Guide

What Is Grok Bot, and Is It Worth It for a Business?

By Ryan Lanyon, Founder and AI Systems Architect, Nexus Digital
Published 2 November 2026

Grok Bot is SpaceXAI's beta app for always-on agents, called Bots, that sign into your apps and websites and work through tasks on a cloud computer of their own. It is a separate product from the Grok chatbot on X. For a business it is worth trying on one narrow job with limited access, while giving it the keys to everything can wait.

SpaceXAI is the company formerly called xAI, and it announced Grok Bot in beta on 11 August 2026. This guide uses the vendors' own documentation as read in October 2026, and names the press where the vendors are silent.

What is Grok Bot?

Grok Bot is a product that lets you set up Bots, each a named AI teammate with one job, its own cloud computer, and memory that carries from one task to the next.

Each Bot runs on a cloud computer with a browser, a file system and a terminal, so it keeps working when your laptop is closed, and Bots can coordinate with each other. The loop has four parts.

1
You give a Bot a name and one job. The documentation lists starter roles and prompts to begin from, and the same Bot is reachable from the desktop app and from your phone.
2
It signs into the tools you already use. SpaceXAI says Bots work across apps and inboxes, including platforms with no clean API. The documentation says to use a connected app where one exists and the browser for services that have none.
3
It works on its own computer. The Bot carries on while you are away, using the browser, files and terminal on that cloud computer.
4
It comes back when you need to decide. For approvals, and for steps such as logins, two-factor codes and payments, the Bot hands the computer to you.

How is Grok Bot different from the Grok chatbot on X?

Grok on X is the chat assistant, and Grok Bot is a separate product that acts on your behalf, with its own computer and its own usage allowance.

SpaceXAI's site lists Grok on X and Grok Bot as separate products, and its announcement says Grok Bot comes with its own usage, separate from your Grok and Cursor plans. The two get mixed up partly because a Bot can also be tagged from X, a feature the documentation says is not available in every region yet. If you searched for Grok Bot hoping to find the chatbot you talk to on X, this is a different thing.

Who can get access to Grok Bot, and what does it cost?

As of October 2026, the vendor documentation says Grok Bot is included with every paid individual Cursor plan and with the Cursor Teams plan.

You can also link an individual SuperGrok, SuperGrok Plus or SuperGrok Heavy subscription, or use a seat on a self-serve Grok Business plan, and sign-in runs through a Cursor account. The Cursor link exists because SpaceX agreed in June 2026 to buy Anysphere, the company behind Cursor, and The Next Web reports that SpaceXAI and Cursor are merging into one company. The desktop app runs on macOS, Windows and Linux, and there are iPhone and Android apps.

Launch reports differ on who could use it. The Next Web reported access limited to SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium subscribers, with an enterprise waitlist, while SpaceXAI's own announcement listed more plans, so read the current plans page yourself.

On cost, neither SpaceXAI's nor Cursor's documentation gives a dollar price for Grok Bot. The announcement says it has its own usage, separate from your Grok and Cursor plans, and the Cursor page says usage resets weekly without listing quotas. Third-party sites quote prices that disagree with each other, so we have left them out, and a trial is the only way to find the real cost for now. Our guide to AI implementation costs in Australia covers the other route, having agents built into your operations.

What can a Bot see and do with your logins?

A Bot acts as the signed-in member, with no identity or credentials of its own, so what it can reach is what that person's accounts can reach.

The one exception is a team-managed connector, which can use team or service-account credentials, and the security documentation sets out the rest.

Passwords stay with you. The Bot does not type credentials or see your password, and for logins, two-factor codes and payments it asks you to take over the computer.
Sessions persist and are shared. The Bot computer belongs to your account rather than to one Bot, so a browser session one Bot uses is available to your other Bots.
App tokens stay off the computer. Connector tokens are held by Cursor's connector backend and are never stored on the Bot computer.
Your own machine is reachable. On the desktop a Bot can run commands and read and move files, with approval asked for each command by default.
Your data sits in the United States. The computers run there today, Grok Bot requires cloud data storage, and the documentation says that with Privacy Mode on, customer data is not used for training.

What a Bot can reach

The signed-in person's access

What their accounts can reach, the Bot can reach

Passwordsstay with you
Sessionsshared across your Bots
App tokenskept off the computer
Your filesreachable from the desktop
Your datastored in the United States

If your business is covered by the Privacy Act, raise the overseas disclosure rule with your privacy adviser before a Bot goes near customer records. Australian Privacy Principle 8 says an organisation must take reasonable steps to make sure an overseas recipient does not breach the other principles before it discloses personal information to them. Whether a Bot working inside your systems counts as a disclosure is a question for that adviser. The same access questions sit at the centre of an AI readiness assessment.

How do Grok Bot approvals work?

When a Bot needs approval, you see the proposed action and its inputs and choose Allow once, Always allow or Deny.

Allow once lets the Bot continue with that action, Always allow can save a matching rule so it stops asking, and Deny blocks it. The documentation adds that an approval controls the proposed action and not work already completed, and that nobody should approve an action whose target or effect they cannot identify.

Auto Review is a second layer, an independent review model that evaluates risky Bot actions before they run and can let them proceed, require approval or deny them. It does not review every side effect, and the documentation names memory writes and most settings changes as examples. It does not say whether Auto Review is on by default, and admin enforcement and team rules are Enterprise only.

