Grok Bot [What It Changes]

TL;DR

• Grok Bot is SpaceXAI's beta product for named, always-on AI teammates that share a cloud computer, sign into the apps a person already uses, finish multi-step jobs, and return for approvals.

• As of the 11 August 2026 launch, SpaceXAI's own pages list access on SuperGrok plans and on Cursor Pro, Pro+, Ultra, and Teams; independent coverage on the same week listed a narrower Heavy / Ultra / Premium set.

• Every Bot on one account shares one computer: files, browser sessions, and logins. Separate Bots are not a security boundary.

• No public accuracy benchmark was published at launch. This is a beta, not a generally available enterprise product.

What Grok Bot is

Grok Bot is a roster of persistent, named AI teammates that live on a cloud computer assigned to your account, sign into tools the way a person would, and keep working after the laptop closes.

That matters to a two-truck HVAC shop, a ten-person marketing agency, and a solo clinic for a boring reason: the work that eats the afternoon is rarely an API. It is the vendor portal with no connector, the insurer inbox, the scheduling page that only works in a browser. If an agent can log into those pages, the constraint that used to stop automation — "this tool has no API" — is no longer the hard stop. The new hard stop is trust: who is signed in, what may send, and whether two clients share a computer.

This hub is current as of 11 August 2026, the launch date on SpaceXAI's introducing post. It is written for an owner who already has a repeatable queue and is deciding whether a cloud-computer agent belongs in that queue at all.

Why now

The old pattern was: prompt, copy, paste, repeat. Grok Bot's pattern is: name a Bot, grant logins, hand it an outcome, review the exception. SpaceXAI built it as an internal prototype for sales outbound, marketing, office operations, and bug fixes, then opened a beta.

The Verge independently describes the cloud-computer model, parallel bots with a coordinator, and the need to let Grok sign into online accounts. Mashable independently confirms the always-on teammate framing and the same early access list it reported. DataCamp independently walks through the docs: one shared machine per account, skills versus routines, and the Cursor distribution path.

What shipped

According to DataCamp, 11 August 2026 is the launch date for the Grok Bot beta. According to the same DataCamp article, iOS 18 is the minimum version it lists for iPhone. According to The Verge, the story posted 12 August 2026 and described availability on desktop and iOS, with enterprise on a waitlist.

According to Mashable, Bots work 24/7 in the background and can run in parallel, including a Bot that manages other Bots. Mashable also links the company-name change to SpaceXAI.

The product site is x.ai/bot. The documentation overview is docs.x.ai/grok-bot/overview. Approvals and the shared-computer warning live on approvals, security, and privacy.

SpaceXAI's launch post lists internal examples: a sales Bot updating CRM notes and drafting follow-ups; an ops Bot seating new hires and processing invoices from Gmail; an engineering Bot reproducing a bug in the product UI and handing a fix to a debugging Bot. Those are company-reported, not third-party benchmarks.

Access and price (public pages, not a quote)

Independent week-of-launch stories and the vendor pages do not match on who is in. Record both.

Sources: SpaceXAI; The Verge; Mashable; DataCamp.

According to the Grok Bot product page, $20 per month is the listed Cursor Pro price, $30 per month is SuperGrok, and $40 per seat per month is Cursor Teams Standard, each billed monthly on that page as of the fetch. The same page says Grok Bot comes with its own usage, separate from Grok and Cursor plans, on the launch post.

Those dollars are vendor list prices on a public page, not a negotiated contract. DataCamp's access list is narrower than SpaceXAI's launch post; check the live plan picker before buying.

How the computer actually works

The overview docs state that each Bot runs on a persistent cloud VM with a browser, filesystem, and terminal. It can use connectors and MCP where they exist, and computer use where they do not. All Bots on the account share that computer.

The approvals page is explicit: do not use separate Bots as a security boundary; sign out when a service should no longer be available; do not send passwords in ordinary chat; take control of the Agent Computer for 2FA, CAPTCHAs, and payment confirmations. Local Mac or Windows execution is a separate setting that defaults to ask-every-time.

DataCamp's FAQ, citing those docs, says a Bot built for one client can technically see files and logins tied to another on the same account, and that the safer move is a separate account per client.

