TR
GitHub

Tarik Razine / Portfolio

ai

OpenAI Dots vs Meta Muse vs Grok Bot: How to Choose

OpenAI dots, Meta Muse and Grok Bot all work while you're away. Compare who can get each, what they do unattended, how they ask first, and what they cost.

· 9 min read
On this page9 sections

Three always-on AI agents shipped in seven weeks: Grok Bot on August 11, Meta Muse on September 8 and OpenAI dots on September 29, 2026. All three keep working after you close the app. So choosing between dots, Muse and Grok Bot is less about which model is smarter. It is about what each one is allowed to do while you are not watching, and how it asks before it acts.

You will probably pay for one, not three. Each one lives in a different ecosystem: your ChatGPT plan, Meta’s apps, or your Cursor/SuperGrok subscription. The wrong pick is not only a wasted subscription. It is an agent with access to your inbox and calendar that acts in ways you didn’t expect.

The short answer: pick dots if your work already runs through ChatGPT, Slack or Teams. Pick Muse for everyday personal tasks if you are in the US. Pick Grok Bot if you want a team of specialist bots and you already pay for Cursor or SuperGrok. Whichever you pick, require approval for anything that sends or pays.

This comparison is based on the vendors’ own pages and documentation, plus first-hand reports from people who used the products. I have not run any of the three agents myself. Sources were checked on September 30, 2026.

Dots, Muse and Grok Bot at a glance

OpenAI dotsMeta MuseGrok Bot (SpaceXAI)
LaunchedSeptember 29, 2026September 8, 2026August 11, 2026
How you get itIncluded in ChatGPT Pro and Business Premium; Enterprise off by defaultFree with a usage limit, paid subscriptions above itIncluded with Cursor Pro, Pro+, Ultra and Teams, SuperGrok, SuperGrok Plus and Heavy, and Enterprise
WhereEverywhere Pro is sold except the EEA, Switzerland and the UK; Business Premium worldwideRolling out in the US (Meta’s announcement)No region limit listed
Talk to itChatGPT on desktop, web and mobile, plus Slack and TeamsMuse app on mobile and Mac, or WhatsAppDesktop (macOS, Windows, Linux) and iOS
One agent or manyOne primary dot todayOne MuseA team of bots with a specialist for each lane
Its computerIts own cloud computer and browserMuse Secure VM, an isolated Linux computerOne persistent cloud computer shared by all your bots

Who can actually get each agent

Availability is the first filter, and the coverage has been sloppy about it.

Dots are not US-only. OpenAI’s help article says dots roll out “to Pro users in markets excluding the European Economic Area, Switzerland, and the UK.” The same article adds that dots “are also available to Business Premium users across all supported ChatGPT regions.” OpenAI’s docs list the Pro 100, Pro 200 and Pro 500 plans as “for users over 18” outside those regions. Enterprise is “rolling out worldwide,” but “off by default and must be enabled by a workspace administrator.” Your first dot is “included in your Pro or Business Premium plan at no extra cost.”

Muse is rolling out in the US. Meta’s announcement says “Muse is rolling out in the US on iOS, Android, and muse.ai.” Meta’s help center adds that the app and website “are not yet available everywhere.” Some press reports mention Canada, but Meta’s own pages don’t list it, so check the app store in your country. On price, the Muse page says it “is available for free with a usage limit,” with a paid subscription above that.

Grok Bot lists plans, not regions. Its FAQ answers “Who is Grok Bot available for today?” with a list of Cursor and SuperGrok plans, and no country exclusion. That is “no limit listed,” not a promise that it works in every country, so check your account before you commit.

So a user in the EU or UK can’t get Muse, and can’t get dots through ChatGPT Pro. Business Premium and Grok Bot remain.

What each agent does while you’re away

All three are marketed as working in the background. The unattended rules differ.

Dots split unprompted work from assigned work. When you are not working with it, a dot does “proactive research” using connected apps “with tools that are restricted to be read-only, which means that they can’t send messages, change app content, or control your browser or computer.” That read-only rule covers the dot’s own initiative, not everything it does. OpenAI’s docs are explicit: “Assigned recurring tasks can include actions you’ve authorized. Their permissions and schedule are separate from proactive research.”

Muse keeps going and checks in. Meta says Muse “keeps working after people close the app, and comes back when something changes or when it needs approval, like before it sends an email or makes a purchase.”

Grok Bot runs around the clock. SpaceXAI says bots run “in parallel, 24/7, even when your laptop is closed,” and “only come back when something needs your approval.” One detail matters for security: “Every Grok Bot shares one persistent cloud computer,” and “Isolation is per user, not per Grok Bot.” The docs add that shared computer files and sign-ins “are not isolated by Bot and may remain on the computer.”

How each one asks before it acts

This is the setting to review before you connect an inbox or a card.

