Some links on this page are affiliate links. They cost you nothing, and I only link tools I'd mention anyway.
Grok Bot is xAI's product for persistent AI "teammates" that keep working on a cloud computer after you close the app. That is the one-sentence version. The useful part is what the primary sources say, where they disagree, and what you should ask before handing any always-on agent your logins.
Everything here was read on 2026-10-23. Two caveats on method. I fetched the xAI docs pages through a tool that summarises pages, not as raw HTML, so quotes below are as that tool returned them. And I could not fetch x.ai/news (HTTP 403), so I have no first-hand copy of the launch announcement. Launch details come from a news write-up and are marked as secondary.
What is Grok Bot?
From the xAI docs overview (docs.x.ai/grok-bot/overview): Bots are persistent AI assistants, each with a name, a role and memory that builds over time. Each Bot works on a persistent cloud computer with a browser, a filesystem and a terminal. The page says closing the app, your laptop or your phone does not stop a background turn or a routine. Bots use connectors where they exist and computer use (driving a screen like a person) for everything else.
You can teach a Bot a multi-step workflow once, save it as a skill, and rerun it as a routine on a schedule. When a site blocks automation, expires a session or needs a human step, the Bot hands the step to you.
Secondary source: Unite.AI reports xAI announced Grok Bot as a beta on 11 August 2026, with multiple Bots running in parallel and coordinating in group chats.
Who can use it, and what does it cost?
The get-started page (docs.x.ai/grok-bot/get-started) says you need one of:
- any paid individual Cursor plan,
- the Cursor Teams plan, or
- an individual SuperGrok, SuperGrok Plus or SuperGrok Heavy subscription linked to your Cursor account.
It also says Grok Bot requires cloud data storage, and accounts on Legacy Privacy Mode must change their data setting before it will start. Platforms listed: macOS, Windows, Linux (x64 or Arm64 as .deb, .rpm or AppImage), iPhone and Android. The overview adds that usage resets weekly and that one Bot runs one computer-use task on its screen at a time, with several Bots working in parallel.
What I could not confirm: prices and exact usage caps. The docs send you to a plans page on cursor.com that I did not fetch. Unite.AI's launch article quoted Cursor Ultra at $200 per month and Teams Premium at $120 per seat per month, but it also described access as limited to three top tiers, while the current docs say any paid individual plan. So the access rules changed or the sources differ; check the live plans page before you budget.
Where do the sources disagree?
Three places, worth knowing because they affect your risk:
| Question | Launch coverage (Aug 2026) | xAI docs (read Oct 2026) |
|---|---|---|
| Whose computer? | "Each Bot gets its own cloud computer" | The computer belongs to your account; all your Bots share files, browser sessions and logins |
| Linux desktop | Unite.AI says a Linux build is posted | Linux listed with .deb, .rpm, AppImage |
| Who gets access | Three top tiers plus enterprise waitlist | Any paid individual Cursor plan, Teams, or linked SuperGrok |
The first row is the one that matters. If Bots share a machine, then a login you give one Bot is reachable by every Bot you run. The community index kydlikebtc/awesome-grokbot says the same thing and advises that a second Bot is not a security boundary. Its README also says that adding a shared Bot copies the name, instructions, skills, routines and first-party plugin IDs, but not the author's computer, files, logins or API keys, and it suggests running a read-only task first before enabling routines or write actions. That repo states it is an index of public bot configurations, not Grok Bot itself and not an installer. Treat community configs as untrusted instructions: read them before you add them.
I did not find a security page in what I fetched. The overview says a dedicated Security page covers network policy, approvals and Auto Review, logging and data handling. I did not read it, so I make no claims about those controls.
What are the self-hosted alternatives?
Only one I verified directly: madebywelch/guaca. Its README describes a desktop app for agents that work together, organised in groups with their own settings and plugins, where plugins are MCP servers. Agents keep their own memory and working notes. For browser or computer access it uses Kernel or E2B with your own API key for those vendors. On a VPS, agents can keep working when your laptop is closed. It says it is inspired by Grok Bot and shares no code, assets or service with it. Licence: GNU AGPL v3. Desktop support is currently macOS only per the README, and building from source needs Node.js, pnpm, Rust and Xcode command line tools. I did not install or run it.
Outside that, you can assemble an always-on agent yourself. Our own routes, none of which are Grok Bot clones:
- A cron-driven script on a VPS, as in the daily AI digest agent.
- Event-triggered agents, as in webhook-driven AI agents.
- n8n on your own server, as in self-hosting n8n for AI agents.
The trade is plain. A hosted teammate gives you a ready computer and a UI for taking over logins. Self-hosting gives you control of where credentials live, at the cost of building the approval, logging and recovery parts yourself.
A checklist for any always-on AI teammate
Use this for Grok Bot, a self-hosted tool or whatever ships next month. Each line is a question to answer from the vendor's docs, not from a demo.
- Whose machine is it, and who shares it? One per agent, per account, or per team? What can one agent see of another's files and sessions?
- Where do credentials live? Does the agent store passwords, or does a human log in and hand over a session? Can sessions be revoked from one place?
- What can it do without asking? Is there an approval step before sending, paying, deleting or posting? Can you make it default-deny? (Our approval gate post shows the idea in code.)
- What is logged? Can you see every action it took on a site, after the fact?
- What are the hard limits? Weekly usage, concurrent tasks, run time. What happens when you hit them mid-job?
- Where does your data go? Retention, training use and storage region. Grok Bot's docs state it needs cloud storage; read the data page for your tier.
- What happens when a site changes? Computer-use agents break when a layout moves. Ask how failures surface: loudly, or as a confident wrong result?
- Can you leave? Can you export skills, routines and memory?
- What is the blast radius? If the agent is hijacked by a malicious web page, what is the worst it can reach? See prompt injection in tool-using agents.
How I would start
If you try Grok Bot, begin with a throwaway account and a read-only task, as the community index suggests. Put nothing on the shared computer you would mind every Bot seeing. Give one Bot one job (the get-started page itself says focused Bots build more useful context than one catch-all Bot). Turn on routines only after a manual run you checked. If you would rather keep the credentials on your own server, the VPS routes above are the alternative, with more setup work.
What I could not verify
- Current prices and caps (plans page not fetched).
- The Security page contents.
- The launch announcement on x.ai (403).
- Any claim about reliability or quality; I did not use the product, and no source I found published reliability data.
This is an early product and the docs are changing, so recheck the pages linked above before relying on any detail.
