ChatGPT Dots vs Grok Bot vs Meta Muse (2026)

Three tech giants shipped the same idea in seven weeks.

xAI’s Grok Bot arrived on August 11, 2026. Meta’s Muse followed on September 8, and OpenAI’s dots landed on September 29. Each promises an AI agent with its own computer that keeps working after you close the app.

The pitches sound identical. The products aren’t. They run on different models, cost anywhere from nothing to $100 a month, and treat your data in very different ways.

Here’s how ChatGPT Dots, Grok Bot and Meta Muse really compare. One of them lists an ad-related detail in its App Store entry that its launch post never mentions.

We haven’t used these agents hands-on yet. This comparison draws on each company’s docs and help pages, checked on October 10, 2026, plus independent hands-on reviews.

Three Agents, Three Different Jobs

ChatGPT Dots are always-on agents inside ChatGPT. OpenAI announced them at DevDay on September 29, 2026, and each one runs on GPT-6 Astra with its own cloud computer and browser.

A dot learns from feedback and keeps working between conversations. You get one per person for now, and you can reach it in ChatGPT, Slack or Microsoft Teams.

Grok Bot is xAI’s standalone agent app, built with Cursor. You create several named Bots that share one persistent cloud computer with a browser, filesystem and terminal.

The Bots work in parallel, message each other and can learn a browser workflow from a demonstration. It’s the same product we covered in Grok Bot vs OpenClaw vs Hermes vs n8n, now with much wider access.

Meta Muse is Meta’s personal AI agent, launched in the US and now available in Canada. It runs on Meta’s Muse Spark model inside a private cloud machine.

Muse handles errands like shopping, bookings and email. It’s free to start on iOS, Android, the web and WhatsApp.

In plain English: Dots is a coworker, Grok Bot is a small automation team, and Muse is a personal errand runner.

ChatGPT Dots vs Grok Bot vs Meta Muse at a Glance

ChatGPT Dots: The Always-On Coworker

(opens in a new tab)

A dot behaves like a teammate who never logs off. You hand it ongoing work, and it keeps making progress between conversations.

In OpenAI’s words, it “asks for your input when a decision needs you.” It researches, analyzes data, prepares documents and builds software, and it can pass bigger jobs to ChatGPT Work or Codex.

OpenAI says dots connect to more than 4,000 apps through plugins. You can also link one personal computer, as long as the ChatGPT app is open on it.

An auto-review step decides whether each action can go ahead, needs your approval or must be handed to you. Password changes and moving money always come back to you.

Early reviews are split. Casey Newton at Platformer called dots “highly capable” and said one did about two hours of work for 15 minutes of his time.

Wired found its shopping test “a bit buggy.” The dot failed a CAPTCHA and couldn’t cancel a subscription.

Grok Bot: A Small Team of Bots on One Computer

(opens in a new tab)

Grok Bot suits people who want several specialists rather than one assistant. You might run a research Bot, an inbox Bot and a reporting Bot side by side, sharing context in group chats.

Its standout feature is teaching by demonstration. You do a browser task once while the Bot records up to ten minutes of on-screen activity. Then you refine the draft skill and schedule it, with up to 50 routines per Bot.

Two details matter more than the marketing suggests. All your Bots share one computer, so files and logins are shared too. xAI’s Grok Bot FAQ says plainly: “Do not use separate Bots as a security boundary.”

The model is a moving target, too. Cursor’s docs say Grok Bot sends each task to whichever model it judges best, from xAI or other companies.

Elon Musk said in October that this includes Claude Opus 5.5. There’s no model picker, and you can’t see which model handled a request.

Meta Muse: The Errand Runner in Your Pocket

(opens in a new tab)

Muse is the consumer play. Meta says it “doesn’t just answer questions, it actually does the work.”

The examples are everyday ones: ordering groceries, booking travel, filling in forms, sending emails and even calling businesses for you.

Everything runs in a private cloud machine that Meta calls the Muse Secure VM. According to Meta’s launch post, a component called Sentinel must approve anything Muse sends to the internet.

Checkout uses one-time card numbers from Link by Stripe, plus Shop Pay and PayPal. Its connectors cover stores such as Walmart, Best Buy and Instacart.

People like it. Muse reached No. 1 on the US App Store, and Apptopia estimated 1.8M iOS downloads in its first 12 days.

Its first month was rough, though. Amazon blocked it as an “unauthorized AI agent,” and a researcher found a macOS flaw that Meta patched the next day. PCWorld advised against installing it on a Mac for now.

What Each One Costs

The prices look simple until you read the metering rules:

Two caveats apply. OpenAI says pricing for extra dots and usage terms after the first month are still to come, so today’s generous limits may not last.

