What Is Grok Bot? How xAI’s Persistent AI Teammates Work

Grok Bot is not the normal Grok chatbot with a new name. It is xAI’s system for creating named AI teammates that can keep their context, use a cloud computer, work across apps, coordinate with other Bots, and run repeatable routines. The product is built around a simple idea: assign one Bot a clear job, give it the tools and boundaries needed for that job, and improve its process over time. Instead of starting from zero in every chat, the same Bot keeps its role, preferences, files, browser sessions, and summaries of earlier work. That persistence makes Grok Bot useful for work such as account research, campaign monitoring, expense reconciliation, bug reproduction, and daily operating summaries. It also raises practical questions about access, approvals, data, and what happens when several Bots share the same computer. Grok Bot in one minute Question Answer What is Grok Bot? A product for creating persistent AI teammates with names, jobs, memory, tools, and a cloud computer. Is it the same as Grok on grok.com? No. xAI’s FAQ explicitly separates Grok Bot from the standard Grok website and mobile apps. What can a Bot use? A browser, filesystem, terminal, connectors, files, and supported websites. Can several Bots work together? Yes. Bots can work in parallel, message one another, use group chats, and hand tasks to one another. Does every Bot get a separate computer? No. Your Bots share one cloud computer, including files, browser sessions, and app logins. Each Bot has its own screen for computer-use work. Can it automate recurring work? Yes. A successful workflow can become a reusable skill, then a scheduled or event-triggered routine. Does it ask before sensitive actions? Grok Bot supports approval requests, Auto Review, secure handoffs, and explicit rules for actions that should stop for the user. Where does it run? The app supports macOS, Windows, Linux, iOS, and Android. Work continues on the cloud computer when your device is closed. Source: xAI’s Grok Bot overview. What exactly is Grok Bot? Grok Bot gives you a roster of AI teammates. Each Bot has a name, one main job, a description of how it should work, and context that develops as it completes tasks. A Bot can receive instructions through text, dictation, or voice chat. It can then use connected tools and websites to complete multi-step work. It keeps you updated, returns drafts and files, and asks for approval when the next action needs human control. The cloud computer is central to the product. It gives Bots a browser, filesystem, and terminal, so they can work in real tools instead of returning only a text answer. A Bot can continue working while your laptop or phone is closed. xAI describes these Bots as durable teammates. “Durable” means the Bot remains available after one task ends. It keeps stable role context, preferences, and summaries of earlier work. The goal is for its usefulness to grow as the working relationship develops. Grok Bot is different from the regular Grok chatbot xAI’s own FAQ makes this distinction clear. Area Grok on the web or mobile app Grok Bot Main experience General chat, reasoning, content, files, voice, and media creation Persistent, named AI teammates with defined jobs Work environment Chat and connected features A persistent cloud computer with browser, files, and terminal Context Conversation and available account context Bot-specific role context, preferences, and summaries that continue over time Automation User-led interactions and product features Reusable skills plus scheduled or event-triggered routines Collaboration A user talks to Grok Several Bots can coordinate, message, share a group chat, and hand off work Control model Product permissions and account settings Task boundaries, approvals, Auto Review, secure handoffs, and team policies Grok Bot is also different from the xAI API. The API lets developers build their own software with Grok models. Grok Bot is an end-user product for setting up and managing AI teammates without building a separate application. How Grok Bot works The basic flow has six parts. 1. Create a Bot around one responsibility You give the Bot a short name, a primary job, and instructions for how it should work. xAI recommends focused Bots instead of one catch-all assistant. For example, a product-performance Bot might be told to investigate performance questions, preserve links and screenshots, separate evidence from assumptions, and never change production settings. A focused role makes it easier to judge whether the Bot did good work. It also limits the information and tools that the Bot needs. 2. Assign a task through conversation You can type, dictate, or start a voice chat. A useful task tells the Bot: The outcome that should be finished The sources, apps, websites, files, or conversations that matter The actions it must avoid or ask about The format of the result The point where it should stop for review “Review this week’s paid campaigns, calculate the gap from target CAC, prepare budget recommendations, and draft a Slack update without changing budgets or sending the message” is much clearer than “manage paid media.” 3. The Bot works on the cloud computer Grok Bot’s cloud computer has a browser, filesystem, and terminal. The Bot can also use supported connectors. When no connector exists, it may use a website through the browser. The browser session can stay signed in, which reduces repeated setup. If a site requires a password, passkey, two-factor code, CAPTCHA, or another sensitive step, the Bot can hand control to you. 4. Several Bots can work in parallel Bots can message each other, share context in group chats, and pass ownership of a task. That allows a research Bot to hand evidence to a writing Bot, or an account-health Bot to ask a data Bot for a current metric. Parallel work does not mean unlimited computer access. xAI says each Bot gets its own screen on the shared cloud computer, and one Bot can run one computer-use task on its screen at a time. 5. You review results and teach lasting preferences When a Bot returns a result, you can correct the analysis, output format, or process. You can tell it which parts should become lasting preferences. For example: “Use this structure for future weekly reports: five findings, source links beside each finding, and a final section called Decisions needed.” 