You commented bots, so here is the link, and then the part the reel did not have time for: what people actually give these bots to do, what runs while you sleep, what costs money and the two places where the reel's one-liners need a footnote.
One bot per job. Not one chatbot.
// Build this with your AI agent
Build this with your AI agent
Want a shortcut? Copy the prompt below into any AI assistant. It asks what you use and what job you want first, then walks you through one step at a time. It buys nothing and never asks for a key.
Read this guide: https://codewithnishant.dev/free/free-grok-bot-alternative/ and the official pages it links to. Help me set up one bot in the free open source Grok Bot alternative the guide describes: 1. Ask which computer I use (Mac, Windows or Ubuntu), which of Claude, Codex or Grok I am already signed into, and the one job I want my first bot to do. 2. Walk me through ONE step at a time: install, first bot, a first task, then the approval setting. 3. Make sure I switch the bot to Ask for approval and set Sending on your behalf to Ask every time before it touches any connected app. 4. If I want a morning schedule, tell me plainly that the app and my computer must be on when it runs. Never ask me to paste a password or API key into this chat. Do not buy, subscribe or install anything without asking me first.
01. The link and the 10 minute setup
The short answer: the tool is called OpenMausBot (it also goes by MausBot, and it used to be called OpenGrokBot). It is an Apache 2.0 desktop app that looks like a chat app: every contact in the sidebar is a bot, and each bot has its own model, memory, computer and apps. It is an independent project, inspired by Grok Bot, and says it has no tie to xAI.
- The repo: github.com/milind-soni/OpenMausBot. On 17 November 2026 the GitHub API showed about 24.5K stars, Apache-2.0, created 11 August 2026 (the day Grok Bot went into beta), latest release v0.1.166 published that day.
- Download: latest release. Mac (Apple silicon and Intel, signed and notarized .dmg), Windows x64 installer, Ubuntu 24.04 .deb or AppImage (beta).
- Windows: the installer is not code-signed yet, so SmartScreen shows "unknown publisher". The README says More info, then Run anyway, once you have confirmed you got it from the official releases page.
The setup, from the install docs and first bot docs:
- Install the app and open it.
- Make sure at least one agent command line tool is installed and signed in:
claude,codexorgrok. Run it once in a terminal and finish its sign-in, restart the app, and it shows up in the model picker. The docs say authentication belongs to that tool: if you get a 401, fix the tool's login, not the app. - Your first bot already exists. A new install starts with a bot called MausBot, your Chief of Staff, whose first message lists what is left to set up.
- Make your own: New or share in the sidebar, then New bot. Use Set me up under its greeting to describe the job in one sentence, and give it a dedicated working folder if it touches files.
- Give it a small first task with a clear outcome. The docs' own example: review this project, run the tests and explain the three highest-impact issues, without changing files yet.
Prefer a terminal or a server? With Node 24 or newer, npx openmausbot runs it, and npx openmausbot serve is the server route for a VPS or an always-on computer (README).
02. The use-case map: one bot per job
The short answer: inbox, research, competitor watching, lead finding, SEO and engineering are the jobs with a written setup you can copy today. Video editing is an idea, not a ready-made bot. I labelled where each one comes from, because most of the public posts list features and very few show a bot doing a real job.
| Job | What the bot does | Needs | Where this comes from |
|---|---|---|---|
| Morning inbox summary | Every weekday at 9, summarises your inbox into the chat | Gmail, a schedule | The README's own example of a trigger |
| Inbox follow-up | Finds sales threads where your team owes the next step, drafts the follow-up, sends nothing on its own | Gmail, optional HubSpot | Playbook "Recover Unanswered Sales Emails" in the project's teams repo |
| Lead finder | Searches the communities you approve, scores leads against your ideal customer, drafts a non-spammy reply every morning | Reddit, optional Google Sheets, a schedule | Playbook "Find Qualified Reddit Leads" |
| Competitor watch | Checks the competitor pages you pick, keeps a dated change log, says watch, respond or ignore | A computer, a schedule, optional Google Sheets | Playbook "Watch Competitor Moves" |
| SEO team | Four bots turn search exports into a ranked list of fixes and content briefs | Your exports, optional Google Sheets | Playbook "Find SEO Growth Opportunities" |
| Engineering | Four bots take a change from intent to a verified release | A working folder | Playbook "Engineering Team" |
| Video editing | Not a documented setup. You would write the instructions yourself and give the bot a computer | A computer, your own prompts | An idea from the reel. I found no README or docs page for it |
Read the labels. The playbooks come from BotMRR, which describes itself as an independent directory. Its repo sits under the same GitHub account as the app, and one playbook lists the project itself as author, so I would not call it neutral. The repo calls the playbooks "reconstructions", not the original creators' bots and not proof of results. I have not run any of them. They are useful because each one is a plain Markdown file that lays out the bots, the apps and the safety rules.