Meta says its Muse subscriptions are “in limited testing.” It hasn’t said how many tokens a typical task uses.

Europeans have one route to dots: a Business Premium workspace. It’s rolling out worldwide, even though Pro excludes the EEA, UK and Switzerland. OpenAI’s dots documentation lists the current rules.

Privacy: What Each Agent Keeps About You

All three need deep access to be useful: your email, your accounts, sometimes your computer. What happens to that data matters more than any feature list.

Muse deserves the closest look. Meta says it “doesn’t share your conversations or the data in your virtual machine with Meta ad systems.”

Yet the app’s App Store listing names Third-Party Advertising as a purpose for data including purchase history, precise location, contacts and browsing history.

The label is self-reported and may cover app-level data rather than what sits in your Muse machine. So it isn’t proof of misuse, but it’s worth knowing before you connect your inbox.

Wired has also reported that Muse keeps “Relationships” profiles on people in your life, including people who don’t use Muse.

Dots and Grok Bot have gaps of their own, as the table shows. Our guide to privacy-first AI automation covers how to limit what any agent can reach.

How Websites See These Agents

Here’s the part most comparisons skip. Every one of these agents browses from a cloud computer, not your home connection, and websites notice.

Grok Bot’s security docs say its Bots use “shared static egress IPs,” with no dedicated address per customer. Those computers sit in the US.

Sites that judge trust by IP see a cloud address shared by many other users. It’s the same signal covered in our guide to how websites detect proxy traffic.

Dots browse in their own cloud browser, separate from yours, and hand sign-ins back to you through a private form. Muse routes every outbound request through Sentinel.

That control cuts both ways. Amazon blocked Muse partly because, it said, the agent doesn’t identify itself.

None of them try to beat bot checks, by design. xAI’s FAQ says the Bot should hand CAPTCHAs and other human checks to you “rather than bypassing them.” Wired’s dot simply failed a CAPTCHA.

Our guide to how anti-bot systems detect automated browsers explains why that’s getting harder. If you run several accounts, see how AI agents manage multiple browser profiles.

Region checks follow the same logic. Tagging @bot on X depends on the country X has on file for your account, not your IP, so a VPN won’t unlock it.

If you manage accounts across sites yourself, our proxy directory compares the residential IPs teams use to keep each one consistent.

Which One Should You Choose?

1Choose ChatGPT Dots If

You already pay for ChatGPT Pro or Business Premium and your work runs through documents, data and chat tools. Dots fits teams that want one dependable agent rather than a fleet.

2Choose Grok Bot If

You use Cursor or SuperGrok, like building automations, and want separate Bots for separate jobs. Accept that you won’t control which model runs.

Keep sensitive logins off the shared computer. Our Grok Bot vs Claude Cowork vs ChatGPT Work comparison shows how it stacks up against other work agents.

3Choose Meta Muse If

You’re in the US or Canada and want help with errands more than work. Read the privacy section above first, then turn off training under Settings, then Data controls, on day one.

4Wait If

You handle client, health or financial data you’re not allowed to share. All three agents are weeks old, and none yet offers the deletion controls or independent audits that kind of data needs.

Mistakes to Avoid with Always-On Agents

1Connecting Your Main Inbox on Day One

An agent with your primary email can read years of history. Start with a secondary account or a narrow folder, and widen access once you trust its judgment.

2Clicking Always Allow Too Early

All three offer some version of “always allow.” Keep approvals on for anything that sends, shares or spends. Give the agent a few weeks of routine work first.

3Assuming Delete Means Forgotten

Deleting a dot doesn’t delete the files it made, deleting a Bot can leave logins behind, and Muse may still remember what you removed. Treat anything you connect as potentially permanent.

4Expecting Agents to Clear Bot Checks for You

CAPTCHAs, 2FA prompts and payment checks come back to you by design. Plan workflows around those hand-offs, or they’ll stall overnight while you sleep.

5Ignoring the Bill After the Launch Window

Dots’ extended limits apply to the first month, Grok Bot overage is billed through Cursor, and Muse counts tokens rather than tasks. Check your usage page weekly for the first month.

Frequently Asked Questions

The Bottom Line

These three agents share a pitch and little else. Dots is the coworker for ChatGPT users, Grok Bot is a toolkit for builders, and Muse is the free errand runner for US and Canadian consumers.

None of them is finished. Start with low-stakes tasks, keep approvals on for anything that sends or spends, and connect your most sensitive accounts last.

For a wider view of agents that browse for you, see our roundup of agentic browsers for AI automation.