That is the SMB-relevant mechanism. An agency with two clients in one Grok Bot account is not running two sandboxes. A clinic that signs a Bot into the EHR and the personal Gmail of the practice owner has put both on one VM.

Skills, routines, and approvals

DataCamp, citing skills, routines, and automations, draws a line: a skill is how to do something and can apply across Bots; a routine assigns one workflow to one Bot on a schedule. The docs describe watching a task once (up to ten minutes, no audio) and turning it into a draft skill.

The approvals page lists actions that should stay behind a boundary: sending messages, publishing, purchases, deleting data, changing permissions, production changes, and accepting legal terms. Auto Review can require approval or always-allow matching actions; if both match, require-approval wins. Personal Auto-review rules store on the current desktop and sync to that computer; verify them on another install.

Grok Bot requires data storage and does not support Legacy Privacy Mode, per the same approvals page. Privacy and training opt-out follow Cursor's privacy policy and security information. Public share links copy configuration, not the computer; the bot-sharing terms apply. Deleting a Bot does not wipe shared-computer files.

Team rollout notes live on Grok Bot for teams and enterprises and identity and access. DataCamp reports that an audit view of what Bots did and a Grok Bot-specific spend cap were listed as coming, not shipped, at the time it read the docs.

USTA analysis

USTA analysis. Inputs: the public Grok Bot page lists Cursor Pro at $20/month, SuperGrok at $30/month, and Cursor Teams Standard at $40/seat/month.

Derived comparison, vendor list only, no discount assumed:

• Cursor Pro list ÷ SuperGrok list = 20 / 30 = 0.67, so the Cursor Pro line is two-thirds the SuperGrok line on that page.

• Teams Standard list ÷ Cursor Pro list = 40 / 20 = 2.0, so one Teams seat is twice one Pro seat on that page.

• Three people on Teams Standard at list = 3 × $40 = $120/month, which is 120 / 30 = 4.0 times one SuperGrok subscription.

This is not a TCO, not a usage-included comparison, and not a claim that Grok Bot is "included free." The launch post says Grok Bot has its own usage. The analysis only divides the three numbers printed on the pricing block we opened.

Where this fits a small workflow

Start with a job that already has a human approval: overnight invoice chasing, NDA collection, Monday pipeline hygiene. Mashable's SpaceXAI examples are exactly those: a Bot sending NDA invites; a Bot finding open invoices and escalating disputes. Do not start with unsupervised outbound to customers.

A marketing agency comparing Plutio alternatives or agency automation tools should ask whether the Bot will sit in the same browser session as two clients. If yes, split accounts. Dispatch-style work in agency dispatch software still needs a person on anything that texts a customer.

HVAC and clinic intake still belong on a path with a review queue. Teams already routing forms through form-to-CRM automation can keep Grok Bot off the send step and on the research-and-draft step. Executive-assistant task automation is the closest analog: calendar, follow-up, and a human on the judgment call.

If that queue already runs in US Tech Automations, plug Grok Bot in as a model-and-computer step behind the same approval node, not as a second system of record. The homepage stack is for the durable path; Grok Bot is a beta teammate that can fail a login.

A day-one setup that does not dump the CRM

The get-started docs and computer-and-apps pages sit next to the overview. Use them as a checklist, not as a reason to connect every tab on day one.

Create one Bot with a job title that matches a single queue: "invoice chase," not "operations." Grant the vendor portal and the shared inbox for that queue only. Run the use-case catalog examples as reading, not as a copy-paste of SpaceXAI's internal sales Bot into a two-person shop.

Turn the first pass into a skill only after you watched it once. The skills and routines split matters: a skill is reusable instruction; a routine is a scheduled assignment. If you schedule first, you have automated a path you never reviewed.

Keep Auto Review on send, publish, pay, delete, and permission changes. The FAQ is the right place to re-read isolation and laptop-off behaviour before you hand a clinic login to a Bot that also has the owner's personal mail.

Enterprise waitlist and sales still go through Cursor contact and the Grok plans picker on the product page. Do not treat a screenshot of those prices as a contract.

A 10-person agency that already uses form-to-CRM tools should land Grok Bot drafts in the same review column as web-form leads. The Bot is not a second CRM. The state of small-business automation still starts with one record of truth.