How a playbook installs
From the README: on a playbook page, choose Add to OpenMausBot. The app opens a review screen before it creates anything. You can also import the same .md file under Teams, then Import, or paste its public GitHub link. Connections stay off until you approve them, routines arrive paused, and packages never carry credentials, conversations or memory.
The maker's own examples
The tool's website has a Solutions page with three bigger jobs: customer calls handled by agents, recruiting groundwork where a recruiter approves each submission, and paperwork and machine checks for manufacturing. Those are the maker's pitch, not customer results. The animated chat on the homepage is labelled an illustrated example.
Moving your Grok Bot bots over
If you already built bots on Grok Bot, the Move from Grok Bot docs describe an import. Grok Bot has no export button, so the app gives you a message to paste into each bot. The bot writes its own setup and the app reads it. It brings instructions, skills (switched on unless a safety check holds one back), scheduled routines (paused), memory and the list of apps. It does not bring chat history, files, sign-ins or group chats. Nothing changes in Grok Bot.
03. Morning schedules, and the catch
The short answer: yes, any bot can run on a schedule, but on the free desktop app it only runs while the app and your computer are on. The reel's "while you sleep" needs one of the two fixes below.
- What you can set: once, on chosen weekdays, or every 5 to 1,440 minutes. You can ask the bot in chat ("every weekday at 9, send me a summary of my inbox") and it sets the schedule up itself, with a card that has Edit, Pause and Undo (README).
- Where results land: one message in the bot's chat. A run with nothing to report stays quiet and shows Nothing new. If a run needs you, it asks in that chat (routines docs).
- The catch: the schedule docs say there is no outside always-on scheduler. The schedule runs inside the app on your computer, so a sleeping computer or a quit app runs nothing, and the app cannot wake a sleeping Mac. While plugged in it keeps the computer awake for the hour before a due routine.
- Fix 1, if your laptop is often off: in the routine's Advanced options, turn on Also run when OpenMausBot starts. Missed runs are covered by one run when you open the app.
- Fix 2, for true always-on: run the server on a machine that stays on, or take a MausBot Cloud plan (section 07). The pricing page says Cloud runs automations around the clock with your laptop off.
- Beyond schedules: webhooks start a bot from an outside request, and app events start it when something happens in a connected app. The desktop keeps up to 500 app events a day, and Gmail is checked every 15 minutes (connected apps docs).
04. A computer for each bot: free local VM vs paid cloud
The short answer: the free option is the Local VM, and it is a Docker or Podman container on your own machine. The paid option is a hosted cloud computer from a third party. Both give the bot a Linux desktop you can watch.
| Option | Cost | What it needs | Good for |
|---|---|---|---|
| This computer | Free | An opt-in on Mac or Ubuntu Xorg. On Windows you choose This computer | Local apps and files you explicitly approve. No isolation |
| Local VM | Free | Docker Desktop, Docker Engine or Podman, plus memory and disk. Each desktop defaults to 4 GiB of memory and 2 CPUs. On Windows, Podman is preferred | Repeatable browser and desktop work that should not run on your own desktop |
| Boat cloud computer | Paid after its trial, billed by Boat | A Boat API key in App Settings | An always-available isolated Linux desktop |
| Your own VPS | Your server's cost | An x86_64 Linux server with Docker, reached over SSH | Infrastructure you already own |
- Local VM setup: open Settings, then Desktops & browser, choose Set up next to Private desktops, prepare the managed desktop image, then give a bot Private desktop as its computer and create it from the bot's Screen panel (Local VM docs).
