AI News

What Is Grok Bot? Features, Availability, Pricing, and How It Works

What is Grok Bot? It is SpaceXAI’s early-beta system for handing real work to persistent AI agents that can operate inside signed-in apps and websites, coordinate with other Bots, remember routines, and keep working after you close your laptop. That makes it materially different from ordinary Grok, which remains primarily a conversational assistant even though it already supports connectors and multi-agent reasoning. Launch access is limited to SuperGrok Heavy, Cursor Ultra, and Cursor Premium Teams subscribers. Individual access costs $200 or $300 per month, while the team plan costs $120 per seat per month. Desktop builds and an iOS app are available; Android is not.

Testing status: This launch-day explainer is based on official materials reviewed on August 11, 2026. Kingy.ai has not independently tested the Grok Bot beta.

Grok Bot at a glance

Developer SpaceXAI; the official footer identifies X.AI LLC
Announcement date August 11, 2026
Release status Early beta
What it does Runs multi-step work across apps, websites, inboxes, and files using a persistent cloud computer
Eligible plans SuperGrok Heavy, Cursor Ultra, and Cursor Premium Teams
Lowest listed price $120 per seat per month for Cursor Premium Teams; individual plans start at $200 per month
Platforms at launch macOS, Windows, Linux, and iOS
Android Coming soon; no date announced
Enterprise Waitlist / contact sales

What is Grok Bot?

SpaceXAI calls Grok Bot an “AI teammate,” but the useful definition is less anthropomorphic. A Bot is an agent that receives an assignment, uses a persistent computer in the cloud, works through authorized apps and websites, and returns either completed work or a request for human judgment.

The official announcement says Bots can work across tools even when those services lack a clean API or Model Context Protocol connection. In other words, the product is designed to operate interfaces rather than depend exclusively on formal software integrations.

That places Grok Bot in the app-action lane described in Kingy’s State of AI Agents 2026: its value is not another generated answer, but whether it can complete a bounded workflow inside the systems where work actually lives.

That distinction is also the main reason to treat the launch cautiously. SpaceXAI has published internal examples and early-user testimonials, not independent evidence that the beta can reliably complete these workflows in other companies.

How Grok Bot works

  1. Create or select a Bot. A user can give each Bot a job, such as sales outbound, expense management, recruiting, or bug reproduction.
  2. Describe the assignment in a message. SpaceXAI positions the interface as a normal conversation rather than a workflow builder.
  3. Give the shared cloud computer access to the required tools. Bots can use its files, browser state, and sign-ins.
  4. Let the Bot execute the steps. It can move among apps and sites, continue while the user is away, and reuse a previously learned routine.
  5. Bring in other Bots when useful. Specialist Bots can share context, pass work, and coordinate in a group thread.
  6. Review the result or make a judgment call. The product can return finished work or pause when approval is needed.

The word shared matters. The product FAQ says a user’s Bots share one persistent cloud computer, including its files, browser, and logins. Isolation is per user, not per Bot. SpaceXAI’s broader wording that Bots have “their own computer” should not be read as a separately isolated machine for every agent.

The main Grok Bot features

  • Persistent execution: SpaceXAI says jobs can continue after the user closes a local laptop. Its “24/7” language does not mean unlimited usage; eligible plans still have weekly allowances.
  • App and website action: Bots can operate interfaces across inboxes, CRM software, product UIs, and other services, including those without a clean API or MCP integration.
  • Learned routines: A Bot can observe a workflow, save the steps, incorporate corrections, and repeat it. The launch material does not explain versioning or recovery when an interface changes.
  • Persistent context: SpaceXAI says Bots learn a user’s voice and edge cases, although Kingy.ai has not measured memory accuracy or drift.
  • Parallel specialists: Internal launch examples cover sales, marketing, operations, finance, product, engineering, and community work. They are not independent customer case studies.

How multiple Grok Bots collaborate

A user can run several specialist Bots at once and place them in the same thread. The Bots can message one another, share project context, pass work, and assign ownership without requiring the user to copy notes between separate chats.

The launch post also describes a hierarchy in which one “chief of staff” Bot coordinates specialists for inboxes, expenses, recruiting, operations, or engineering. A research Bot might gather evidence, a communications Bot draft from it, and a coordinator decide what needs approval.