If the Bot must watch you work, cap the session, sit in a test account, and write down which URL it saw. Then delete the recording if the vendor does not say how long it lives. The approvals page is silent on a lot of retention detail; that silence is a reason to keep production data off the first week.

Parallel Bots are useful only after the first Bot is boring. A coordinator Bot that messages specialists is how SpaceXAI talks about octopus arms. A small shop that turns that on before it has an approval rule will multiply mistakes across inboxes.

Honest limits

No public accuracy benchmark shipped at launch. Company demos are the cases that worked. DataCamp lists open questions: long messy workflows, shared-computer trust, missing audit and spend-cap features, and error propagation when one Bot's output is another Bot's input.

Linux desktop, Android, and iPad were unsupported in DataCamp's launch write-up. Enterprise was a waitlist on Verge, Mashable, and the launch post.

Grok Bot is not the Grok chatbot. DataCamp states they are sold separately.

The SpaceX–Cursor acquisition was described by DataCamp as expected to close in the third quarter, not confirmed closed, at the time of that article. Do not treat Cursor as a completed merger on the strength of a download URL pointing at downloads.cursor.com.

Signal vs Speculation

Signal (sourced). SpaceXAI launched Grok Bot on 11 August 2026. Bots share a cloud computer, can use apps without an API, run in parallel, and come back for approvals. Independent coverage from The Verge, Mashable, and DataCamp confirms the teammate framing and the desktop/iOS beta. Public list prices on x.ai/bot include $20 Cursor Pro, $30 SuperGrok, and $40 Teams. Docs warn that Bots are not a security boundary.

Speculation (our read). If the shared-computer model holds, agencies and clinics will either buy a second account per client or keep Grok Bot off anything with a second tenant's credentials. Our read: the 12–36 month path for small teams is "draft and chase," not "send and close," until audit logs and spend caps that DataCamp flagged as coming actually ship. Our read is not a reliability score.

Rival products named in the coverage — OpenAI ChatGPT Work, Anthropic Claude Cowork, Microsoft Copilot Tasks, Google Gemini Spark — show a category, not a winner. DataCamp's Grok Bot tutorial and Grok 4.6 notes are adjacent product pages, not proof that Grok Bot finished a third-party workload.

Industry figures, not list prices.

Industry figures, not list prices.

Industry figures, not list prices.

According to AICPA, 62% of firms reported cloud-workflow adoption.

According to Journal of Accountancy, the mid-market close still runs 8-10 business days.

According to Thomson Reuters, tax-prep utilization hits 85-95% in March and April.

Key Takeaways

• Grok Bot is a named cloud-computer teammate in beta, not a chat window with extra buttons.

• One account, one computer: do not use extra Bots as client isolation.

• Public list prices on the product page start at $20 Cursor Pro and $40 Teams per seat; usage is separate.

• Independent outlets and the vendor still disagree on the exact plan list; check the live picker.

• Keep send, pay, delete, and production changes on approval.

FAQ

Is Grok Bot the same as the Grok app?

Direct-answer first sentence: no; DataCamp states they are separate products. You can hold one without the other.

Can it work when the laptop is off?

Direct-answer first sentence: yes, because the work runs on the cloud computer, not on the device. SpaceXAI and DataCamp both describe background and scheduled work continuing after the app closes.

Are two Bots isolated from each other?

Direct-answer first sentence: not on the same account. The overview and approvals docs say all Bots share files, sessions, and logins.

Who can get it today?

Direct-answer first sentence: it depends which page you trust. SpaceXAI's 11 August launch post includes SuperGrok plans plus Cursor Pro through Ultra and Teams Standard and Premium; Verge, Mashable, and DataCamp listed SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium. Enterprise remains a waitlist on those pages.

Should a Bot get the password to the bank portal?

Direct-answer first sentence: no; the approvals docs tell you to take control for passwords, 2FA, CAPTCHAs, and payment confirmations. Use a secure secret request when the connection supports it, not chat.

What to do next

Pick one overnight draft job with a morning review. Use a dedicated account if more than one client can appear in the browser. If you want that review queue on a durable path, route the exception through agentic workflows instead of leaving the only copy of the work inside a beta Bot. US Tech Automations belongs on the approval and record step, not as a slogan on the launch.