OpenAI Dots vs Meta Muse vs Grok Bot
Compare OpenAI Dots, Meta Muse and Grok Bot using their launch announcements: computers, channels, connected apps, access and approvals.
OpenAI Dots, Meta Muse and Grok Bot are personal agents that can keep working after a message ends. Each vendor describes a computer, connected services and a way to bring work back for review. The useful comparison is the path into your existing tools, the channels where you talk to the agent, and the controls you can inspect. This is a launch-day comparison based on the vendors' own announcements, not a benchmark of task success.
| Question | OpenAI Dots | Meta Muse | Grok Bot |
|---|---|---|---|
| Vendor and home | OpenAI, in ChatGPT | Meta, in Muse and WhatsApp | Grok Bot, on x.ai and its own apps |
| Agents at launch | One primary dot; teams of dots are a future plan | One personal Muse in Meta's announcement | Multiple Bots can coordinate in a team |
| Computer | Its own cloud computer and browser | Muse Secure VM with its own browser | Its own computer |
| Main channels announced | ChatGPT, Slack and Teams; limited texting beta | Muse app and WhatsApp | Grok Bot desktop and iOS apps |
| Work while away | Scheduled tasks and proactive research | Continues projects and returns for changes or approval | Bots can work across apps and return for approval |
| Connected services | ChatGPT plugins and connected apps | Apps a person connects to Muse | Tools and accounts a person signs into for a Bot |
| Launch access | Gradual eligible Pro and Business Premium rollout; Enterprise beta requires an admin | US rollout on iOS, Android and web; free for most needs | Beta for eligible SuperGrok and Cursor subscribers on desktop and iOS |
| Launch usage | First dot included in eligible Pro and Business Premium plans; dot usage does not count toward eligible plan allowances for the first month | Most needs free, with subscriptions for more use | Bot usage is separate from Grok and Cursor plan usage |
The OpenAI Dots announcement, Meta Muse announcement and Grok Bot announcement support the broad rows above. Check each vendor's current plan page before using a launch article as a purchase guide.
Where Dots stand out
Dots live inside ChatGPT. The same product family supplies the conversation, plugins, Codex and ChatGPT Work tasks. OpenAI says a dot can hand work to Codex or ChatGPT Work, with those tasks counted against the usual usage limits. A dot can use a cloud computer, and local computer access is optional. You can reach it across ChatGPT, Slack and Teams after configuring those channels.
The controls are part of the story. OpenAI describes Custom Rules, Activity View, action review, separate sign-in handling and read-only proactive research. The Dots setup guide uses one documented handoff from WorkerKit: deliver a report to a personal email account connected to the dot.
Where Muse stands out
Meta presents Muse as a consumer personal agent available in its own app and WhatsApp. At launch, Meta says Muse is rolling out in the US on iOS, Android and web, with most needs free and subscriptions for more use. Meta describes a dedicated Secure VM, connected accounts, ongoing goals and a Sentinel check before outside actions. Muse can ask for approval before sensitive steps. Its consumer context and WhatsApp channel may be the practical draw for people already working there. Read our Meta Muse and AI workers guide for the WorkerKit path specific to that product.
Where Grok Bot stands out
Grok Bot's launch emphasizes a team of bots that can coordinate across apps, work on their own computers and come back for approvals. The announcement names eligible SuperGrok and Cursor subscriptions for its beta, with desktop and iOS apps. Our Grok Bot and AI workers guide covers the separate WorkerKit MCP setup used by that client. Check the vendor's current access and plan documentation for availability, since the launch offer can change.
Which one should you try?
Choose based on the workflows and channels you already use. If ChatGPT and its plugin ecosystem are central, start with Dots. If WhatsApp is your main personal channel, examine Muse. If you want the Grok Bot desktop experience, test Grok Bot against your real tasks. Give each one the same small job, then compare the work produced, the approvals requested and how clearly you can see what happened.