This is not the same as ordinary Grok’s multi-agent reasoning. In ordinary Grok, parallel agents can divide a difficult question and merge their findings into an answer. Grok Bot is presented as a persistent operational team that acts across a shared working environment and hands off ongoing jobs.

The arrangement also concentrates access. Because Bots share one per-user cloud computer, they may share browser sessions, files, and authenticated tools. Buyers should understand that boundary before connecting sensitive accounts.

Grok Bot availability and supported platforms

Platform Launch status Official detail Caveat
macOS, Apple silicon Available Dedicated download Launch page currently serves version 0.16.0
macOS, Intel Available Listed under “More downloads” Separate Intel build
Windows Available Windows 10 and 11, x64 No ARM build listed
Linux Available Debian and Ubuntu, x64 .deb The FAQ omits Linux even though the download dialog offers it
iPhone and iPad Available Dedicated iOS App Store listing Feature parity with desktop is not documented
Android Coming soon Shown in the download dialog No release date
Web Unconfirmed No dedicated web client is identified Ordinary Grok’s web app should not be assumed to support Grok Bot

The Linux entry is a genuine documentation mismatch. The official FAQ says users can talk to Grok Bot on macOS, Windows, or iOS, while the same product page’s download dialog supplies a Linux package. The most accurate launch-day description is that a Linux build is available, but the FAQ does not explicitly include Linux among its supported conversation surfaces.

Access is also plan-gated. The announcement names SuperGrok Heavy, Cursor Ultra, and “Cursor Teams Premium.” The current product and pricing page reverses the team-plan wording to Cursor Premium Teams, which is the label used here. Free Grok, standard SuperGrok, SuperGrok Plus, and ordinary Cursor Pro are not listed as eligible.

Grok Bot pricing

Plan Launch-day price Grok Bot access Usage note
SuperGrok Heavy $300 USD/month Included Weekly usage included; exact Bot allowance not published
Cursor Ultra $200/month Included Product page says “extended limits on AI tokens”
Cursor Premium Teams $120/seat/month Included Adds team billing, analytics, marketplace, and SAML/OIDC SSO
Enterprise Not published Waitlist / early access Contact Cursor sales or the relevant account team

These are the US monthly list prices displayed by the Grok Bot product page and SuperGrok subscription selector on August 11. They are subscription prices, not xAI API token rates.

The product FAQ says subscriptions include weekly usage and bills additional use according to token cost. It does not publish the included Grok Bot allowance, the full extra-usage rate card, or the difference in Bot limits among the three eligible plans. A “Get started for free” download button on the page does not override the paid eligibility rules.

Grok Bot vs ordinary Grok

Area Grok Bot Ordinary Grok
Primary role Complete ongoing work across tools Chat, search, reason, create, and analyze
Working environment Persistent per-user cloud computer Conversation on web, iOS, or Android
Third-party tools Can operate signed-in app and website interfaces Can reach supported services through connectors
Persistence Designed to keep jobs running when the laptop is closed Conversation history and settings sync across devices
Repeatable work Can observe and save routines Supports instructions and conversational context
Multiple agents Specialist Bots can message, coordinate, and hand off work Multi-agent mode parallelizes reasoning for a response
Human control Can return for approvals and route sensitive actions through Auto Review User directs the conversation and connected-tool requests
Access Three high-priced plans at launch Free to start, with paid SuperGrok tiers

Ordinary Grok is not merely a chatbot. Its official overview includes live search, file analysis, image and video creation, voice, connectors, and multi-agent reasoning. Grok Bot’s claimed difference is sustained execution: it acts in the work environment, remembers routines, and coordinates handoffs rather than stopping at an answer.

For more background on the underlying workplace push, see Kingy’s analysis of Grok 4.5 and SpaceXAI’s agent strategy.

Privacy, permissions, and control

Giving an agent access to workplace accounts changes the risk profile. The official FAQ says Grok Bot uses Cursor authentication, SSO, and privacy mode. It says the cloud computer is encrypted in transit and at rest, provides a training opt-out, and can send sensitive actions through Auto Review before execution. Enterprise administrators can configure data-loss prevention, certificates, proxies, and network controls at boot.

