关键字 · BOT

All guides / Open source

AnyModelBot: run a whole team of AI bots on your own computer, any model, pay per use

Open-source app for Windows, macOS and Linux: many AI bots, each with its own persona, model, workspace and memory, reachable from Telegram, Slack and Discord.

AnyModelBot is a team of AI bots that runs on Windows, macOS and Linux. Each bot has its own persona, default model, workspace and memory. You can chat with them in the app, or connect them to WeChat, Feishu, Telegram, Slack and Discord. Your conversations, memory, files and API keys stay on your own device.

The app itself is free and open source (MIT license), made by modelflare-ai, with around 4,400 stars on GitHub. Models can run through their Modelflare service (sign in and pay per use) or through your own official API keys.

01What you get

  1. As many bots as you need, each with its own persona, model, workspace and avatar. A persona is just text, and templates are only a starting point.
  2. Real work: bots read and write files, run commands in a sandbox, and send results back to the chat as files.
  3. Three permission levels per bot: Read-only, Workspace (the default: it writes only in its own folder and asks before writing outside it) and Full access (no sandbox, but it still asks before deleting a whole folder).
  4. Group chats: add a few bots to a group, @ whoever you want to answer, and bots can hand work to each other.
  5. Background tasks and rules: "Every morning at 8, check this price and tell me if it drops below 200" becomes a rule that checks on schedule and only pings you when it matters.
  6. Ready-made bots: Company Research, PPT Master (turns documents into editable slides), Film Studio (short videos), US Stock Research and more.
  7. Computer use plus image and video generation: computer control runs on the bundled Cua Driver, and you can make a bot ask before every click.
  8. Any model: GPT, Claude, Gemini, Grok, DeepSeek, Kimi, GLM and more in one picker, switchable per conversation.

02How to install and run it

Easiest: download the installer for your platform from the GitHub Releases page (.exe for Windows x64 / arm64, .dmg for macOS Apple silicon / Intel, .AppImage for Linux). On Windows, if a security warning appears the first time, click "More info", then "Run anyway". The macOS app is signed and notarized, so just drag it into Applications and open it.

You can check the download with: sha256sum -c SHA256SUMS --ignore-missing

Run it with Docker on a NAS or server (x64 and arm64):

  1. docker run -d --name amb -p 3600:3600 -v amb-data:/data ghcr.io/modelflare-ai/anymodel-bot
  2. docker exec amb cat /data/ui-token, then open http://<host>:3600/?token=<token>

Run from source (needs Node.js 22.13+ and Go 1.26.8+):

  1. git clone https://github.com/modelflare-ai/AnyModelBot.git
  2. cd AnyModelBot
  3. npm install
  4. npm run setup (fetches and builds the built-in gateway)
  5. npm start (local console: http://127.0.0.1:3600/)

03How to start using it

  1. Step 1: on first launch a browser opens so you can sign in (or sign up) to Modelflare; after you authorize, you're sent back to the app with no key to copy. Prefer not to use Modelflare? Add your own official API keys in Settings (OpenAI, Anthropic, Google Gemini, xAI, DeepSeek, Kimi, GLM, Qwen, MiniMax and Mistral are supported).
  2. Step 2: create a bot, write a short persona and pick a default model, or grab a ready-made bot from Discover.
  3. Step 3: to use it from your phone, go to Settings › Chat channels and add Telegram, Slack or Discord (paste a token), or scan a QR code for WeChat.
  4. Step 4: hand it longer jobs, for example by saying "do it in the background". You keep chatting and the result comes back to the same conversation.

04Good to know

  1. The app itself has no subscription fee; you pay for model usage.
  2. Through Modelflare, requests are forwarded to the provider you pick and Modelflare keeps request logs. With your own official keys, requests go straight to the provider without passing through Modelflare.
  3. What leaves your device: the prompts, files and tool results a model needs to answer, screenshots when you ask a bot to look at your screen, and messages that pass through chat platforms.
  4. The local console listens on 127.0.0.1 only. With Docker, don't expose port 3600 directly to the internet.
  5. Telegram, Slack and Discord are direct messages only for now, and computer use on macOS is in Beta.
  6. Features depend on the model you choose, your operating system and the permissions you grant.
  7. License: MIT.

05Links

Related guides