Module 1: Fundamentals

What Is Grok Bot

Lingxi Li, an engineer at SpaceXAI who builds Grok Bot with Grok Bot, describes it as a highly capable engineering intern with its own computers.

A Bot signs in to apps and websites and uses them the way you do, on a cloud computer that stays on. It finishes a job end to end and only comes back when something needs your approval.

A Bot has a name, a job, and a memory

According to SpaceXAI, a Bot is a durable AI teammate with four parts. It has a name, a job, its own conversation, and working context that develops over time.

The job is what separates one Bot from another. SpaceXAI recommends a separate Bot when the work has a distinct goal, set of tools, working style, approval boundary, or schedule.

The docs describe a job in operational terms. The example Bot on the Get started page is an engineer:

Name: Piper
Job: Product performance
Description: Investigate product-performance questions using our
observability tools. Preserve links and screenshots, separate evidence
from hypotheses, and return a short summary with the highest-impact
issue first. Never change production settings.

Lingxi Li runs five engineer Bots this way. Each one owns one area, such as the iOS app, the desktop client and CI/CD, or infrastructure.

They perform best inside a single domain. The specs and design principles they carry are sharpest there.

You work with a Bot by messaging it

You give a Bot work the way you give work to a teammate. You send it a task, the relevant context, and access to the tools or files it needs.

The Bot works across the apps, repos, and systems involved and posts updates in the conversation. When it needs an approval, it stops and asks you.

In Lingxi’s setup, a Bot starts a cloud coding agent and reads its transcript. It then reviews the proof attached to a PR and checks for failed CI runs or merge conflicts.

All of your Bots share one computer

Each Bot runs on a persistent cloud VM with a browser, a filesystem, and a terminal. The docs call this the Bot’s computer.

Every Bot on your account uses the same computer. You can reach it from the desktop and mobile apps.

The computer belongs to your account, not to an individual Bot. The docs are direct about the consequence. Treat a login or a file on the computer as available to all of your Bots.

Each Bot gets its own screen. Screens are separate work surfaces, not separate security boundaries.

What a Bot remembers

Named Bots keep memory, files, browser sessions, and preferences across turns. Context compounds. It does not reset on every task.

A Bot retains stable preferences, important facts, and summaries of its work. It does not replay every prior message.

SpaceXAI is clear that memory is not a substitute for an authoritative source. Their guidance:

A good first handoff

SpaceXAI suggests you start with a real task that touches several tools and has a clear result. Their example:

Pull this week's Strategic Prospects PG List from Salesforce. Skip
anyone already in a sequence. Research the top 5 accounts across the
web, Slack, Databricks, and Sumble, pull contacts, and draft LinkedIn
and email in my voice, and leave me drafts to approve by tomorrow
morning

After the Bot delivers, you review its work and give corrections. Then you turn the process into a reusable skill or a scheduled routine. The lesson on how to get started walks through your first task.

How Is Grok Bot Different

Six things separate Grok Bot from the agent tools you already use. Each one comes from the SpaceXAI docs.

Avi Chawla’s Grok Bot Masterclass sums up the design decision under all six in one line. State moved out of the context window and into a machine that stays on.

Think of it as an office. The cloud computer is the office, and each Bot is a coworker with their own desk.

There are different people, jobs, and notebooks, but one filing cabinet and one set of keys.

The same analogy explains the trade-off. Shared keys make handoffs free, and they also mean the roster shares one set of consequences.

The Masterclass frames it as a trade about who absorbs the operations. A self-hosted agent such as Hermes gives you inspectable files and real isolation, and asks you to run the machine. Grok Bot removes the maintenance and takes the isolation away with it.

What that means for your coding agents

Lingxi Li, an engineer at SpaceXAI, uses Grok Bot as the layer above the coding agent rather than a replacement for it. Each of their Bots can create cloud agents, read transcripts, review proofs attached to PRs, and send follow-ups.

Before Grok Bot, Lingxi could manage 15 cloud agents at a time by hand. Now a fleet of Bots manages more than 200 at once. The lesson on the outer-loop Bot shows how that setup works.

How It Works

This lesson explains how Grok Bot does its work. It draws on the SpaceXAI docs.

The computer runs in the cloud and keeps its state

Grok Bot works from a persistent cloud computer with a browser, a command line, files, and connected tools.

You can open Agent Computer from a conversation to watch the shared desktop. You can then close the app or the laptop, and cloud work continues.

SpaceXAI describes a durability contract. Three things survive updates and recovery. Everything else is replaceable.

Durable project files belong in /workspace. SpaceXAI does not publish the operating system, the image contents, or the machine specification.

Your local computer is separate. A Bot runs commands on it only when you enable that capability and approve each use.

Your Bots share sessions, files, and credentials

Every Bot on your account uses the same computer. They share browser sessions, files, and command-line credentials, and one Bot can continue from work another Bot saved.

A research Bot saves a file and a writer Bot picks it up. Nothing needs a second setup.

A Bot acts through a connector or the browser

A connector gives a Bot a structured path into a supported service. Connectors appear as Plugins in the app and belong to your account, not to one Bot.

The browser handles everything else. That means vendor portals, older internal tools, and anything without a clean programmatic interface.

SpaceXAI’s advice is connector first, browser as the fallback. A connector is often more reliable than clicks through a website.

In chat, type @ to attach a connector to a task and / to reference a saved skill.

The Bot hands you the keyboard for sensitive steps

When a Bot reaches a step it should not complete alone, it asks you to take over the computer. SpaceXAI lists five cases.

• A password or passkey
• Two-factor authentication
• A CAPTCHA
• A payment or identity check
• A site that explicitly requires a human

You take control, complete only the blocked step, and tell the Bot to continue. Do not paste passwords or one-time codes into chat.