Two more settings matter for a team on a self-serve plan. Action Recording, which keeps a record of Bot actions for 90 days, is off by default, and audit logs are Enterprise only, so self-serve Teams do not get that log. The documentation also shows an example rule that asks for approval before any external email goes out, the kind of rule you would write on day one rather than assume exists.

What was the Grok Build report?

The Next Web reported on 14 July 2026 that a security researcher found Grok Build, a separate coding tool, uploading entire Git repositories, committed secrets included, to a cloud storage bucket.

The researcher, publishing as cereblab, analysed version 0.2.93. The Next Web's report says the tool had been marketed with the claim that nothing from your codebase was transmitted to xAI's servers during a session, and that the captured traffic contradicted it. Elon Musk confirmed the uploads and said earlier Grok Build user data would be deleted, and The Next Web notes that no independent audit has confirmed the deletion.

This is a report about a different product, and nothing in it says Grok Bot behaves the same way. We cite it for what it teaches about any new agent tool, which is that a vendor's description of what leaves your systems is a claim you check for yourself. With Grok Bot, you decide what the Bot is given, and you read the approval prompts and any recorded actions afterwards.

How we handle the same question in our own system

We run agents inside our own business, which makes the approval question a practical one for us. We let our own agents write email drafts into Gmail, and none of those drafts goes out until a person approves it.

The agents that work from our task board follow the same rule at larger scale. We sort all 982 actions our agents' tools offer into four classes: 503 reads and 329 writes that run freely inside an approved plan, 36 terminal actions such as sending or switching on a campaign that need a person's tap, and 114 destructive ones that stay blocked even with approval unless a separate setting is switched on. Building a thing and drafting a thing are cheap, so they run freely, and the irreversible step is where a person stops the line.

We freeze the list of actions an agent proposes the moment a person approves it, and the code that runs the list uses no model, so nothing outside what the person read can run.

In Grok Bot an approval controls one proposed action, and an Always allow rule can widen it to anything that matches. Ours asks a person to approve a whole plan they have read. We have not used Grok Bot ourselves yet, so this is the standard we would hold it to.

Is Grok Bot worth it for a business?

For most 10 to 50 person Australian businesses, Grok Bot is worth a small, supervised trial now. A broad rollout should wait until it leaves beta and you have read its access settings.

The table sorts the situations we would expect a business of that size to be in.

Situation	Our read	Why
One owner or operations lead with a repetitive task in a web app that has no good API	Try it now, on that one task	The reach is one person's accounts, and the same person reviews every approval.
Several staff each wanting Bots on shared company logins	Wait	A Bot acts as whoever is signed in, and audit logs are Enterprise only.
Work that touches customer, client or financial records	Wait for a privacy check	The computers run in the United States, and overseas disclosure rules may apply.
Anything that sends email, spends money or publishes	Only behind approval rules you have written and tested	Auto Review does not cover every side effect, and Always allow stops the questions.
A decision that needs a monthly cost	Trial first	There is no public dollar price, and usage resets weekly.

A trial worth running starts with one job you can time. Give the Bot a login made for that job, keep approvals on, and run it for two weeks. Then compare the minutes you spent reviewing its work with the minutes it saved.

Five checks come before you set it up.

1
Which login it will use. Create one for the job with only the access that job needs.
2
What each approval covers. Read the proposed action, and treat Always allow as a standing permission.
3
Where the data goes. Ask your privacy adviser before customer records are involved.
4
Who can see what the Bot did. Decide in advance how you will review its work, since audit logs are Enterprise only.
5
What it costs in usage. Measure a fortnight of real use before you plan around it.

The product is moving, so check the vendor pages again before you decide. If you want help working out which jobs in your business an agent could safely take on, the Strategic Assessment is where we start, and the first conversation is an AI Opportunity Call.

Sources

FAQ

Grok Bot is SpaceXAI's product for running always-on AI agents, called Bots, that sign into your apps and websites and work on a cloud computer of their own. Each Bot has a name, a job and memory that carries between tasks, and it hands work back when a person needs to decide something. It has been in beta since 11 August 2026 and is separate from the Grok chatbot on X.

No. Grok on X is the chat assistant inside X, while Grok Bot is a separate product with its own cloud computer and its own usage allowance, and it acts in your apps on your behalf. SpaceXAI's own site lists them as separate products. A Bot can also be tagged from X, which is one reason people confuse the two, though the documentation says that feature is not in every region yet.

Neither SpaceXAI's nor Cursor's documentation gives a dollar price for Grok Bot as of October 2026. It is included with paid Cursor plans and can be reached through a linked SuperGrok subscription or a Grok Business seat, with its own usage that resets weekly. Prices quoted on third-party sites disagree with each other, so check the plans and billing page before you decide.

Not by default, in our view, because a Bot acts as the signed-in member, so it can reach whatever that person's accounts can reach. Give it a login made for one job, keep approvals on for anything that sends or spends, and ask your privacy adviser before customer records are involved, since the computers run in the United States. This is our reading of the vendor documentation as of October 2026.

Businesses with several staff sharing company logins, work that touches customer or financial records, or any job that sends email or spends money should wait or start behind strict approval rules. Audit logs are Enterprise only and the computers run in the United States, so those cases need more control than a self-serve plan gives you today. A single owner testing one narrow task is the lowest-risk way in.

Want this against your own numbers? The faster version is a call. We map where the hours are going in your business, then show you what an implementation would return.

Book an AI Opportunity Call

Want the next one in your inbox? Join the weekly AI newsletter and get the free AI security guide.