Free Grok Bot Workshop on Nov 7: How to Set Up a 24/7 AI Agent Safely

Always-on AI agents went from a demo to something you can install in a few months. Grok Bot, xAI's team of named agents with their own cloud computer, launched in beta on August 11, 2026. Within weeks, OpenAI, Meta and Anthropic shipped their own versions.

That speed is the problem. People are connecting an agent to their inbox before they have decided what it may do on its own. So on Saturday, November 7, 2026 at 6:30 PM IST, we are running a free 90-minute live session on exactly that: how to set up Grok Bot, and how to do it safely. Register on the workshop page, it takes an email address.

TL;DR: what the free Grok Bot workshop covers

Agents that act on your behalf need rules. See how explainx helps secure your agents →

What is Grok Bot, and why isn't it just @grok?

Grok Bot is a separate app from the Grok chatbot. xAI, now operating as SpaceXAI, describes it as "your team of always-on agents. They have their own computer, work inside tools and apps like you do, and keep working 24/7." One explainer put the difference in five words: Grok talks, Grok Bot does.

Here is the mechanism, in plain language:

• A shared cloud computer. Every Bot you create shares one persistent computer with a browser, files and a terminal. Each Bot gets its own screen. xAI's FAQ says isolation is per user, not per Bot.
• Plugins, browser and takeover. Bots act through plugins such as Gmail, Slack and Notion, through the browser for sites with no API, and through "computer takeover" when a login, 2FA code or CAPTCHA needs you.
• Routines. A routine is a scheduled or event-triggered task. A Bot can own up to 50, at least five minutes apart, and every run uses your allowance.
• Memory. Bots remember preferences and conversations. The docs warn that memory can go stale.

It is also not where some people expect it. Grok Bot does not run inside Telegram or WhatsApp. You reach it through its desktop and mobile apps, Slack, X (by tagging @bot), and email. We covered one recent addition in Grok Bot getting its own email address.

What we build in the live demo

The demo is deliberately boring, because boring is the safest place to start. We build a daily briefing routine: read-only, scheduled, no sending, no buying.

The prompt we work from looks like this:

Every weekday at 7:30 a.m. my time, send me a briefing with no more than
five items on AI agents and my industry. Use these sources. Each item gets
one sentence and a link. If nothing important happened, say
"Nothing material today." Do not pad it.

The last line carries the most weight. Without a "nothing to report" rule, an agent that must produce a briefing tends to invent one on a quiet day. In the session we will:

1. Create a Bot and give it its standing boundaries in the first message.
2. Set the schedule and check the timezone, because routines use it.
3. Run a test and read the output critically.
4. Correct it in plain language ("shorter, skip funding news") and watch the Bot remember.

Run a read-only routine like this for a week before you add anything else. It teaches you how your Bot behaves while the cost of a mistake is a bad summary.

Safety first: the rules we set before connecting anything

This is the longest part of the session, and the reason we are running it. The method is simple: read-only first, drafts second, approved actions last.

Door ajar with a small knocker, an approval prompt before an agent acts, for Grok Bot Ask-first rules

Ask-first rules

Grok Bot's Auto Review checks shell commands, app calls, computer use and delegation before they run. xAI is clear that it is model-based and "should complement, not replace, least privilege." It also does not review memory writes.

So we add narrow Ask-first rules for the five verbs that cause damage: send, publish, delete, purchase, and change anything in production. That mirrors xAI's own guidance. Two details are worth knowing. Approvals that come from a routine or another Bot expire after about ten minutes, and the action then does not run. And emails sent from routines wait as drafts.

Least privilege, because Bots are not a boundary

xAI's docs say not to use separate Bots as a security boundary. Because every Bot shares one computer and one set of logins, a Bot you never meant to give your inbox can still see it. The fix is to not connect the account at all. Connect read-only sources first, drafting tools second, and money last.

Prompt injection

Browsing agents can be hijacked by hidden text in a page or email that says "ignore your instructions." Palo Alto Networks' Unit 42 reported on March 3, 2026 that indirect prompt injection "is no longer merely theoretical but is being actively weaponized." Our defaults:

• Keep Bots that read the open web read-only.
• Put Ask-first rules on every outbound action.
• Never tell a Bot to "do whatever the email says."

We walked through why this matters in a disputed viral case about a Grok Bot posting bank balances to Slack: whether or not the claim holds, the permission lessons do.

Saved logins and cards

Since October 23 a Vault stores logins, cards and contact info that Bots can fill in. Our read on how the Vault handles payments is that you should keep cards out until your purchase rules are in place, and use a low-limit or virtual card when you do add one.

Cost

The usage pool is weekly and shared with Grok chat and Cursor. Users report that long jobs and frequent routines burn it fast. Set a monthly extra-usage limit before any long job, run fewer and smarter routines, and use plain Grok chat for one-off questions.

Green key fitting a narrow opening in a fence, least-privilege permission boundary for a 24/7 AI agent

If you want a layer that watches agent runs and stops a bad step before it lands, AgentBeam is the agent security platform from the explainx.ai team.

How does Grok Bot compare with Claude Duet, ChatGPT dots and Meta Muse?

About 15 minutes of the session is a side-by-side. The short version, as of October 2026:

Availability differs by region: Grok Bot's @bot tagging is not yet available for accounts based in the UK, the EU, Australia, Iceland, Liechtenstein or Norway, and Duet is a US-only waitlisted preview. Prices and limits change often, so we state the date for each number.

For the long version, read our four-way proactive agent comparison. On the safety side, see what happened when ChatGPT dots sent an unapproved email and the agent privacy promises of Muse and dots. If you would rather run your own, there is an open-source Grok Bot alternative too.

Who should join, and who should skip it

Join if you use ChatGPT, Grok or Claude and want to move from chatting to delegating recurring work. Join if you are a creator or founder who wants an agent watching X, your inbox or your reports. Join if you already have an agent connected and are not sure your guardrails are enough.

Skip it, or just join for the Q&A, if you already run several Bots and routines. The first hour is an introduction.

Nothing here promises income or "fully automated" anything. These are betas. They make mistakes, users report outages and stuck computers, and the interface changes weekly. We say so in the session and tell you which claims are confirmed and which are only announced.

How to join

1. Open the Grok Bot workshop page and register with your email.
2. Optional: have a Grok Bot plan or its free trial ready if you want to follow along.
3. Bring one recurring task you would like an agent to do. We will use a few for live feedback.
4. If you cannot attend live, register anyway and the recording will be sent.

You can also see other live sessions on the explainx.ai workshops page.

Related reading

• Claude Duet vs OpenAI dots vs Meta Muse vs Grok Bot: which proactive agent
• Grok Bot gets its own email address: what is verified
• Grok Bot Vault: saved logins and cards
• Disputed Grok Bot bank balance story and the permission lessons
• ChatGPT dots and the unapproved email
• Agent privacy promises: Meta Muse and OpenAI dots
• Open-source Grok Bot alternative

Official sources: xAI's Grok Bot launch post and the Grok Bot documentation. Palo Alto Networks Unit 42, "Fooling AI Agents: Web-Based Indirect Prompt Injection Observed in the Wild," March 3, 2026.

Pricing, regional availability and feature details are accurate as of October 31, 2026. Grok Bot changes weekly, so check the current docs before you rely on any number here.