We set up Grok Bot in steps: name the job, confirm it's the right tool, create the bots, sign them into only what they need, set approvals, teach the workflow, test on real work, and hand off.

We have set up Grok Bot for clients, and because each bot has a cloud computer of its own, the work is less about installing software and more about briefing a teammate well: agree on the first job, get the official app in place, create the bot or bots the job needs, log them into only the tools they need, set permissions, review, and approvals, teach the workflow by describing it or showing it once, test with real work, and hand off. Product facts here come from x.ai/bot and the official Grok Bot guides, current as of November 2026.

Step 1: name the job and the owner

As with any agent, we start with one job in one sentence and a named owner. Grok Bot's official pages describe bots that keep context on how you work and come back when your approval is needed. That only helps if someone is there to answer. We agree on who approves and how quickly.

Step 2: confirm Grok Bot is the right shape

Grok Bot is a hosted teammate. Its computer runs in the cloud, so it can keep working when your laptop is closed, and it can use websites and apps the way a person would. That suits clients who want nothing to run themselves and who work in tools that are awkward to automate any other way. If a client wants the assistant and its memory only on hardware they own, we recommend an open-source agent instead. See How to choose an AI agent for your business.

The official FAQ says Grok Bot is generally available for Enterprise and for teams on the Teams plan, with sales involved to set up Enterprise access for an organization. We help the client work out which access path applies. We are not the vendor.

Step 3: get the app in place and create the bots

Grok Bot is used from official desktop and iOS apps. We help the client get set up, then create the bot or bots the job needs. Official pages describe running more than one bot in parallel, for example one on a project and another on outbound work. For a first setup we usually create one bot with one clear job, and add others later.

Step 4: log the bot into only what it needs

Because the bot works inside real apps and websites, we log it into the specific tools the job requires and nothing else. When a sign-in needs the client, for example a CAPTCHA or a two-factor code, official guides describe the bot handing the desktop back, and sending a secure form for a password or API key rather than asking for it in chat. We use those paths. Passwords and codes are never pasted into a conversation.

Step 5: set permissions, review, and approvals

Official guides describe bots kept in check by permissions, a reviewer, and allow and block lists, and the official FAQ says sensitive actions can go through Auto Review before they run. We configure those for the job, then add our own approval rule on top: the bot may read, sort, and draft; anything that leaves the business, moves money, deletes data, or changes access waits for the owner. See Approval rules for AI agents.

Step 6: teach the workflow

Official guides recommend describing a workflow the way you would do it, or recording a demo on your screen so the bot can follow along once and save it as a routine. In practice we do both. We write a short brief with the client covering what starts the job, where information lives, what good looks like, and when to stop and ask. If the workflow lives in a website that is hard to describe, the client shows it once.

The quality of that briefing matters more than any setting. A vague instruction like "handle my email" produces vague work. A specific one like "each morning, list new inquiries, draft a reply to each using our FAQ, and hold them for my approval" produces work you can check.

Step 7: test with real work

We run the job against real, recent work with approvals on and review every result with the client. We test the awkward cases too: a request outside the bot's job, an email that contains instructions from a stranger, and a site that asks for a sign-in. We adjust the brief until the bot's drafts need few edits.

Step 8: hand off

Handoff covers how to message the bot, how approvals reach you, how to open the bot's computer when you want to see what it is doing, how to pause it, and which accounts it can reach. The client gets written notes. The bot is a teammate with an owner, not a black box.

What happens after handoff?

The first change most owners want is to let the bot do one routine thing on its own. We loosen approvals one action at a time. The second is often another bot for a related job. Because Grok Bot can run several bots in parallel, adding a second job does not mean overloading the first bot.

How is privacy handled?

The official FAQ says Grok Bot uses the same Cursor SSO, auth, and privacy mode used in that product family; that the cloud computer is encrypted in transit and at rest, with training opt-out; and that enterprise admins can set DLP, certificates, proxies, and network controls. We repeat that as official policy, not as our own audit. Our general checklist is in Securing an AI agent: basics.

Questions

Do I need a computer running all the time for Grok Bot?

No. The bot works on a cloud computer of its own, so it keeps working when your laptop is closed.

Can Grok Bot use websites that do not have an integration?

Official pages say that once you log a bot in, it uses apps and websites the way you would, including tools that are harder to navigate than a simple API.

What happens when it hits a login or CAPTCHA?

Official guides describe the bot handing the desktop back to you for steps like a CAPTCHA or two-factor login, and sending a secure form for passwords or keys.

Do you only set up Grok Bot?

No. We have also set up OpenClaw and Hermes Agent for clients, and we recommend what fits.