- Boat bills while it is on. The app puts it to sleep after 10 minutes of nothing using it, and wakes it in a few seconds when a bot needs it (cloud computer docs). The README says it plainly: a cloud computer may incur charges.
- Watch it, take over: every computer shows a live screen preview, and a bot that needs you to sign in to a website asks with a card instead of asking for the password in chat. Starting a preview does not give the bot control.
- Heads up: the Local VM does not work from the published server Docker image, because that image has no Docker inside. Use cloud desktops there.
- Speed: computer use is slow across this whole field. A hands-on review from August 2026 said the same about this tool and about Grok Bot. That review is three months old, so judge it yourself.
05. Allow and Deny cards, and the default to change
The short answer: the cards are real, but they are not what you get out of the box. Per the approval levels doc, every new bot and thread starts at Full access.
| Level | What it does |
|---|---|
| Ask for approval | Requests approval for commands and file changes. This is where you get the Allow and Deny cards |
| Auto-accept edits | Approves file edits itself. Other actions can still ask. Offered where the provider has this mode |
| Approve for me | Uses the provider's automatic review to approve routine actions and ask about the rest |
| Full access (the default) | Runs the provider in its most permissive mode, including potentially destructive or sensitive work. Real questions and missing credentials still reach you |
- Switch your first bots to Ask for approval. Pick it in the bot's settings and it applies to the bot and all its threads at once, or in the composer for one thread. That is my advice for the first week, not a rule from the docs.
- Then set Permissions, Sending on your behalf, to Ask every time. The docs say connected-app sends (email, messages, posts, payments) follow this setting at every level. Send without asking, or no setting at all, sends with no daily limit (connected apps docs).
- Remember what Full access skips. At Full access a bot's own schedule changes, team setup and skill writing apply without a second approval. Each shows as one line in the chat with Undo, but deleting a bot has no Undo.
Approvals add friction at volume. A review from August called them the right default but expected them to get annoying. Start strict, then loosen one bot at a time.
06. Connect Gmail, Slack and Notion
The short answer: in the official desktop app you connect an app from the sidebar and sign in through your browser. No key to find.
- Open Connected apps from the sidebar.
- Search for Gmail, GitHub, Slack, Notion or another service, and select Connect.
- Finish the OAuth sign-in in the browser window that opens.
- Back in the app, ask a bot to use it. You can add several accounts of the same app, each with its own label such as work or personal.
The connected apps docs say the signed desktop app registers with the maker's managed service and you do not need to paste a Composio key. The README's credentials table lists a Composio project key as an optional extra, and a saved key wins over the managed service. The count of apps differs by page: the README says 500+, the docs say hundreds, so I would not quote a number.
07. What costs money
| Piece | Cost | Source |
|---|---|---|
| The app | Free, open source, Apache 2.0 | README |
| The AI itself | Your own Claude, ChatGPT or Grok plan, or an API key. Provider fees and limits apply | Pricing page |
| Local VM | Free, uses your own memory and disk | Local VM docs |
| Boat cloud computer | Third party, paid after its trial | README, cloud computer docs |
| MausBot Cloud (optional) | Starter $15, Pro $49, Max $99 a month in USD, AI usage not included | Pricing page, read 17 Nov 2026 |
| Voice replies and calls (optional) | Your own ElevenLabs, Fish Audio or xAI key | README |
Cloud plans on the pricing page: Starter keeps unlimited bots with 5 awake and 2 working at once, about 100 computer-hours a month. Pro is $49 with no caps on awake bots and 300 computer-hours. Max is $99 with 800. These are the maker's numbers on one day, so recheck before you pay.
08. Your first week: one bot, one job
- Day 1. Install, sign in to one agent tool, make one bot, switch it to Ask for approval.
- Day 2. Pick one job that repeats, like an inbox summary or a competitor check. Describe it in one sentence with Set me up.
- Day 3. Connect only the one app that job needs. Set Sending on your behalf to Ask every time.
- Day 4. Run it by hand twice and read every output.
