GitHub - modelflare-ai/AnyModelBot: 🤖 Open-source, local-first alternative to Grok Bot, OpenAI dots and Claude Duet. Your own AI bots on any model: GPT, Claude, Gemini, Grok, DeepSeek, Kimi, GLM and 70+ more through one Modelflare key. Reach them in the app, WeChat or Feishu; they remember you, work in the background, run commands and use your computer. Pay per use.
github.com
🤖 Open-source, local-first alternative to Grok Bot, OpenAI dots and Claude Duet. Your own AI bots on any model: GPT, Claude, Gemini, Grok, DeepSeek, Kimi, GLM and 70+ more through one Modelflare ke...

Disclosure: I’m on the team that builds this.

AnyModelBot is a local-first, open-source (MIT) app for Windows, macOS and Linux (Docker image too). A “bot” is a persona with its own default model, its own workspace folder and its own memory. It can read and write files, run commands and operate the desktop, so most of the design work went into deciding what it may do without asking:

  • Three permission tiers per bot: read-only; workspace (default — writes only inside its own folder, anything outside pauses for approval); full access (unsandboxed, but recursive deletes still ask).
  • Approval cards: when an action steps outside what the bot may do on its own, the task waits for your OK. Computer use can be set to ask before every UI change.
  • Conversations, memory, files and keys stay on your device.

Demo: the bot tidies a folder and asks before moving files

Honest limits: it’s young (v0.1.x). Models are pay-per-use through a hosted model service, so prompts, file contents and tool results needed for an answer leave your machine to the model vendor you pick; the README is explicit that local-first isn’t an absolute security claim. Which features work depends on the model, OS and permissions you grant, and macOS computer use is still Beta.

Feedback on the permission model especially welcome — what would you want a desktop agent to always ask about?

  • anymodelbot@programming.devOP
    1·
    19 hours ago

    Fair question. AnyModelBot didn’t launch from zero. It’s built by the team behind Modelflare, an AI API gateway with 200,000+ registered developers. On launch day we announced it to that whole user base and featured it in Modelflare’s project showcase. That audience is exactly who AnyModelBot is for: developers who already use multiple AI models every day. So a big early spike in stars, and especially in downloads, was expected. Download numbers are the metric we watch most, and you can track the releases on GitHub. If you try it and find a bug or want a feature, please open an issue at https://github.com/modelflare-ai/AnyModelBot/issues and we’ll prioritize it.