Back to Blog

Grok Bot and Agentic Chat Teams Explained, How Persistent AI Teammates Work

In this article

• Quick Answer
• Key Takeaways
• What Is Grok Bot?
• How Do Agentic Chat Teams Work in Grok Bot?
• What Can a Team of Bots Actually Do?
• Which Grok Models Matter for Grok Bot?
• Who Can Use Grok Bot and What Does It Cost?
• How Secure Is Grok Bot for Business Use?
• Grok Bot or Custom AI Agents, Which Should You Choose?
• How to Pilot an Agentic Chat Team
• Limits Worth Knowing Before You Roll It Out
• How Unico Connect Helps
• Frequently Asked Questions
• Sources

This guide explains how Grok Bot works, what an agentic chat team is good for, what it costs, how the security model holds up, and when a business should build its own agents instead. Every fact was checked against SpaceXAI and Cursor documentation on 22 September 2026.

What Is Grok Bot?

Grok Bot is a desktop and mobile app for working with AI teammates. SpaceXAI describes a Bot as a durable teammate with a name, a job, its own conversation, and working context that develops over time. You message a Bot the way you would message a colleague, give it a task and access to the tools it needs, and it works through the job across apps and websites, coming back when something needs your approval.

SpaceXAI says it designed the product around five concepts and kept everything else beneath the interface.

1. Bots. Persistent agents with their own identity, memory, runtime, and tools.
2. Chats. The conversation you use to work with a Bot.
3. Prompts. Instructions that can be used once, saved as Skills, or triggered automatically as Routines.
4. Tools. Access to software through APIs, connectors, the shell, or computer use.
5. Artifacts. The documents, designs, code, and data a Bot creates or changes.

The desktop app runs on macOS, Windows, and Linux, and the mobile app runs on iPhone, iPad, and Android. Cursor, which SpaceX acquired on 14 August 2026, runs the cloud computers and meters usage on your Cursor account. Both SpaceXAI and Cursor publish the Grok Bot documentation, and Grok Bot is sold through both SpaceXAI and Cursor plans.

How Do Agentic Chat Teams Work in Grok Bot?

An agentic chat team is a group of Bots working toward one shared outcome in one visible conversation. The Grok Bot documentation describes the mechanics precisely.

The practical pattern is small. Give one Bot ownership of an end to end outcome, add a specialist only when a role is stable, and use a group chat when the handoff itself needs to be visible. The documentation gives the example of a website launch group with a launch coordinator, a content editor, and an analytics reviewer, where the coordinator assigns work and the group keeps every handoff in one thread.

What Can a Team of Bots Actually Do?

SpaceXAI says it built Grok Bot as an internal prototype first. Its launch post describes a sales Bot updating the CRM with call transcript notes and drafting follow ups, an operations Bot seating new hires and processing invoices that arrive in Gmail, and an engineering Bot reproducing a bug in the product, filing the ticket, and handing the fix to a debugging Bot.

At the enterprise launch on 3 September 2026, SpaceXAI said thousands of organizations had adopted Grok Bot, named Legora, Supermicro, and ServiceTitan as customers, and said the heaviest use is outside engineering. Its examples by function give a sense of where persistent agents fit.

• Sales. Researching target accounts overnight and leaving LinkedIn and email drafts for morning review.
• Recruiting. Building a morning shortlist with outreach queued for approval, and submitting interview scorecards from call recordings.
• Marketing. Pulling webinar questions and briefing each account executive on what their customers asked.
• Finance. A procurement Bot that monitors vendor spend and renewals, which SpaceXAI says surfaced tens of thousands of dollars in savings.
• Engineering. Watching pull requests for bugs, security findings, failing builds, and merge conflicts until each one is ready for review.

These are vendor examples rather than independent case studies, but the common thread is useful. Every one of them is multi step work that crosses several tools and ends in a draft or a change a person can check.

Which Grok Models Matter for Grok Bot?

SpaceXAI released Grok 4.7 on 21 September 2026 and says it trained the model to natively understand the Grok Bot harness, which makes it better at conversational tasks and general knowledge work. Grok 4.7 uses a new, larger base model than Grok 4.6, has a 500,000 token context window and a May 2026 knowledge cutoff, and costs 2 dollars per million input tokens and 6 dollars per million output tokens for prompts under 200,000 tokens. A fast variant runs at twice the output speed for twice the price. Grok 4.6, released on 12 August 2026, focused on long running agents.

Grok 4.7 wins two of the seven benchmark rows SpaceXAI published.

SpaceXAI reported Grok 4.7 at xHigh effort except on DeepSWE, where it used high effort, and the other two models at Max. Against Grok 4.6, reported at High effort and the same price, Grok 4.7 improved on every benchmark row, for example from 40.4% to 46.3% on CursorBench and from 20.3% to 38.0% on Terminal-Bench.

Fable 5.1 leads four rows, GPT-5.6 Sol leads DeepSWE, and Grok 4.7 leads on legal work, electrical engineering, and price. Inside Grok Bot you do not pick a model, because Cursor manages model selection and there is no model picker, so these API prices matter more for custom agents built on the Grok API than for Grok Bot plans.

Developers who want an agent team through the API rather than the app can use grok-4.20-multi-agent-0309, a beta model in which multiple agents collaborate in parallel on deep research tasks. It has a 1,000,000 token context window, supports function calling and structured outputs, costs 1.25 dollars per million input tokens and 2.50 dollars per million output tokens below 200,000 prompt tokens, and runs in the us-east-1 and us-west-2 regions.

Who Can Use Grok Bot and What Does It Cost?

Grok Bot has no separate subscription and no price per Bot. Access comes with a paid Cursor plan or a linked individual Grok or X subscription, and each plan includes weekly Grok Bot usage that resets every week.

