How to Set Up Grok Bot for Your First Task

You need an eligible account, a supported client, and a Cursor data setting that permits cloud storage. Current Cursor Help supports ongoing access through SuperGrok Heavy, Cursor Ultra, or an eligible Premium team seat. Pro, Pro Plus, and a Standard team seat alone do not include Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition.. Verify any special offer in the live plan flow.

Check privacy before downloading anything. Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. requires cloud data storage, and accounts using Privacy Mode (Legacy)A Cursor data setting that blocks Grok Bot because the product requires cloud data storage. Press Enter for the full definition. must change to a supported Cursor setting before the product can start. On a team, that belongs with the admin who owns the policy. A member should not make the decision by clicking through onboarding alone.

Pick one real task before setup too. The onboarding flow asks which tools you use and suggests teammates, but those answers do not connect or modify the tools. Having a task ready keeps the first Bot tied to an outcome instead of becoming a general helper with no stable job.

Open the Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. access page at cursor.com/bot/onboarding, choose the correct desktop build, and complete Cursor authentication in the browser. The app checks for updates automatically, with a manual Check for Updates action under Settings and Beta.

1. 1On macOS, choose Apple silicon or Intel, open the disk image, drag Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. to Applications, and open it.
2. 2On Windows, choose x64 or Arm64, run the installer, and open Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. from the Start menu.
3. 3Select Get started. If the app is already open, use Sign In with Cursor from Settings.
4. 4Finish authentication in the browser. Complete your organization's normal SSOSingle Sign-On. One company login (usually via SAML or OIDC) instead of a separate password per tool. Press Enter for the full definition. flow when required.
5. 5Return to Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. and let the shared cloud computer finish its background setup.

The first-run tour explains Bots, the shared computer, and routines, then asks about your tools. It ends at Meet a future teammate. None of those tool choices authorize an integration on their own, which is a useful pause. You can create the role first and connect only the systems its first task actually uses.

If the app sends you back to sign-in or says the plan is ineligible, check the Cursor account shown in the browser before changing subscriptions. Teams can contain several seat types, and a personal account may not be the same account that carries the Premium seat.

Create one Bot around a narrow, recurring outcome. Choose a suggested teammate or Create your own, then give it a short name, one primary job, and a description of how it should work. The description holds durable responsibilities and boundaries. Task-specific details belong in the conversation.

A useful description might say: investigate product-performance questions using approved observability tools, preserve links and screenshots, separate evidence from hypotheses, return the highest-impact issue first, and never change production settings. That gives the Bot a method and a hard stopping point without scripting every click.

Keep the roster small. SpaceXAI recommends adding another Bot only when a stable specialist role appears. Several named Bots can share context and files, but they also share credentials, and extra roles make it harder to remember which routine owns a recurring action. One end-to-end owner is a cleaner starting point.

A strong first request names the outcome, sources, constraints, deliverable, and review point. Keep the source read-only and ask for a result you can verify independently. Attaching a document and requesting a sourced summary is the shortest route because it needs no external login.

• Outcome: the work that should be finished, not a topic to explore.
• Sources: the exact files, apps, websites, or conversations it may use.
• Constraints: what it must avoid and which uncertainty should stop the run.
• Deliverable: the file, linked brief, draft, or evidence pack you expect back.
• Review point: the moment when the Bot must ask before continuing.

Use an acceptance check you can perform in a minute. For a document summary, require every date, decision, and open question to cite its page or section. For an analytics review, ask for direct chart links and a comparison period. A result with source pointers is much easier to correct than a polished paragraph whose evidence you have to reconstruct.

Name a lasting preference when you revise the result, such as keeping source links inline or ending every report with Decisions needed. The Bot can retain stable working preferences, but current facts should stay in the source system. Ask it to reopen those sources on consequential work instead of trusting an old memory.

Use Settings and Plugins for supported connectors, or let the Bot reach a website from its cloud browser. When authentication needs a password, passkey, two-factor code, CAPTCHA, or payment confirmation, open Agent Computer and take control for that step. Return control after the sign-in completes.

1. 1Open Agent Computer from the conversation when the Bot asks for authentication.
2. 2Take control and enter the password, passkey, verification code, or CAPTCHA yourself.
3. 3Confirm the signed-in account and permissions before returning control.
4. 4Tell the Bot to continue, then review the first action it proposes in the connected tool.

Do not paste passwords or one-time codes into ordinary chat. A supported connection may show a secure secret request whose value is masked, excluded from the transcript, and not shown to the model. The docs are careful about its scope: it is a handoff for that connection, not a general password manager.

The browser session persists on the shared computer. That saves repeated sign-ins, and it also means every Bot on the account can use the same session. Connect a scoped service account when the source system supports one, and sign out or revoke authorization when the workflow ends.

Stabilize the one-time task first. A skill records how to do it, while a routine tells one Bot when to run it. Saving the method before adding the schedule gives you something concrete to test and edit.

1. 1Run one safe task and correct the result until it meets the acceptance check.
2. 2Ask the Bot to save the process as a skill with inputs, steps, validation, output, and approval rules.
3. 3Test the skill on a second input and define missing-data, stale-data, retry, and partial-completion behavior.
4. 4Ask the owning Bot for a schedule or supported event trigger, including its time zone and exact input source.
5. 5Use Test run with safe inputs. It performs real work, so keep write actions behind approval.
6. 6Inspect the next run and recent history, then pause the routine when a website, connector, or source format changes.

Teach a task offers another way to draft the skill. From a one-to-one Bot conversation, open the computer, choose Teach a task, describe the result, and demonstrate the browser workflow once. Recording lasts up to ten minutes and captures visible computer interaction without microphone audio. Review what it created before testing because one example rarely contains the failure cases.

A Bot can own up to 50 routines, and the app keeps the 20 most recent run records for each. Those are capacity limits, not a target. A short list with clear owners is easier to audit, and deleting a routine has no undo. Deleting its Bot also removes the routines that Bot owns.

Event-triggered routines use Cursor account integrations and are separate from a Slack or GitHub pluginA Cursor marketplace package that bundles MCP servers and skills (sometimes sub-agents and hooks); one click installs all of it into your Cursor instance. Press Enter for the full definition.. Keep the matching rule narrow. A listener for every new message creates irrelevant runs and extra usage; a specific channel, phrase, and ticket link gives the Bot an event it can recognize and you can explain later.

The iPhone app reaches the same Bots, conversations, routines, connectors, and cloud computer as desktop. On iOS 18 or later, you can start work, dictate or attach a message, review a result, answer a question, and take over the computer for a sign-in or approval while away from your desk.

Routine management is split. On iPhone you can inspect a routine's schedule, next run, and instruction, then pause or resume it. Editing the schedule or instruction, viewing full run history, testing, and deleting still require desktop. Teach-by-demonstration also remains a desktop workflow.

Enable notifications if an unattended job may need a decision. Push delivery is still rolling out, so keep the in-app attention state in the review habit during beta. The cloud work continues whether the phone app is open or closed.