AI Agents

Grok Bot Review: xAI's Always-On AI Agents Explained

Reviewed August 16, 2026 · ToolStep Editorial Team

On August 11, 2026, xAI launched Grok Bot, a beta product that turns AI from a chat session into a standing piece of infrastructure: each "Bot" gets its own persistent cloud computer with a browser, filesystem, and terminal, signs into the tools you already use, and keeps working after you close your laptop. It is the first major joint product of the SpaceX–Cursor integration, and xAI's entry into the "digital labor" category alongside OpenAI's ChatGPT Agent, Anthropic's Claude Cowork, and Google's Gemini Spark. This explainer covers what actually shipped, how the always-on architecture works, what it costs, and the trust questions you should ask before handing over your logins.

How we evaluated: This comparison is based on official product documentation, published specifications, vendor benchmarks, and current pricing — following the ToolStep testing protocol. We did not run hands-on tests for this comparison, and we do not claim testing that did not happen.

Quick Answer

Grok Bot is xAI's beta launch of always-on AI agents. Every Bot gets a persistent cloud computer, signs into your existing tools with a secure login handoff, and runs multi-step jobs end to end — only surfacing when something needs your approval. It ships today with SuperGrok Heavy ($300/month), Cursor Ultra ($200/month), and Cursor Teams Premium ($120/seat/month) on desktop and iOS, with a wider rollout promised alongside Grok 4.6. It is the most ambitious "AI teammate" product shipped by a major lab so far, but it is an early beta with no independent benchmarks, no published security audit, and pricing that assumes your time is worth $200+ per month.

What Is Grok Bot?

Grok Bot is xAI's always-on AI agent product, launched in beta on August 11, 2026. According to xAI's launch post "Introducing Grok Bot," it is framed as "AI teammates you can give real work to" — a deliberate contrast with the Grok chatbot most people know from X.

The distinction matters. According to xAI's developer documentation, a single Grok Bot is "a persistent, named agent, one AI teammate" — separate from the shared chat interface of the Grok app. Where a chatbot session ends the moment you close the tab, a named Bot keeps its state: files it has created, browser tabs and login sessions it has open, and preferences it has learned from watching how you work. In short: Grok talks, Grok Bot does.

xAI says the feature began as an internal prototype that "took off across the company," with internal teams building Bots for sales outreach, marketing campaigns, office operations, CRM cleanup, invoice processing, email drafts, and bug reproduction before it became a public product. That origin story positions Grok Bot as a proven internal tool being opened up rather than a speculative launch.

Grok Bot is also the first big product to come out of the SpaceX–Cursor tie-up. SpaceX, which folded xAI into itself earlier in 2026, agreed in June to acquire Cursor's maker Anysphere for a reported $60 billion, and Grok Bot is bundled with both Grok and Cursor subscriptions while the deal heads toward closing.

How Grok Bot Works

The workflow is designed to feel like delegating to a colleague rather than prompting a model:

  1. You hand off a task. You message a Bot from desktop or phone, describing what you want done in plain language — no workflow builder to configure first. The Bot asks clarifying questions, drafts a plan, and starts working.
  2. The Bot works on its own computer. Each Bot runs on a dedicated cloud computer with a real browser, filesystem, and terminal — not on your device. It operates the tools and websites the job touches the way a person would.
  3. It runs end to end, 24/7. Because the work happens in the cloud, jobs keep running when your MacBook lid is closed. The Bot chains together the steps needed to finish and reports back when done.
  4. It returns only for approval. When human judgment is required — a purchase, a send button, a decision with consequences — the Bot pauses and asks you.

One design choice worth flagging: Grok Bot does not let you pick which underlying model runs a given task. It selects automatically based on the job. For most users that is a convenience; for technical teams evaluating it for production workflows, the lack of a model-selection mode is a real limitation.

What Can Grok Bot Do?

According to xAI's announcement and first-look walkthroughs from launch week, Grok Bot's capabilities center on multi-app, multi-step knowledge work:

xAI's own internal use cases — CRM cleanup, sales research, invoice processing, email drafts, bug reproduction — give a realistic picture of the sweet spot: repetitive, multi-step, cross-app office work where the cost of an error is low and the value of automation compounds.

Persistent Computer / Agent Capabilities

The technical core of Grok Bot is what xAI calls a persistent cloud VM: a virtual machine with a browser, a filesystem, and a terminal that stays running independently of whether your own device is on. This is the single biggest architectural difference between Grok Bot and most AI agents shipped before it.

Most agent tools today — including well-known coding and research agents — run inside a chat window on your device or in a sandboxed session that evaporates when the task ends. Grok Bot runs jobs on a dedicated cloud instance that persists. According to xAI's developer documentation, all of a user's Bots share one persistent cloud computer isolated to the account rather than to an individual Bot, which is what lets different Bots hand off files, browser sessions, and app logins to each other without repeating setup work.

Why persistence matters more than it sounds:

Grok Bot vs ChatGPT Agents

OpenAI's ChatGPT Agent already drives a virtual computer to complete multi-step tasks, and it remains the most direct comparison. Both products let an AI operate a browser and tools autonomously; the differences are in architecture and positioning.

