How Do You Use Grok Bot?
Grok Bot is a persistent, always-on AI agent from xAI, launched in beta on 11 August 2026. Each agent gets its own dedicated cloud computer. You use it by signing it into the tools you already work in, then demonstrating the workflow you want repeated instead of describing it in a prompt.
That last part is the product. Grok Bot is not a faster chat window. It is marketed as an AI teammate that keeps memory across sessions, coordinates with other agents, and runs 24/7.
Two things are worth knowing before you spend an hour looking for a signup button. It is subscription-only, with no public API and no free tier. And it is a different product from the free @grok chatbot on X, which causes constant confusion.
For the parts of your stack that do have free credit programs, AI Perks tracks $7.7M in credits across 194 companies.
Grok Bot, @grok and the Grok API Are Three Different Products
Only one of the three is Grok Bot. The other two are a free chatbot inside X and a paid developer API, and neither gives you a persistent agent.
| Product | What it actually is | How you get it |
|---|---|---|
| Grok Bot | Persistent agent with its own cloud computer, memory across sessions, 24/7 operation | Subscription tiers only. No API, no free tier |
| @grok on X | Free conversational assistant inside the X app | Free, no setup |
| Grok API | Developer API for building your own Discord, Telegram or internal bots | Paid access through the xAI console, priced per model |
If you searched "how to use Grok bot" hoping to build a Discord bot, the third row is what you want, not the first. Those are separate products with separate billing, and the naming overlap is the single biggest source of wasted time on this topic.
Grok 4.6, a separate release, shipped one day after the Grok Bot beta, on 12 August 2026. A model version and an agent product are two different announcements, and the two landed within 24 hours of each other.
Which Subscription Actually Unlocks Grok Bot
xAI's own page lists Grok Bot as included with SuperGrok, SuperGrok Plus, SuperGrok Heavy, and with Cursor Pro, Pro+, Ultra, Teams Standard and Teams Premium. Secondary coverage disagrees and describes a much narrower set of top tiers.
Here is the discrepancy stated plainly, because most articles pick one side and present it as settled:
| Source | What it says about access |
|---|---|
| xAI official product page | Broad. Included across SuperGrok, SuperGrok Plus, SuperGrok Heavy, Cursor Pro, Pro+, Ultra, Teams Standard, Teams Premium |
| Secondary press coverage | Narrow. Effectively top tiers only, with figures around $300/month for SuperGrok Heavy, $200/month for Cursor Ultra, $120/seat/month for Cursor Teams Premium |
Treat xAI's own page as authoritative. Then check the plan you are actually on, inside the product, before you upgrade on the strength of a number in any article, including this one. Tier lists for a beta product change faster than the articles describing them.
The Cursor bundling is not random. SpaceX acquired xAI in an all-stock merger announced 2 February 2026, and SpaceX and xAI acquired Anysphere, the company behind Cursor, in June 2026. That is why an editor subscription carries an agent product.
Platform availability at launch was desktop and iOS, with Android described as planned and enterprise access by waitlist.
Teaching Grok Bot a Workflow: Demonstrate, Do Not Describe
The core interaction is demonstration, not prompting. You show the agent the workflow, it retains what it learned in memory across sessions, and it runs the same thing later on its own cloud computer.
That changes how you should brief it. A prompt describes an outcome. A demonstration encodes the tools it signs into and the order of operations.
Start with one bounded loop. Pick a task with a clear trigger, a clear finish, and a visible output you can check. Not "manage my inbox", but "every morning, pull yesterday's failed payments into this sheet".
Give it the real logins it needs, and nothing else. The agent signs into tools and websites on its own cloud computer. That is the useful part and the risky part. Scope credentials to the one workflow.
Run it supervised before you run it unsupervised. The whole pitch is 24/7 operation without supervision, which is exactly why the first few cycles deserve supervision.
Add a second agent only after the first is boring. Coordination between agents multiplies both the output and the failure modes.
