🚀 Connect your communication app to AI in just one minute

This repository provides a Chat bot (Telegram, Discord, Slack, Lark（飞书），钉钉, 企业微信, QQ, 微信) that integrates with LLM API to provide AI-powered responses. The bot supports openai deepseek gemini LLMs, making interactions feel more natural and dynamic.
中文文档
Китайская документация

感谢阮一峰老师的weekly，很荣幸能登榜：https://github.com/ruanyf/weekly
感谢linux.do 社区，佬友们很给力，给一个我的介绍贴链接：https://linux.do/t/topic/1128110
Thanks to the Reddit community as well, even though a few of my subreddits got banned 😅: https://www.reddit.com/

• 🤖 AI Responses: Uses LLM API for chatbot replies.
• ⏳ Streaming Output: Sends responses in real-time to improve user experience.
• 🏗 Easy Deployment: Run locally or deploy to a cloud server.
• 👀 Identify Image: use image to communicate with LLM, see doc.
• 📄 Identify Document: send txt/pdf/csv/html documents and ask questions; chat media is stored locally per conversation, see doc.
• 🎺 Support Voice: use voice to communicate with LLM, see doc.
• 🐂 Function Call: transform mcp protocol to function call, see doc.
• 🧩 Skills: load local SKILL.md instructions into the unified /task workflow, see doc.
• 🛠️ Conversational System Operations: inside the /task workflow, manage runtime settings, MCP, skills, and cron jobs, and run commands and file operations inside safety boundaries, see doc.
• 🌊 RAG: Support Rag to fill context, see doc.
• 🌞 AdminPlatform: Use platform to manage MuseBot, see doc.
• 🌛 Register: With the service registration module, robot instances can be automatically registered to the registration center doc
• 🌈 Metrics: Support Metrics for monitoring, see doc.
• 🐶 Cron: Support Cron to trigger LLM, see doc.

easiest way to use: https://www.youtube.com/watch?v=4UHoKRMfNZg
deepseek: https://www.youtube.com/watch?v=kPtNdLjKVn0
gemini: https://www.youtube.com/watch?v=7mV9RYvdE6I
chatgpt: https://www.youtube.com/watch?v=G_DZYMvd5Ug

NOTE: All providers are accessed through the OpenAI-compatible API format (chat completions, images, audio). Aliyun (Qwen) and Vol (Doubao) are the recommended/fully-supported providers. Third-party relay services (OpenRouter / OrcaRouter / 302.AI) have been removed.

1. Clone the repository

   git clone https://github.com/yincongcyincong/MuseBot.git
   cd MuseBot

2. Install dependencies

    go mod tidy

3. Set up environment variables

    export TELEGRAM_BOT_TOKEN="your_telegram_bot_token"
    export DEEPSEEK_TOKEN="your_deepseek_api_key"

Run the bot locally:

 go run main.go -telegram_bot_token=telegram-bot-token -deepseek_token=deepseek-auth-token

Use docker

  docker pull jackyin0822/musebot:latest
  chmod 777 /home/user/data
  docker run -d -v /home/user/data:/app/data -e TELEGRAM_BOT_TOKEN="telegram-bot-token" -e DEEPSEEK_TOKEN="deepseek-auth-token" -p 36060:36060 -p 18080:18080 --name my-bot  jackyin0822/MuseBot:latest

 ALIYUN:
 docker pull crpi-i1dsvpjijxpgjgbv.cn-hangzhou.personal.cr.aliyuncs.com/jackyin0822/musebot:latest
 chmod 777 /home/user/data
  docker run -d -v /home/user/data:/app/data -e TELEGRAM_BOT_TOKEN="telegram-bot-token" -e DEEPSEEK_TOKEN="deepseek-auth-token" -p 36060:36060 -p 18080:18080 --name my-bot  crpi-i1dsvpjijxpgjgbv.cn-hangzhou.personal.cr.aliyuncs.com/jackyin0822/musebot:latest

command: (doc)[https://github.com/yincongcyincong/MuseBot/blob/main/static/doc/param_conf.md]

You can configure the bot via environment variables:

Here’s the English version of your environment variable table: If you use parameter. Please use lower letter and underscore. for example: ./MuseBot -telegram_bot_token=xxx

