What if you could keep Grok Bot's team-of-agents interface and choose a different model and computer for each job? That's the idea behind MausBot, the independent project on GitHub as OpenMausBot. In Andrew Warner's walkthrough, creator Milind Soni spins up a Linux desktop, opens Blender on his own laptop with permission, schedules a task from a calendar, and starts a group of bots on a website QA pass. The useful question is where that flexibility saves work, and which costs and responsibilities it puts back on you.
MausBot is a free desktop workspace for running several AI agents with different models, tools and schedules. It is not xAI's Grok Bot made open source, and it does not include unlimited inference or cloud hours. Start with one bot and a task you can check. Add separate computers, schedules or teammates only when the task calls for them.
Watch Andrew Warner's Full Walkthrough
A Bot, a Model and a Computer Are Different Things
Think of a bot as a recurring job with its own instructions, conversation and tool access. The agent engine runs the work, and available models depend on the engine and provider you connect. Soni shows different models assigned to different bots, plus separate threads so one interview or project does not get mixed into the next. Start with an engine you already use, such as Claude Code or Codex. Judge any replacement on a completed job and its review cost, not on how many names appear in the model picker.
The computer setting is separate. Leave visual desktop control off for tasks that don't need it, use an isolated local or cloud desktop for browser work, or opt in to control of a supported host for something like the Blender demo. An isolated desktop does not isolate accounts you intentionally sign into. Use a test account and keep production credentials out of a trial. The approval documentation also makes an important distinction: MausBot displays requests supported by the engine, but the inbox is not a universal barrier before every action.
What Is Free, and What Keeps Running?
The project's licensing notes put the core under Apache 2.0, with a separate source-available license for the enterprise directory. The official pricing page lists the desktop app at $0 and individual managed Cloud plans from $15/month as of 18 November 2026. Starter lists two bots working at once and 100 monthly computer-hours. Model usage is separate; subscriptions, provider rate limits and infrastructure capacity still apply.
| Where MausBot runs | What to expect |
|---|---|
| Your desktop | Start without a MausBot hosting bill. The app must be running when a local routine is due. |
| Your always-on host | Work can continue after you shut your laptop. You manage the server, uptime, updates and access. |
| MausBot Cloud | MausBot hosts the workspace. Check plan allowances, team access and what the model provider bills separately. |
A remote Linux desktop is a place for a bot to click and type; it isn't necessarily the server that runs the bot's schedule. The routines guide explains the local-versus-hosted difference. The phone app is a companion to a running host, and Tailscale, shown in the interview, is a remote connectivity option, not a hosting service. Also, local-first describes where workspace data is stored: using a hosted model can still send it prompts and relevant context. See the official FAQ before connecting sensitive data.
What the Live Demos Actually Show
- Parallel website QA. At 10:39, Soni dispatches a QA group of about ten bots to his own site. For your first test, give each bot a different normal user journey on authorized staging. Require a reproducible finding and have a person verify it. The video doesn't measure coverage, accuracy or savings.
- Charts from a large dataset. At 12:54, Soni describes pulling roughly one million rows and generating charts. That's one workflow demonstration, not a guaranteed file limit or evidence that every row fits in a model's context. Ask for the analysis code, row counts and filters before trusting the chart.
- Sharing work with a team. The workspace walkthrough shows people sharing bots and discussing work alongside them. Give the team a named owner for each routine and keep public deployment separate from analysis. Soni shows publication options, and the docs describe publishing to Netlify through your own account. Review the artifact and its data before it goes live.
Moving From Grok Bot or Importing a Template
The official migration guide explains what the import carries: bot instructions, skills, memory notes and supported routines. It does not bring along chats, computer files or sign-ins. Scheduled routines arrive paused. Move one bot first, verify the settings, reconnect only the necessary apps, and turn off its old schedule before enabling the new one to avoid duplicate work.
The separate community templates repository offers reusable team setups and playbooks. Treat an imported skill like a script someone else wrote: review its instructions, included scripts and requested connections before enabling it.
Start With One Bot and One Reviewable Result
- Get the desktop build for your system from the official download page or GitHub Releases. The install guide covers macOS, Windows and Ubuntu.
- Connect one supported engine and create one bot with a dedicated working folder. Leave unrelated accounts and host control disconnected. If visual browsing is needed, use a test desktop.
- Run the check below manually before scheduling it. Compare it with a human pass: did it cover the three paths, avoid unauthorized changes and save time after review? Track model and computer cost per report you accept.
Check three staging paths
A read-only release check with a human reviewer.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | Meet MausBot |
| 00:54 | A separate computer for each agent |
| 01:48 | Optional control of the host computer |
| 02:24 | Bring your own model |
| 03:09 | Choosing models for different jobs |
| 04:48 | Multiple bots and separate threads |
| 05:51 | A familiar chat interface |
| 06:18 | Scheduling from the calendar |
| 07:39 | Tasks and the approval inbox |
| 08:24 | Mobile companion |
| 09:00 | Running agents around the clock |
| 10:21 | Managing agents in parallel |
| 10:39 | Parallel browser QA demo |
| 11:42 | The agent organization view |
| 12:54 | Analyzing and charting a large dataset |
| 14:15 | Publishing an app online |
| 14:51 | Local, cloud and team workspaces |
| 16:12 | Humans and agents in the same workspace |
| 17:15 | Connecting devices with Tailscale |
| 18:36 | Moving from Grok Bot |
| 19:30 | Sharing and importing templates |
| 20:15 | Where to try MausBot |
Resources and Source Notes
The embedded interview and supplied transcript establish what Andrew Warner and Milind Soni demonstrate. Zapier is identified as the video's presenter. The official site, repository and documentation establish current product details. Follow the MausBot project or Milind Soni for release announcements. JQ AI SYSTEMS did not independently reproduce the interview's demos or audit the product. Evidence cutoff: 18 November 2026.