One fair early criticism: a reviewer noted the automatic model router "wasn't great" in the early beta. That is a UX complaint about a beta, not a reason to avoid the product, but it does mean you should verify which model handled a task rather than assume.
What to Hand an Agent First, and What Not To
The best first task is repetitive, low-stakes, and produces output you can verify in under a minute. The worst first task is anything irreversible.
| Good first workflow | Poor first workflow |
|---|---|
| Daily report pulled from two dashboards into one doc | Anything that sends external email |
| Monitoring a queue and flagging anomalies | Anything that moves money |
| Reformatting and filing incoming documents | Anything that posts publicly |
| Cross-checking two systems for mismatches | Anything with no verifiable output |
The pattern in the left column: the agent gathers and arranges, a human decides. That split survives contact with a beta. The right column does not.
Teams pairing agents with model APIs quickly run into a second bill, which is where AI Perks is useful.
No Access to Grok Bot? Use the Separate Grok API
If your plan does not include Grok Bot, the honest alternative is the Grok API, which is a different product with a different access model and is what you would use to build a bot yourself.
A few things that are currently true and worth knowing:
- A $25/month new-account API credit has been reported as current for xAI
- An optional data sharing arrangement provides a recurring $150/month credit in exchange for sharing API request metadata and outputs
- xAI has no dedicated startup credit program comparable to what OpenAI or Anthropic run
On token pricing, be careful what you read. Published figures vary widely across model generations and contradict each other from source to source. Pricing varies by model, so check the xAI console directly rather than trusting a table in an article, including this one.
That gap matters. The direct competitors in the persistent-agent category, ChatGPT Agent and Claude Cowork, sit inside ecosystems with substantial credit programs behind them.
| Stack layer | Credit programs available |
|---|---|
| Model APIs (Anthropic, OpenAI, Google) | Yes, several per provider. See what is available |
| xAI and Grok API | No dedicated startup program |
| Cloud and GPU compute | Yes, large programs. See what is available |
| Dev tools, data, observability | Yes, dozens. See what is available |
Agents are expensive to run because they run constantly. Covering the model and compute layers with credits is the difference between a pilot you can afford and one you quietly switch off.
Frequently Asked Questions
Is Grok Bot free to use?
No. Grok Bot has no free tier and no public API. It is included with certain xAI SuperGrok and Cursor subscription tiers, and enterprise access runs through a waitlist. The free @grok assistant inside X is a separate, simpler product and is not the same thing.
Does Grok Bot have an API?
No. Grok Bot is a subscription product with no public API. If you want to build a Discord bot, a Telegram bot or an internal tool, you use the separate Grok API, which is billed per model through the xAI console. Credits for other layers of that build are tracked at getaiperks.com.
How do I teach Grok Bot a workflow?
You demonstrate it rather than describe it. Grok Bot learns workflows by demonstration, retains memory across sessions, and repeats the work on its own cloud computer. Start with one bounded, verifiable loop, supervise the first runs, and expand only after it is reliably dull.
Which plan do I need for Grok Bot?
xAI's own page lists SuperGrok, SuperGrok Plus, SuperGrok Heavy and several Cursor tiers including Pro, Ultra and Teams. Some coverage describes narrower, top-tier-only access. Check your own plan inside the product before upgrading, since beta tier lists move faster than the articles about them.
Are there free credits for the Grok API?
A $25/month new-account API credit has been reported as current, plus a recurring $150/month credit for opting into data sharing. xAI runs no dedicated startup credit program comparable to those at OpenAI or Anthropic. Other providers do, and AI Perks tracks $7.7M in credits across 194 companies.
What are the alternatives to Grok Bot?
ChatGPT Agent and Claude Cowork are the direct competitors in the persistent-agent category. All three carry real running costs, so the practical question is which ecosystem gives you credits on the model and compute layers underneath. Compare programs at getaiperks.com.
Agents run constantly. Make the credits, not your card, pay for it.