Instead of long flag/env lists, put your configuration in a json file and point the bot at it with -conf or the CONFIG env var:

go run main.go -conf=conf/examples/telegram_deepseek.json

# docker: mount the config file
docker run -d -v /home/user/data:/app/data \
  -v /home/user/mybot.json:/app/mybot.json \
  -e CONFIG="/app/mybot.json" \
  -p 36060:36060 -p 18080:18080 --name my-bot jackyin0822/musebot:latest

Two formats are supported:

Flat (recommended — top-level keys use the lowercased-underscore form of the env names):

{
  "telegram_bot_token": "your-telegram-bot-token",
  "type": "deepseek",
  "deepseek_token": "your-deepseek-api-key",
  "smart_mode": true,
  "admin_user_ids": {"123456789": true}
}

Sectioned (same layout as the config persisted by the admin panel: base/llm/audio/photo/video/rag/tools/register sections for fine-grained control):

{
  "base": {
    "telegram_bot_token": "your-telegram-bot-token",
    "type": "deepseek",
    "deepseek_token": "your-deepseek-api-key"
  },
  "llm": {
    "temperature": 0.7
  }
}

Precedence: env vars > config file > command-line flags > defaults. Keys missing from the file keep their defaults; restart to apply changes.

conf/examples/ ships a few classic configs ready to copy and edit:

If you are using a self-deployed llm, you can set CUSTOM_URL to route requests to your self-deployed llm.

support sqlite3 or mysql

if DB_TYPE is sqlite3, give a file path, such as ./data/telegram_bot.db if DB_TYPE is mysql, give a mysql link, such as root:admin@tcp(127.0.0.1:3306)/dbname?charset=utf8mb4&parseTime=True&loc=Local, database must be created.

choose a language for bot, English (en), Chinese (zh), Russian (ru).

llm_conf
photo_conf
video_conf
audio_conf advanced_conf (third-party models / smart_mode / task auto-approval)

clear all of your communication record with deepseek. this record use for helping deepseek to understand the context.

retry last question.

choose txt/photo/video/recognize model type.

choose txt/photo/video/recognize model.

show current model type and model.

calculate one user token usage.

/edit_photo will update you photo base on your description.

allows the bot to chat through /chat command in groups, without the bot being set as admin of the group.

multi agent communicate with each other!

The image runs two processes under supervisord: the main MuseBot service and the MuseBotAdmin management platform. Both restart automatically if they crash.

1. Build the Docker image

    docker build -t MuseBot .

2. Run the container

     docker run -d -v /home/user/xxx/data:/app/data \
       -e TELEGRAM_BOT_TOKEN="telegram-bot-token" \
       -e DEEPSEEK_TOKEN="deepseek-auth-token" \
       -p 36060:36060 -p 18080:18080 \
       --name my-bot MuseBot

3. Ports and volumes

   • /app/data: persist this volume — sqlite database, admin database (muse_bot_admin.db), media files and the auto-generated admin session key.
   • 36060: main service HTTP (webhook callbacks, config API, SSE chat).
   • 18080: admin web UI.
   • ffmpeg/ffprobe, nodejs/npm and libopus are pre-installed for voice, video and skill features.

4. First admin login The default account admin / admin is created on first start. After login the admin UI shows a change-password prompt (you can click "Later" to skip, but changing it soon is recommended). Log in at http://<host>:18080 (see the admin doc).

The built-in web_search task agent uses DuckDuckGo with Bing fallback by default. To use Parallel Search MCP instead, set WEB_SEARCH_PROVIDER=parallel before starting MuseBot, or pass -web_search_provider=parallel on a fresh installation. The setting is also saved as tools.web_search_provider in the bot configuration. Environment values override the saved setting.

Parallel searches use Fast mode at https://search.parallel.ai/mcp, without an account or API key. The task agent receives titles, source URLs and relevant excerpts in the existing search result format. This provider supports search; it does not add a page-fetch tool. Anonymous usage is subject to service limits, and errors are returned to the task without switching providers.

Use default to retain DuckDuckGo/Bing, or disabled to block built-in web search. Custom MCP servers and their tool settings are separate. Each /task or /mcp research run has its own opaque Parallel session identifier, reused by its agents and subsequent searches within that run.

Feel free to submit issues and pull requests to improve this bot. 🚀

telegram-group: https://t.me/+WtaMcDpaMOlhZTE1 , or you can have a try robot Guanwushan_bot. every body have 10000 token to try this bot, please give me a star!

QQ群：1031411708

MIT License © 2025 jack yin