Manage a local AI bot team like a chat app
You can line up multiple AI bots side by side, similar to a messenger contact list, and assign different tasks based on the models, computers, and connected apps each bot uses. OpenMausBot follows the format of closed personal agent services like Grok Bot or Meta Muse, but it is an open-source alternative that keeps all data and processing on your local machine. It reuses your existing Claude, Codex, or Grok accounts, and you can make direct decisions via approval cards when a bot attempts risky actions.
Local-first design and agent integration
The app manages all agent processes through a small harness server running on 127.0.0.1. Chat logs, keys, and events are stored in the ~/.openmausbot folder rather than the cloud, giving users data sovereignty. Bots run via the claude, codex, or grok CLI installed on your computer, reusing existing logins and subscriptions without a separate proxy or new account. You can also specify a custom CLI binary in the settings screen to use different builds or wrappers as the engine.
How bots work with their hands
Each bot can operate a cloud Linux desktop, an isolated local VM, or directly manipulate your computer if platform safety boundaries are authenticated. On macOS and Ubuntu Xorg, host control is enabled after explicit opt-in, while Ubuntu Wayland remains disabled until safety issues are resolved. Composio connects over 500 apps, such as Gmail, Slack, and GitHub, allowing bots to use them as tools. Risky actions like shell commands or file modifications are displayed as inline cards in the chat screen, where users can allow or deny them.
Team composition and automation features
You can separate bots by work, personal, or project categories using channels, with each channel having its own unique transcript and work folder. You can import team compositions as markdown files from BotMRR to install bots, channels, and playbooks all at once. If you configure the Jev decision model, it selects the appropriate bot to respond to unmentioned messages within a few hundred milliseconds. Additionally, you can have bots automatically perform tasks at set times or when app events occur through schedule and webhook triggers.
Installation and requirements
Release builds are provided for macOS (Apple silicon & Intel), Windows (x64), and Ubuntu 24.04 (x64), with Ubuntu being a beta version. To run from source, you need Node 24 or higher and pnpm, and at least one agent CLI among claude, codex, or grok must be installed and logged in. You can run the harness server and the app separately using the pnpm dev:server and pnpm dev commands. Desktop app packaging is performed with the pnpm package:mac, pnpm package:win, and pnpm package:linux commands.
Pre-use checks
This project is in its early stages, but the core loop from messaging to computer use is functional. Ubuntu is a beta version with feature limitations, so you should check the support scope according to your desktop environment. Sensitive features like host control require explicit opt-in and are disabled in Wayland environments due to safety gates. It is distributed under the Apache 2.0 license, but the enterprise/ folder uses a source-available license, so you must delete that folder to get a fully open-source edition. Using cloud computers or voice synthesis features may require separate accounts and costs for third-party services like Boat or ElevenLabs.
