Works seamlessly in both private chats and group settings.
MuseBot is a Go service that wires messaging platforms — Telegram, Discord, Slack, and several Chinese platforms (WeChat, DingTalk, Lark, QQ) — to a swappable LLM backend. It handles text, image generation, voice, and video in a single deployable binary, with SQLite or MySQL for conversation history. The target audience is developers who want a self-hosted AI bot across platforms they already use.
The platform breadth is the real selling point: covering both Western and Chinese platforms in one Go binary is rare and genuinely useful for teams that straddle both worlds. MCP-to-function-call translation is a smart design — plug in any MCP server and MuseBot exposes it as a native tool call to the LLM without per-platform glue code. Per-user token budgeting is built in from the start, not bolted on, and Prometheus metrics with a bundled Grafana dashboard means you're not flying blind. SQLite as the default DB with no external dependencies makes local setup genuinely fast.
Private TLS certificates are committed to the repo under admin/shell/certs/ — those keys should never have been there, and even as 'example' certs this sets a bad precedent for anyone cloning and deploying. The admin panel is a completely separate Go binary with its own DB connection and its own auth layer, which doubles operational complexity and creates state sync risk if both processes touch the same tables concurrently. The WeChat and QQ integrations rely on HTTP callback mode from Tencent platforms that are notoriously unstable for developers — they require a public HTTPS endpoint, break on API changes, and have almost no community documentation in English. Running SQLite under concurrent LLM requests from multiple users will hit write contention fast; the switch to MySQL is documented but not the default, so most people will hit this wall before they realize it.