- Day 5. Turn it into a schedule, with Also run when OpenMausBot starts on if your laptop sleeps.
- Day 6. Add the second bot only if the first one earned it.
- Day 7. Decide: stay on the free app, add a Local VM, or look at a Cloud plan.
New to the idea? Start with what an AI agent actually is. A different route to a personal agent is in the Hermes Agent setup guide, and the Meta side of this story is in the OpenMuse guide.
09. Honest limits
- It is young. Release numbers went from v0.1.161 on 15 November to v0.1.165 on 16 November and v0.1.166 on 17 November, and the repo shows over 1,100 open issues on GitHub. The README itself says "early but real" and lists rough edges: hosted and mobile connectivity is still being built, and webhooks use a local receiver on your computer.
- Keys on the command line route. For the terminal and server install, the docs say API keys are saved as plaintext in a private
config.json. The packaged desktop app encrypts your AI provider keys through the operating system instead. Some keys still sit in plaintext there, such as ones given to one engine and MCP header values. - Local-first is not local models. Claude, Codex and Grok still send your prompts to their providers. Your transcripts and keys live in
~/.openmausboton your machine. - Subscription terms. Bots run the official command line tools you are signed into. I did not find a ruling on whether your provider's plan terms cover that use, so read your plan's terms before you build anything big on it.
- One reviewer's August test (link): tagging one bot from another bot's chat did not work. Re-test it yourself.
- Platform gaps: Ubuntu is a beta, bot control of the computer is disabled on Ubuntu Wayland, and the Windows installer is unsigned.
- No crypto. The project says it has no token, and any coin using the Maus name is not theirs.
FAQ
What is the free open source Grok Bot alternative?
It is OpenMausBot (also called MausBot, formerly OpenGrokBot): an Apache 2.0 desktop app where every contact in a chat sidebar is an AI bot with its own model, memory, computer and apps. It runs on the Claude, Codex or Grok command line tool you are already signed into. It is an independent project and says it has no affiliation with xAI.
Is it really free?
The app is free and open source. You still pay for the AI itself through the Claude, ChatGPT or Grok plan or API key you already have. A local virtual machine on your own laptop costs nothing extra but needs Docker or Podman. The Boat cloud computer is a third-party paid service after its trial, and the optional MausBot Cloud plans start at $15 a month.
Do my bots run while my computer is asleep?
Not on the free desktop app. Its own docs say the schedule runs inside the app on your computer, so a sleeping computer or a quit app runs nothing. For always-on bots you either run the server on a machine that stays on, or pay for a MausBot Cloud plan.
Will a bot ask before it does something risky?
Only if you set it up that way. The docs say every new bot starts at Full access, where the provider runs in its most permissive mode. Switch the bot to Ask for approval to get Allow and Deny cards, and set Permissions, Sending on your behalf, to Ask every time so emails and messages wait for you.
Do I need a Composio key to connect Gmail and Slack?
Not in the official packaged desktop app. The docs say it registers with the maker's managed connected-apps service, so you connect an app and finish sign-in in your browser. A source build, or anyone who wants their own Composio project, pastes a project key instead.
Can I move my Grok Bot bots over?
Yes, the docs describe a Move from Grok Bot import. It brings instructions, skills, scheduled routines (paused), memory and the list of apps. It does not bring chat history, files, sign-ins or group chats.
What I checked, and what I did not
On 17 November 2026 I read the OpenMausBot README, its installation, first bot, Local VM, routines and connected apps pages, the approval levels and schedule docs, the pricing page, the teams repo and the GitHub API for stars and releases.
I did not install or run the app, any playbook or any schedule, so nothing here is a result of mine. The use-case table says where each job comes from. Playbooks are reconstructions, the Solutions page is the maker's pitch, and the video editing row is an idea. The app ships almost daily, so if a menu looks different on your machine, the linked docs are the source of truth. This is an independent project and is not affiliated with xAI, Meta, OpenAI or Manus, and neither am I.
One bot, one job, approvals on, and read the first ten outputs yourself. Then let it run.
// Free newsletter
I send out guides like this every week
Real setups, real sources, no hype. Drop your email and I'll send you the next one.