What It Does

OpenMausBot (short for MausBot, formerly OpenGrokBot) is an open source alternative to Grok Bot, and in many ways an open source counterpart to products like Meta Muse, OpenAI dots, and Manus Cue. The idea is simple but powerful: it packs several real agents into a single chat app. Every bot in the sidebar is a Claude or Codex instance running on your own machine, each with its own personality, model, cloud computer, and connected apps. You talk to them like contacts in a list, watch them work in real time, and approve sensitive actions by hand. The project sells itself on bring-your-own-agent and local-first: your models and compute come from the environment you already own, and you are not locked to any single vendor. In practice, you bring the API keys and the local horsepower, and it orchestrates that group of agents into a team that can research, code, triage, or simply chat alongside you.

The Problem It Solves

Closed AI companion products, Grok Bot and Meta Muse among them, keep agents locked inside the vendor's servers. Users cannot see how things run, cannot control their own data, and cannot plug their own workflows in. OpenMausBot hands that control back through local-first execution and bring-your-own-agent. Each bot runs on your machine and gets its own cloud computer, a Docker-backed virtual machine, so you keep the multi-agent experience without handing your chat history, credentials, and app connections to a third party. Compared with products that "just work but keep everything in the cloud," its pitch is exactly "my agents, my rules." It also ships signed or notarized installers across macOS (both Apple Silicon and Intel), Windows, and Ubuntu, so ordinary users can deploy it without building an environment from scratch, which matters because most people who want a private AI assistant are not willing to compile Rust toolchains first.

Language and Architecture

The main language is TypeScript with strict mode enabled, so the codebase is fairly disciplined. The front end uses React 19 and the desktop shell is built on Electron, which lets one codebase cover macOS, Windows, and Ubuntu. The agent runtime executes locally (Claude or Codex), and each bot gets a cloud computer, a virtual machine, provided by Docker. That means different bots are isolated at the environment level: one bot breaking its cloud computer does not affect another, and you can give a research bot and a coding bot completely separate sandboxes. Mobile-related hints involve Expo. The repository was created on August 11, 2026, and is actively iterated (the latest push landed on November 8, 2026). Distribution covers .dmg, .exe, .deb, and AppImage, with topics tagged ai-agents, claude-code, codex, computer-use, docker, electron, and self-hosted, which neatly summarizes the three trends it rides at once.

License

Apache-2.0.

It lands squarely on the 2026 trend of "AI companions and multi-agent desktop apps," and answers the privacy and control anxiety around closed AI products with a fully open, local-first stance, while also riding the traffic of products like Grok Bot. It gained roughly 1,955 stars that day, placing second on the list. More important, it tells a complete story of "multi-agent plus local execution plus auditability," matching exactly the mood of people who want an AI team but do not want to hand their house keys to a big company. The narrative is doing a lot of the work, but the execution is real enough that curious users can actually run it today rather than just admire the demo.

GitHub repository https://github.com/milind-soni/OpenMausBot, official site https://www.openmausbot.com.

If you want more on the "put AI agents on the desktop" idea, see our write-ups on agent-desktop and on building desktop apps with front-end frameworks like React.