The launch material does not fully explain individual retention defaults, credential handling, permission granularity, audit records, or the default line between autonomous and approval-required actions. A training opt-out is not the same as no training by default.

Before connecting consequential systems, use a bounded workflow, limited permissions, review gates, and a reversible test environment. Kingy’s AI Agent Adoption Playbook explains that safer rollout pattern in detail.

What is confirmed—and what remains unanswered

Confirmed at launch

  • Grok Bot is an early beta available through three paid subscription tiers.
  • Bots can work through one persistent cloud computer shared per user.
  • Multiple Bots can coordinate, message one another, and share context.
  • Users can demonstrate workflows and save them as routines.
  • macOS, Windows, Linux, and iOS downloads are available.
  • Android is marked as coming soon.
  • Enterprise users are directed to a waitlist or sales contact.

Still unanswered

  • The precise weekly allowance and extra-token rates for Grok Bot.
  • Regional availability and whether all eligible subscribers receive immediate access.
  • An Android release date or confirmed dedicated web client.
  • Mobile-versus-desktop feature parity.
  • A complete list of supported apps and account types.
  • Plan-by-plan differences in Bot limits and concurrency.
  • Enterprise launch timing and pricing.
  • Default approval thresholds, failure recovery, and practical beta reliability.

Who should consider Grok Bot now?

The strongest early fit is an existing eligible subscriber testing a bounded, repetitive workflow. Sales research, CRM hygiene, inbox triage, expenses, and bug reproduction match the stated design because each crosses systems and produces a reviewable output.

Everyone else should wait for evidence. Prices are high, usage economics are incomplete, and the product is explicitly a beta. What matters is whether a Bot can finish work repeatedly, safely, and cheaply inside a messy real account.

Grok Bot FAQ

Is Grok Bot the same as Grok?

No. Ordinary Grok is a conversational assistant with search, creation, connectors, and multi-agent reasoning. Grok Bot adds persistent cloud-computer access, ongoing app-level execution, learned routines, and specialist Bots that can coordinate work.

Who can use Grok Bot?

At launch, SpaceXAI lists SuperGrok Heavy, Cursor Ultra, and Cursor Premium Teams subscribers. Enterprise users are directed to a waitlist or sales contact.

How much does Grok Bot cost?

SuperGrok Heavy is $300 per month, Cursor Ultra is $200 per month, and Cursor Premium Teams is $120 per seat per month. Plans include weekly usage, with additional usage billed by token cost; the full Bot allowance and extra-usage rates are not publicly detailed.

Does Grok Bot work on Windows, Mac, Linux, iPhone, or Android?

The official download page offers macOS builds for Apple silicon and Intel, Windows 10 and 11 x64, a Debian/Ubuntu Linux package, and an iOS app for iPhone and iPad. Android is marked “Coming soon.”

Can multiple Grok Bots work together?

Yes, according to the launch materials. Bots can run in parallel, message one another, share context in threads, coordinate in group chats, pass assignments, and involve the user for judgment calls.

Do Grok Bots share a computer?

Yes. The FAQ says every Bot belonging to one user shares a persistent cloud computer, including its files, browser, and logins. Isolation is per user, not per individual Bot.

Is Grok Bot available to enterprises?

Not as a general launch. The announcement offers an enterprise waitlist, and the FAQ says broader access is expected later. Neither a binding release date nor enterprise price is published.

Is Grok Bot safe to connect to work accounts?

SpaceXAI lists encryption, Cursor authentication and privacy controls, training opt-out, Auto Review, and enterprise network controls. Those claims are not a substitute for testing permissions, approval rules, logs, and recovery in your own environment.

The bottom line

Grok Bot is more than ordinary Grok with a different name. Its persistent cloud computer, saved routines, app-level action, and Bot-to-Bot handoffs are a coherent attempt to move from AI assistance to delegated work.

The launch is also deliberately narrow and expensive. SpaceXAI has confirmed the product shape, eligible plans, and core access surfaces, but not the usage economics or operational reliability that will determine whether it is genuinely useful. Existing eligible subscribers have a reason to test one bounded workflow. Everyone else can afford to let the beta produce evidence first.

Official sources

Sources last checked August 11, 2026. Featured image: Kingy AI editorial illustration.