Dots: four rule levels

OpenAI’s help article lists four behaviours for custom rules: “Take action without asking,” “Take action if pre-approved,” “Ask before taking action” and “Hand off to you.” Some tasks never move to the agent: “Certain sensitive tasks, such as changing a password, always stay with you.” One detail is easy to miss: “Disconnecting an app does not delete information your dot has already obtained from it.”

Muse: ask by default, with a separate gatekeeper

Muse “is designed to ask your permission before taking certain actions, like sending messages, making purchases, or sharing information with a connected app.” For each request you can “allow an action once, always allow it, or deny it.” Behind that sits a second agent: “Nothing Muse does reaches the internet unless the Sentinel approves it.”

Grok Bot: rules you write

In Grok Bot’s approval screen, “Allow once lets the Bot continue with that action, Always allow can save a matching rule, and Deny blocks it.” You can also write “Ask first” rules, which “always stop matching actions for you.” Sensitive actions “can go through Auto Review before they run”. Note the “can”: you set it up.

The trap is the same in all three: “always allow” is one tap away. After a week of approvals, it is tempting to allow everything, and the agent stops asking.

Paying for things

Muse has the most specific checkout design in the primary sources. Meta says “a one-time card number is generated at checkout so your real card is never seen by the merchant or your agent.” It also says eligible purchases are covered by Link’s purchase protections. Meta calls that “a first for AI agents”, which is Meta’s claim, not a verified fact.

Dots let you “set rules for what your dot can share, purchase, or access.” Grok Bot’s pages don’t describe a checkout flow at all.

What it costs to run

Dots and Muse follow their plan rules. Dot conversations “don’t count toward your ChatGPT usage limits,” but tasks it starts in Codex or ChatGPT Work “count toward your usage limits as usual.”

Grok Bot works on metered usage, and Cursor’s help pages explain it:

  • Paid access includes weekly usage that “resets weekly.”
  • When it runs out, work continues as on-demand usage billed through Cursor, if on-demand is on.
  • “A limit is not a hard stop in the middle of a run”: a bot already working can finish past your monthly limit.
  • The free trial “is a usage credit rather than a set number of days, though a 7-day window also applies.”

What early users ran into

Vendor pages describe the design. These first-hand reports show what happened in use:

  • Dots (Casey Newton, Platformer): his dot “also emailed my speaking agent (with my permission) to clarify a potential scheduling conflict in November.” Newton discloses that his fiancé works at Anthropic.
  • Muse (Daniel Howley, Yahoo Finance): when shopping, Muse moved forward with the purchase, “prompting me to hit the buy button”. It stopped at the payment step and handed it back to him.
  • Muse (Alex Kirshner, Slate): “Authorized to try, Muse failed; I took over and logged in, then passed the sticks back to the agent.”
  • Grok Bot (Sanghoon Kang, Cursor forum bug report): delegated cloud agents were running while the parent bot looked idle, so “I assumed the workflow had stalled and manually asked the CEO to inspect its delegated tasks.”
  • Grok Bot (jjcm, Hacker News): “since using Grok Bots my inference spend has 2-3x’d.”

Tommaso Nervegna (Sorted Pixels) raises a design point that applies to all three: “Every agent you add creates an approval queue, and an approval queue is a backlog with worse ergonomics.” More agents means more approvals waiting for you.

Which one should you pick

This is my recommendation from the documentation and the reports above, not a benchmark.

  • Pick dots if your work already runs through ChatGPT, Slack or Teams. You can message your dot in Slack and Teams, and its unprompted background research is read-only by design. In the EU or UK, you need Business Premium.
  • Pick Muse for everyday personal tasks such as shopping, bookings and errands, if you are in the US. It has the clearest default: ask before sending or buying, with a separate Sentinel checking what leaves the machine.
  • Pick Grok Bot if you want a specialist for each job and you already pay for Cursor or SuperGrok. SpaceXAI describes “a chief of staff” with “a specialist for each lane.” Its own docs warn that a job such as “General Helper” gives the bot less guidance. Budget for metered usage, and remember that all your bots share one computer.

Key takeaways

  • Availability decides first: Muse is rolling out in the US, and dots on ChatGPT Pro exclude the EEA, Switzerland and the UK. Grok Bot lists no region limit.
  • “Read-only” in dots covers its unprompted research, not tasks you assign it.
  • All three can ask before acting, and all three let you “always allow”. Review those rules before connecting email or payments.
  • Grok Bot is metered: weekly usage, then on-demand, and a run can finish past your monthly limit.
  • More agents means more approvals to handle. Start with one.

If you run coding agents rather than personal ones, my Herdr CLI walkthrough covers keeping several agents visible from the terminal. For a local, self-hosted route, see OpenClaw with Gemma 4 and Ollama.

Sources