OpenMausBot: your own team of AI bots

The problem

These release notes cover seven commits from our demo shop. I didn't write them. A bot did, on my own machine, in about a minute. And before it ran a single command, it asked me. Today: OpenMausBot.

What OpenMausBot is

OpenMausBot is an open source app, Apache licensed, that gives you a team of AI bots in a chat window. It started as an open alternative to Grok Bot from xAI. There is no extra subscription: each bot runs on an agent you already have, like Claude Code or Codex, and its data stays in a folder in your home directory.

The solution at a glance

A bot is three things. A soul: standing instructions and the rules it never breaks. A model, which you can switch per bot. And a computer to work on, your own or an isolated one. On top of that come approvals before risky steps, and routines that run on a schedule.

Install

Installing is a download from mausbot dot com, for Mac, Windows and Ubuntu. Or one npx command, with Node 24. It needs one connected agent. I use Claude Code, which was already logged in.

Building it

Let's build one. Plus, new bot. He gets a name, Remy, a title, release notes writer, and one line on what he is for. Under soul go the standing instructions: read the commits, group them into new, fixed and under the hood, write for users, and never push. The model is Claude Sonnet five, but every agent on this machine shows up here. Permissions stay on ask for approval, so nothing risky runs without me. And under routines, I give him a schedule: weekly release notes, every week, with one line on what to do. Create bot, and there he is: Remy, ready for his first job. I gave Remy one sentence: release notes for the demo shop, from master up to episode four. Before he touched anything, he asked. The card shows the exact command, a git clone of the public repo, and I allowed it once. A minute later: four new features, two fixes, four changes under the hood, and a note that this covers seven commits. I checked it against GitHub. It's right. The whole run cost twenty-nine cents on my Claude account.

Refactoring

One bot is nice. A team is better. Under templates, the community shares complete teams, an engineering team, a growth team, ready to load. I'll build my own. A second bot: Vera, release editor. Her soul says: turn release notes into a short announcement, and check every claim against the code first. Now a group chat. I call it release desk, and add Remy and Vera. Then a shared brief that every member reads on each turn: Remy writes the notes and hands them over, Vera checks them and writes the announcement. Remy is the lead, so he answers unless I mention someone else. Same request as before, now to the whole desk, plus a short announcement for the README. Remy writes the notes and, on his own, passes them to Vera with an at mention. Vera doesn't take his word for it. She clones the repo herself and asks before every command: the log, the diff, the form object, the job, the views. Nine approvals, all read only. Then her verdict: every claim checked against the code, all confirmed, nothing changed. And the announcement, ready for the README. Vera worked for almost three minutes. The whole desk cost fifty-eight cents.

Good to know

A few things to know. You pay your model provider, per run, so keep an eye on the cost. How much a bot can ask before it acts depends on the engine and the tool. API keys you enter are stored in plain text in config dot json. And routines only run while the app is running, or on a server you set up for it.

Wrap-up

That's OpenMausBot: your own bots, on your own machine, on accounts you already have. Give Remy a routine, and the release notes write themselves, every week.