For a job that must run on a defined schedule with its own app grants, model, budget and record of each run, consider a separate WorkerKit worker. A personal agent can then interpret the result instead of carrying every recurring job inside one broad context. Our OpenAI Dots guide gives one concrete example. A direct WorkerKit plugin inside Dots has not been verified as of September 29, 2026, so the guide uses email delivery instead of claiming an MCP setup that may not work.
Compare one real task before choosing
A launch video shows what a product is designed to do. Your own workflow shows whether it does the job with the access and review process you need. Pick an ordinary task that takes 15 to 30 minutes and can be checked by a person. For example, ask each agent to prepare a briefing from a meeting invitation, recent email and a project document. Give each the same source access and the same definition of done. Then check the result, the actions it attempted and the controls you had to use.
| Test | Good evidence | What to record |
|---|---|---|
| Source access | The agent cites or links the messages and documents it used | Missing context and accidental access to unrelated data |
| Work completed | The briefing answers the same concrete questions for each product | Corrections needed before it is useful |
| Approvals | You can see what the agent proposes before consequential steps | Unclear drafts, duplicate actions or unexpected prompts |
| Continuity | It can resume the project after you leave and return | Lost context and repeated setup |
| Cost and limits | You can see which plan allowance the task uses | Whether the workflow remains practical at daily volume |
For a standing version of that task, a meeting prep worker is a useful separate baseline. It has a defined job, schedule and receipt, so you can compare a repeatable worker output with the broad personal-agent output. A different example is the follow-up tracker, where the benefit is catching a missed reply without asking a personal agent to search manually every morning. Review each kit's actual permissions before installing it.
How do safety controls differ?
All three vendors describe limits and approvals, but they use different mechanisms and language. OpenAI says dots have Custom Rules, app permissions, read-only proactive research and a separate action review. Meta describes Muse Secure VM and a Sentinel agent that checks outside actions, with permission before sensitive steps. Grok Bot's launch announcement says bots return when approval is needed. Those descriptions are not interchangeable security audits. They are starting points for the same hands-on test: inspect the work, ask what access the agent used and check whether it pauses before a consequential action.
The relevant comparison is also shaped by your organization. If access is controlled by a workspace admin, check whether the agent and each connected app are permitted there. If a task needs a narrow identity for just one job, a WorkerKit worker can hold its own app grants and cost ceiling. That does not replace the personal agent's safeguards; it gives the recurring job another explicit boundary. Our guide to rolling out an AI worker safely describes how to start with a small, reversible job.
FAQ
Is OpenAI Dots the same thing as ChatGPT Agent Mode?
OpenAI presents Dots as always-on agents that keep context and ongoing responsibility between conversations. Its launch article and help guide are the best source for the current Dots experience. Treat a dot as its own product surface rather than assuming every ChatGPT feature or limit behaves identically inside it.
Which one works in WhatsApp?
Meta's Muse announcement names WhatsApp as a main messaging channel. OpenAI's Dots launch names ChatGPT, Slack and Teams, with texting on a limited beta. Check current regional availability before relying on any channel for daily work.
Do all three have their own computer?
Their launch announcements each describe a computer for the agent. OpenAI calls it a dot's cloud computer, Meta calls it Muse Secure VM, and Grok Bot says bots have their own computer. That shared headline does not mean their browser, file access, sign-in or approval behavior is identical.
Which one should run recurring work?
Each vendor describes continuing work, and Dots and Muse explicitly describe work between conversations. Choose after testing the exact recurrence, permissions and review path you need. A separate worker schedule is useful when one job should have its own instruction, access and receipt.
Can they share one WorkerKit fleet?
WorkerKit exposes a fleet through its dashboard, API and MCP surfaces, but compatibility depends on each client. Our Muse setup and Grok Bot setup describe their documented routes. The Dots setup currently uses email delivery because direct WorkerKit plugin access in Dots has not been verified.
Sources: OpenAI Dots, OpenAI Dots help, Meta Muse, and Grok Bot.