6. A reliable process becomes a skill or routine A skill is a reusable set of instructions for completing a task. It can capture the inputs, sequence, decision rules, checks, output format, and approval boundary. A routine tells one Bot when to run a workflow. It can run on a schedule or, where supported, after an event such as a Slack message or GitHub notification. xAI recommends starting with a one-time task, correcting it until the result is reliable, saving the method as a skill, testing it again, and only then creating a routine. The shared cloud computer is useful, but it changes the security model One detail deserves special attention: all of your Bots share one cloud computer. That shared environment includes files, browser sessions, and app logins. It makes handoffs easier because one Bot does not need to sign in again or request files that another Bot already produced. Each Bot keeps its own role context and conversation, while the user’s files, browser sessions, terminal, and app logins sit on one shared cloud computer. It also means separate Bots are not separate security zones. xAI tells users not to treat one Bot as a boundary from another Bot on the same account. If a sensitive file or authenticated browser session is present on the shared computer, it may be available to the rest of the user’s Bot roster. Teams should respond with process controls: Connect only the systems a workflow needs. Use scoped service accounts when a source supports them. Remove sensitive temporary files after the task. Sign out when a service should no longer be available. Revoke connectors that are no longer needed. Keep high-impact actions behind approval. Skills, routines, and teach-by-demonstration Skills and routines turn a useful Bot from a flexible assistant into a repeatable operating system for a narrow job. What a strong skill contains xAI recommends including: When the skill should be used Required inputs and access The steps to follow How to validate the result The expected output Actions that require approval Skills are available across your Bots, though a Bot still needs the relevant connector or signed-in session to use one. Teaching a workflow by demonstration Where the feature is available, you can demonstrate a browser process instead of describing every click. Grok Bot records visible computer interaction for up to ten minutes, then drafts a skill from that demonstration. The draft still needs review. One example may not show unusual cases, failure handling, or sensitive decisions. Add those rules before scheduling the process. Building a routine A routine needs more than a time and task name. Define: The Bot that owns the routine The schedule and time zone, or a narrow event trigger The current source of input The expected result The approval boundary What to do if a source is missing or stale xAI’s current documentation says a Bot can own up to 50 routines. The app keeps the 20 most recent run records for each routine. A routine’s test run performs real work. It may navigate websites, call tools, or change files. Use safe inputs and retain approval for writes while testing. Practical Grok Bot use cases xAI’s documentation suggests roles that own a repeatable result, not a loose area of interest. Sales outbound research A sales Bot can review a CRM list, score accounts against an ideal customer profile, find likely contacts, and draft personalised outreach. It should skip people already in an active sequence and stop before sending messages. The useful outcome is a review-ready list with evidence, not a large pile of scraped names. Paid-media monitoring A paid-media Bot can pull campaign spend and performance, compare results with monthly budgets and target acquisition costs, recommend changes, and draft an internal update. Budget changes should remain behind approval. A recommendation routine can run regularly without giving the Bot authority to move money. Expense reconciliation An expense Bot can match receipts from an inbox with entries in an expense system, flag missing categories, check policy exceptions, and draft follow-ups. Totals should reconcile to the source, and every exception should cite the relevant policy. Product-performance investigation A technical Bot can review dashboards, traces, release information, and source-control links to investigate a performance problem. It can return screenshots, evidence, and a ranked hypothesis list without changing alerts or production settings. Bug reproduction A bug-reproduction Bot can read an issue, test it in a staging environment, capture exact steps, record expected and actual behaviour, and prepare a small test case. It should avoid production customer data and use approved test accounts. Chief-of-staff digest A Bot can review approved Slack channels, email, calendars, meeting notes, and planning documents, then produce a source-linked list of changes and decisions. Feedback about what was useful and what was noise helps the digest improve. A marketing example for an Indian business Consider a marketing consultant managing campaigns for several clients. One Bot could own the Monday performance review. Its job would be to open approved ad and analytics accounts, read the current targets, compare the last seven days with the previous four weeks, flag campaigns that moved outside agreed thresholds, and prepare client-ready observations. The Bot should not change a budget, pause an ad, upload a creative, or message a client. Those actions stay behind approval. A practical first request could be: Review the campaigns listed in this sheet. Use the linked ad accounts and analytics views. Compare spend, leads, cost per lead, qualified leads, and conversion rate for the last seven days against the previous four-week average. Separate tracking issues from performance issues. Prepare a five-point client update with source links. Do not edit campaigns, budgets, spreadsheets, or messages. Stop when the analysis and draft are ready for review. Once that task is reliable across several clients, the process can become a skill. A Monday-morning routine should only be added after the Bot handles missing data, account errors, and date ranges correctly. Approvals and Auto Review Grok Bot supports several ways to keep important actions under human control. Boundaries in the request The first control is plain language. Tell the Bot what it may prepare and what it must not execute. xAI recommends explicit approval boundaries for sending messages, publishing, purchases, financial transfers, deletion, permission changes, production changes, and accepting legal terms. Action approval