Grok BotChatGPT Agent
ArchitectureFleet of named Bots on a persistent shared cloud computerAgent mode driving a virtual computer per task
PersistenceAlways-on; state survives sessions and device shutdownTask-oriented; session ends when the job is done
Multi-agentYes — group chats, chief-of-staff orchestrationSingle agent per session
Credential handlingSigns into your real apps via login handoff, including API-less sitesOperates in its own sandboxed environment
Model choiceAutomatic — no user model selectionTied to selected ChatGPT models
Entry priceFrom $120/seat/month (Cursor Teams Premium)Included with ChatGPT Plus at $20/month

The honest summary: ChatGPT Agent is cheaper and more proven for individual task automation, while Grok Bot bets on standing teams of agents with real account access. For a deeper look at how OpenAI's latest model family stacks up against Anthropic's, see our GPT-5.6 vs Claude comparison.

Grok Bot vs Claude

Anthropic's counterpart is Claude Cowork, and Claude models (Opus 5, Sonnet 5) remain the reasoning backbone inside many competing agent harnesses. According to launch-week coverage, Grok Bot's differentiation is again architectural — the persistent shared computer and the fleet model — rather than raw model quality. Independent efficiency data from Artificial Analysis on the underlying Grok 4.6 model (released the day after Grok Bot) shows it completing tasks in roughly half the reasoning turns of Claude Opus 5, which matters for long-running agent workloads, while Claude retains a lead on some coding benchmarks — see our Grok 4.6 vs Claude vs GPT-5.6 for coding breakdown and our Claude vs ChatGPT comparison for the model-level picture.

One Claude-specific consideration for August 2026: new Claude models launched on or after August 2 generate text with an invisible watermark under the EU AI Act rollout, while xAI has made no comparable announcement for Grok outputs. If output provenance matters for your workflow, that difference is worth factoring in — see our Claude AI watermark explainer.

Privacy and Security

An agent that holds your credentials and acts autonomously inside your accounts is powerful — and that is exactly why it deserves scrutiny before you deploy it on anything sensitive. The main considerations:

Practical guidance: start Bots on low-stakes, reversible work (research, drafting, data cleanup) before granting access to anything that sends messages, spends money, or touches production systems.

Pricing / Availability

Grok Bot is not sold as a standalone plan. Beta access is bundled into three existing paid tiers:

PlanPriceGrok Bot Access
SuperGrok Heavy$300/monthYes (beta)
Cursor Ultra$200/monthYes (beta)
Cursor Teams Premium$120/user/monthYes (beta)
EnterpriseWaitlistNot yet available

At launch, Grok Bot ships as dedicated apps for macOS and Windows, an iOS app (iOS 18+), and terminal access on Linux. Android is listed as "coming soon." Elon Musk posted that xAI will widen the beta once early issues are fixed, alongside the release of Grok 4.6 — which shipped on August 12, one day after Grok Bot's launch (see our Grok 4.6 review).

Who Should Use Grok Bot?

Good fit

  • Teams already paying for Cursor Ultra or SuperGrok Heavy — Grok Bot is included, so the marginal cost is zero
  • Founders and operators with repetitive multi-app workflows: CRM cleanup, sales research, invoice processing, email triage
  • Developers who want a standing agent for bug reproduction and issue triage across tools
  • Anyone who specifically needs work to continue overnight or across time zones

Poor fit

  • Anyone not already on a $120+/month xAI or Cursor plan — there is no affordable entry point today
  • Workflows that require a specific model choice or fine-grained control over agent behavior
  • Regulated or sensitive environments that cannot tolerate unaudited credential access
  • Users who need Android or iPad support today

Final Verdict

Grok Bot: 3.5/5 (beta)

Grok Bot is the most complete vision of "AI as infrastructure" shipped by a major lab: persistent cloud computers, fleet orchestration, real account access, and a messenger-style interface regular people can actually operate. The launch demos look like real delegation rather than autocomplete, and the internal-origin story lends credibility. But it is an early beta with no independent benchmarks, no published security audit, no model selection, and pricing that gates it behind $120–$300 monthly plans. If you already have access through Cursor or SuperGrok Heavy, it is worth putting a Bot on low-stakes recurring work this week. Everyone else should wait for the wider rollout — the category is moving fast enough that six months will sort the winners from the demos.

FAQ

What is the difference between Grok and Grok Bot?

Grok is the chatbot: you ask, it answers, the session ends when you close the tab. Grok Bot is a separate product — persistent, named agents that run on their own cloud computers, sign into your tools, and keep working between sessions. Per xAI's documentation, they are distinct products with separate apps.

Does Grok Bot work when my computer is off?

Yes. Because each Bot runs on a cloud computer rather than your device, jobs keep running when your laptop is closed or your phone is in your pocket. You can start a task from the iOS app and check results later.

Can I try Grok Bot for free?

No. There is no free tier or trial. Beta access requires SuperGrok Heavy ($300/month), Cursor Ultra ($200/month), or Cursor Teams Premium ($120/user/month).

Do Grok Bots remember things?

Yes. Bots keep stable preferences and role context between conversations and learn routines by watching how you work. However, xAI's documentation explicitly warns against treating Bot memory as an authoritative source — for anything that matters, ask the Bot to verify against current data.

How is Grok Bot related to Grok 4.6?

Grok 4.6 is the underlying frontier model released August 12, 2026, one day after Grok Bot's launch. Elon Musk tied Grok Bot's wider beta rollout to the Grok 4.6 release. See our Grok 4.6 review for the model details.

Who are Grok Bot's main competitors?

OpenAI's ChatGPT Agent, Anthropic's Claude Cowork, and Google's Gemini Spark occupy the same "digital labor" category, alongside earlier independent entrants like Hermes Agent and OpenClaw. For how the underlying models compare, see our coding model comparison and Gemini 3.7 Flash review.

Reviewed August 16, 2026 · Updated August 16, 2026 · ToolStep Editorial Team