SuperGrok Heavy and X Premium+ links also qualify, while SuperGrok Lite does not include Grok Bot. Plans do not stack, so linking SuperGrok on top of a Cursor plan adds no usage. When weekly usage runs out, Grok Bot stops until the reset unless on demand usage is turned on, in which case extra usage is billed through Cursor and counts toward your on demand monthly limit. Routines spend usage every time they run, and Cursor warns that an hourly schedule or a Slack listener on a busy channel can use a week of usage in a day.

A free trial is available as a usage credit with a seven day window, and it never turns into a paid plan. Enterprise customers enable Grok Bot through their Cursor account team, and SpaceXAI gave Grok and Cursor Enterprise customers two weeks of free usage from 3 September 2026. On 29 August 2026 it added an X integration, and paid Grok Bot users get free X API credits to start. SpaceXAI launched Grok Bot as a beta, so expect features and limits to keep changing.

How Secure Is Grok Bot for Business Use?

The security model is more detailed than most agent products publish, and it has clear boundaries.

• Isolation. Each user works in an isolated environment, and a Bot has no access by default, reaching only the accounts you sign it into. Between your own Bots, the computer is shared, so treat anything on it as available to every Bot.
• Connector tokens. Plugin OAuth tokens are held on the Cursor connector backend, and a Bot invokes tools without receiving the tokens.
• Enterprise controls. Network Controls, Team Setup, Action Recording, Allow Local Egress, enforced Auto Review, audit logs, OpenTelemetry export, the MCP allowlist, and SCIM are Enterprise only. Action Recording stores Bot actions, including scrubbed shell commands, for 90 days. Teams without a network policy default to allowing all destinations.
• Prompt injection. Auto Review checks Bot actions against the request when enforcement is on, outside content is marked as untrusted data, and network policy, per action approvals, and per user isolation sit underneath. Cursor states plainly that these controls reduce but do not eliminate the risk. Our guide to AI agent security and prompt injection covers the same defenses for custom agents.
• Hosting. Bots run only on Cursor hosted cloud computers. On premises deployment, deployment inside your own perimeter, and bring your own image are not supported today. The computers run in the United States today, and Cursor notes this is not the same as its US only data residency program, which does not apply to Grok Bot by default.
• Certifications. Anysphere, the company behind Cursor, holds ISO/IEC 27001 and ISO/IEC 42001 certifications issued by Schellman, and Grok Bot is in the current scope.

For regulated teams, the hosting line usually decides the question before anything else does. If customer data may not leave your own cloud account, Grok Bot is not an option today, however good the controls are. Our guide to governing AI agents at enterprise scale covers identity, access, and audit for agent fleets.

Grok Bot or Custom AI Agents, Which Should You Choose?

The honest answer is that they solve different problems, and many companies will use both.

Our guide to computer use or building the integration covers the first row of that table in depth. A sensible path is to pilot a job in Grok Bot to learn what the work actually involves, then decide whether it stays there or becomes a production agent inside your own stack. Our comparison of single agents and multi agent systems, our guide to the OpenAI Agents API, and our guide to AI agent development cost cover that second step.

How to Pilot an Agentic Chat Team

1. Pick one real job with a clear finish line. Choose work that spans a few tools, such as a weekly pipeline review that ends in drafts for approval.
2. Describe the Bot in operational terms. Put durable rules in the Bot description, such as never sending external messages without approval, and keep task details in the conversation.
3. Start with one owner. Add a second Bot only when a specialist role is stable, and keep the roster small.
4. Use a group chat for visible handoffs. Assign one owner per stage in the kickoff message so accountability is clear.
5. Keep consequential actions behind approval. On Enterprise, turn on enforced Auto Review, set a network allowlist, and send audit logs to your security tooling.
6. Measure before you scale. Track time saved, how often drafts need correcting, and how often a Bot hands a step back, then turn stable processes into skills and routines.

Limits Worth Knowing Before You Roll It Out

• It launched as a beta. Features and limits have changed several times since August 2026.
• Some sites resist automation. A site can block automation, expire a session, or require a human step, and the Bot hands those back to you.
• Memory is not a source of truth. The documentation advises asking a Bot to reopen current data for consequential decisions rather than relying on what it remembers.
• Deleting a Bot does not wipe the computer. Files and sign ins on the shared computer may remain after a Bot is deleted.
• Stopping does not undo. Telling a Bot to stop ends the work but does not reverse actions it already completed.
• Shared templates can leak configuration. A public template shows the Bot identity, description, skills, and routines, so remove keys and internal details before sharing.

How Unico Connect Helps

We build custom AI agents for companies that need them inside their own stack, with approval boundaries, audit logging, and evaluations designed in from the start. If a Grok Bot pilot proves a workflow is worth automating properly, our agentic AI development team can turn it into a production agent that runs where your data lives, and our AI development services cover the models, retrieval, and integrations around it. For sales, finance, and operations work like the examples above, our AI automation team builds the same workflows inside your own systems, and you can hire AI engineers to extend your team. To compare the market, see our list of top agentic AI development companies or talk to our team.

Frequently Asked Questions

Sources

Product facts come from the SpaceXAI announcements of Grok Bot on 11 August 2026, wider plan access on 26 August, the X integration on 29 August, Grok Bot for Enterprise and Designing Grok Bot on 3 September, and Grok 4.7 on 21 September. Limits and security details come from the SpaceXAI Grok Bot documentation and the Cursor Grok Bot documentation and Grok Bot security pages. Plan prices come from Cursor pricing and SpaceXAI pricing, usage and trial rules from the Grok Bot plans and billing and routines help pages, and API pricing from the xAI model documentation. All were read on 22 September 2026.