When an action needs approval, the conversation shows the proposed operation and its inputs. You can allow it once, save a matching permission where available, or deny it. An approval only controls the proposed next action. It does not undo work that has already happened. Auto Review Auto Review evaluates tool calls and computer actions before they run. You can create “Ask first” rules or narrow automatic-allow rules. If both match, “Ask first” wins. xAI warns against broad permissions such as allowing everything in the browser. Websites and tool behaviour change, and model-based review should support least-privilege access rather than replace it. Secure handoffs for secrets Passwords, passkeys, two-factor codes, CAPTCHAs, and payment confirmations should be entered by the user. Grok Bot can hand over its computer so the user can complete the step and return control. xAI says ordinary chat should not be used to send a password or one-time code. Supported secure secret requests mask the value and keep it out of the transcript and model context. Local-computer access The shared Bot computer runs in the cloud. Access to your own Mac or Windows computer is separate. The default local execution setting is “Ask every time.” Users and administrators can apply stricter settings. How to get started with Grok Bot According to xAI’s current documentation, access is available through: Every paid individual Cursor plan Cursor Teams A linked individual SuperGrok, SuperGrok Plus, or SuperGrok Heavy subscription The desktop app supports macOS, Windows, and Linux. Mobile support is available for iPhone, iPad, and Android. The setup process is: Install the Grok Bot desktop or mobile app. Sign in with the relevant Cursor account. Link an eligible SuperGrok account if that is the source of access. Create a Bot with a short name, one job, and clear working instructions. Assign a safe first task with a defined result and review point. Take over the browser for passwords, passkeys, codes, or CAPTCHAs. Review the result and state any preference that should last. Save the process as a skill after it works reliably. Add a routine only after testing current inputs, failure states, and approval behaviour. Grok Bot requires cloud data storage. xAI says accounts using Cursor’s Legacy Privacy Mode must move to a supported data setting before Grok Bot can start. What Grok Bot remembers xAI says a Bot can retain stable preferences, role context, and summaries of earlier work. Conversations and learned context stay separate for each Bot. Files, browser sessions, and direct handoffs move context across Bots because they share the cloud computer. That makes collaboration easier, but it is another reason to remove access and sensitive working files when a project ends. Memory should not replace verification. For a price, policy, deadline, customer status, or production fact, tell the Bot to check the current source. Limits to understand before automating work Grok Bot can reduce manual coordination, but it still operates inside practical limits. Websites can block automation. A site may restrict browser control, expire a session, or require a human step. A Bot can make a reasoning error. Persistent context does not guarantee current or correct information. All Bots share one computer. Separate Bot names do not isolate files, sessions, or credentials. Routines can repeat a mistake. A weak workflow becomes more costly when it runs automatically. Usage resets weekly. Access is included through eligible plans, but the amount of work still depends on the plan’s allowance. Cloud storage is required. Legacy Privacy Mode is not supported. External actions create more risk. Sending, publishing, purchasing, deleting, and production changes need tighter controls than research and drafting. The safest adoption path is simple: read first, prepare second, act only with a clear approval rule. Who should use Grok Bot? Grok Bot is a good fit when work has all or most of these qualities: It repeats often. It uses several websites, files, or apps. The output has a clear definition. A human can review the result. The process benefits from remembering preferences. The workflow can fail safely when data is missing. It is a weaker fit for unclear work with no reliable source, high-impact decisions that cannot be reviewed, or tasks where every case is materially different. Frequently asked questions Is Grok Bot the same as Grok on X or grok.com? No. xAI describes Grok Bot as a separate product for persistent AI teammates on a cloud computer. The normal Grok apps centre on the general Grok assistant. Can Grok Bot keep working when my laptop is closed? Yes. Work runs on the cloud computer. Closing the app or your device does not stop a background turn or routine. Can several Grok Bots work at once? Yes. They can reason, use connectors, work with files, and coordinate in parallel. Each Bot has its own screen on the shared computer. Does each Bot have a private computer? No. Your Bots share one cloud computer. Files, browser sessions, and logins on that computer may be available to all of your Bots. Can Grok Bot use any website? It can use many browser-based services, including some without a connector. A website may still block automation, require a sign-in, expire a session, present a CAPTCHA, or require a manual step. Can a Bot learn a process that I demonstrate? Yes, where teach-by-demonstration is available. The feature records visible computer interaction for a limited period and drafts a skill. You should add exception handling, approval rules, and validation before using it as a routine. How much does Grok Bot cost? xAI says access is included with eligible paid Cursor plans and linked eligible SuperGrok subscriptions. Usage resets weekly. Pricing and allowance details can change, so check the current plan page before subscribing. Can Grok Bot publish or send messages without asking? That depends on the task instruction, available tool, saved approval rule, Auto Review result, and team policy. For safer use, require approval for external communication, publishing, financial actions, deletion, permissions, and production changes. Sources xAI, Grok Bot overview, updated 21 September 2026. xAI, Get started with Grok Bot, updated 21 September 2026. xAI, Grok Bot use cases, updated 11 August 2026. xAI, Skills and routines, updated 14 September 2026. xAI, Approvals, security, and privacy, updated 28 September 2026. xAI, FAQ for Grok website and apps, updated 27 August 2026.