A skill records how, a routine decides when

A skill is a reusable set of instructions for how to do a task. Skills are available across your Bots.

A routine belongs to one Bot and says when to run a workflow. It runs on a schedule or after an event, such as a Slack message or a GitHub notification.

The docs prescribe the order. Start with a one-time task, make it reliable, and save the method as a skill. Only then automate it.

For event-driven routines, SpaceXAI recommends a narrow matching rule, not a broad listener such as “every new message.”

A Bot can learn a workflow by watching you

Where Teach a task is available, you open a Bot’s computer view and describe the result. You perform the workflow once, then stop the recording.

The Bot turns up to ten minutes of visible computer interaction into a draft skill. It does not record audio.

Treat the result as a draft. SpaceXAI advises you to add the decision rules, failure handling, and approval boundaries that one example cannot show.

Bots talk through messages, groups, and handoffs

You message a Bot the way you message a teammate. The transcript shows tool activity, created files, questions, and approval requests next to normal messages.

A direct message from you takes priority over background work and can redirect the current turn. “Stop now” ends work at once, but it does not undo completed actions.

A group chat holds two to six Bots. Type @ and a name when one Bot owns the request. Write normally and the Bots decide who responds.

A Bot can also message another Bot directly. The receiver wakes, does the work, and replies later, and you see the handoff in the conversation.

SpaceXAI advises a single owner at each stage. Parallel handoffs create duplicate work and noisy updates.

Get Started In 15 Minutes

You need three things. An eligible plan, the Grok Bot desktop app for macOS, Windows, or Linux, and one app or website for a first task. You can also use Grok Bot on mobile.

Grok Bot requires cloud data storage. Accounts on Legacy Privacy Mode must move to a supported data setting first.

Install and sign in

Download the app from the Grok Bot downloads page, open it, and choose Get started. Sign-in finishes in a browser window, and single sign-on follows your usual organization flow.

On first use the app asks which tools you use. Those answers shape the first teammate suggestions and connect nothing by themselves.

Your cloud computer provisions in the background. The last step opens Meet a future teammate.

Create one narrow Bot

Pick a suggested teammate or choose Create your own. Give the Bot a short name, one primary job, and a description of how it should work. Piper is the docs’ example.

Focused Bots build more useful context than one catch-all Bot. Add more Bots later, when work splits into distinct roles.

Write a strong first request

According to SpaceXAI, a strong request has five parts:

Run a task that needs no login

The docs suggest you attach a document and try this. It gives a result in about five minutes and confirms the machine is healthy.

Summarize this document in five bullets. List every date, decision,
and open question in a separate section. Cite the page or section for
each item. Do not change the source file.

Run a task in a real tool

Open our analytics dashboard and compare new-user activation for this
week with the previous four weeks. Identify the largest step-level
change and draft a short investigation plan with links to the relevant
charts. Do not change any dashboards. Ask me to sign in if needed.

When the Bot reaches a login, it hands you the computer. You take control, enter the password, passkey, or two-factor code, and return control.

The session then persists for every Bot on your account. This is the moment the shared boundary becomes concrete.

For supported services, install a connector from Settings, then Plugins. It is the more reliable path where one exists.

Review the result and save what works

Ask the Bot to revise anything incomplete or in the wrong format, and name a lasting preference explicitly:

Use this format for future weekly reports: five bullets, source links
inline, and a final section called "Decisions needed."

Once a result is repeatable, ask the Bot to save the method as a named skill. Only after that create a routine.

Your first-session checklist

The steps above, in order. Item 7 is the sign that you have something to reuse.

Before you sign in to anything sensitive, read the lesson on what to watch out for.

What To Watch Out For

Seven caveats before you sign in to real tools. Most come from the SpaceXAI docs. The framing of the first comes from Avi Chawla’s Grok Bot Masterclass.

Separate Bots are not a security boundary

The Masterclass puts it in one line. Free handoffs and shared credentials are the same fact seen from two sides.

The SpaceXAI rule follows from that. Do not place a credential or file on the computer if another Bot on your account should not use it.

Keep local execution off

A Bot runs commands on your laptop only when you enable that capability and approve it. The setting lives under Settings, then General, then Agent, then Execution on Local Computer.

The Masterclass advises you to set it to never allowed, unless a Bot has a specific reason to work on your local files. This does not restrict the cloud computer.

Never paste secrets into chat

The takeover flow exists so passwords and one-time codes stay out of the transcript. For supported connections, a secure secret request masks the value and keeps it away from the model.

Some sites expire a session or ask for verification again. Have the Bot pause and notify you rather than attempt a bypass.

Browser automation inherits every fragility of the site

Where no connector exists, a Bot falls back to pixel-level control of the browser. A layout change, a new consent prompt, or a session timeout can break a workflow that ran fine before.

Prefer a connector when one exists. Re-test a routine after a website, connector, or source format changes.

Automate only after a task is reliable

A routine inherits every unstated assumption in the task it came from. Resist the urge to automate on the first good output.

A test run performs real work. Use safe inputs and keep write actions behind approval. A deleted routine is gone at once, with no undo.

Auto Review is model-based

Where enforcement is available, Grok Bot evaluates tool calls and computer actions before they run. Require Approval rules win over Always Allow rules.

Two caveats apply. The review is model-based, so the docs position it as a complement to least privilege, not a replacement. Personal rules live on the current desktop, so a second installation needs its own check.

Browser clicks burn tokens

Sam Sokolin, who builds Grok Bot at SpaceXAI, notes that repeated clicks in a web app burn a lot of tokens. The lesson on how to cut token spend covers the fix.

Before you sign in to real tools

The seven caveats as seven checks. The first three matter before the first login.