Meta Muse and OpenAI dots are personal agents: each belongs to one person, works on a cloud computer of its own, remembers across conversations and acts without being prompted. Muse launched on 8 September 2026 as "the world's first personal AI agent built for everyone". dots launched on 29 September at DevDay as "remarkably capable, always-on agents built to handle everything". TechCrunch described dots as having a "bubbly, cartoonish persona, similar to Meta's recently released Muse agent".
The resemblance is real, but the two products are built for different lives. This comparison uses each company's own pages and press coverage as of 30 September 2026. For the other five products in the category, see personal agents in 2026, compared.
Muse keeps one person's agent and data inside an isolated VM with a separate permission service; a dot works from its own cloud computer and posts into shared channels that only its owner can direct.
Where does each agent run?
Muse runs in the Muse Secure VM, "a dedicated, virtual machine (VM) that houses both the agent and a person's data". Meta's engineering post describes an isolated Linux box with a file system, a terminal and a Chromium browser. Muse writes its own code and tools there, and launches swarms of subagents. The agent runs in a container that is separate from the security services around it. A service called Sentinel controls every connector action and all network egress. Muse only ever holds surrogate tokens, which are exchanged for real credentials at the network boundary. Meta has also announced a Confidential VM, encrypted "with a key only they hold, so not even Meta can access it".
A dot works from "their own cloud computer, their own browser, and the apps you've connected". You can open that computer and watch the dot work. If you allow it, the dot can also connect to your own PC. OpenAI says it can use your saved passwords without exposing them to the underlying model. So both keep secrets away from the model, but Meta has published far more about how.
What can each one reach?
Muse connects to email, calendar, Instagram and Facebook, and it can write its own connectors. Meta has built it around commerce. At launch it had the entire Shopify catalogue. Connect added Walmart, Best Buy, Sephora, Wayfair and Instacart, along with Notion, GitHub and Box, and payments through Shop Pay, PayPal and single-use Stripe Link cards. The Small Business edition, announced on 29 September, adds Asana, Canva, Figma, QuickBooks, Slack, Stripe and Zoom. Not every retailer is willing: Amazon blocked Muse from its site within two weeks of launch.
dots connect to what OpenAI says is more than 4,000 apps through its plugin ecosystem, including Slack and Google Drive. The bigger difference is where you meet the agent. A dot lives in ChatGPT on desktop, web and mobile, can be started from Codex, and works in Slack and Microsoft Teams, where it can post in channels. Muse lives in its own apps for iOS and Android, on muse.ai and in WhatsApp. Glasses support, a voice mode and a pocket device called Muse Charm have been announced.
Side by side
How do they ask before acting?
Muse "checks with the person before sensitive actions like sending an email or making a purchase". The default allows ordinary browsing and "stops for anything hard to undo". The dialog comes from Sentinel, directly in the app's interface rather than in your conversation with Muse. That matters because a prompt-injected agent cannot fake its own approval. Grants can be one-time, session-scoped, task-scoped, time-bounded or perpetual. Meta runs a bug bounty that pays up to $300,000, including up to $130,000 for a prompt injection that affects a single user.
dots begin with built-in rules for when to act and when to ask. You can add Custom Rules in plain language, such as "never send emails". The background research a dot does uses read-only tools, and a monitoring system can pause or stop it. OpenAI's approach is easier to shape. Meta's is more fully described. For all five products' approach to approvals, see allow once, always allow, deny.
Which one should you pick?
Pick Muse if the agent is for your own life: shopping, returns, bookings, inbox and calendar. It is also the better pick if you care most about how your data is isolated. It is free to start, and it reached number one on the US App Store ten days after launch. TechCrunch, citing Sensor Tower, counted more than 3.4 million downloads by 25 September.
Pick dots if you already pay for ChatGPT Pro or Business, and your work happens in ChatGPT, Codex, Slack or Teams. That is where a dot's context carries across. Much of what a dot does "was already possible through Codex", as TechCrunch put it. What dots add is an agent that stays on and comes to you with work "sometimes before you even think to ask", as OpenAI put it.
What neither of them does
Neither lets another person give your agent work. Muse has one owner and no sharing feature, and that includes the Small Business edition. A dot can post where your team can see it, but only you can direct it. That is the right default for an agent that holds your email, your cards and your memories. It leaves a real gap, though, when the agent worth sharing is not your concierge but the working setup on your laptop: the coding agent with the repository checked out and the credentials in place.
That is the gap muti fills. A room around the agent already on your machine lets people you trust send it work, each with their own permission. Pushes to a shared branch, deletions, spending and anything touching credentials always stop for you, whoever asked. See borrowing an agent.
Frequently asked questions
Is Meta Muse free?
Muse has a free tier. Meta says it is "free for most of what people need", with subscription plans, and Alexandr Wang told CNBC they cost $20 and $100 a month. It requires users to be 18 or older, and at the end of September it was available in the US and Canada.
Can OpenAI dots work in Slack?
Yes. At launch, dots work in Slack and Microsoft Teams as well as in ChatGPT and Codex. According to OpenAI's help centre, when a dot posts in a channel, the channel's other members can see its messages, but only the dot's owner can direct it.
Do Muse and dots run on my computer?
No, both run on a cloud computer. A dot can connect to your own PC if you allow it. Muse runs in its Secure VM, and TechCrunch reported computer use on the Mac. If you want an agent that works on files and tools that are already on your machine, see local-first agents vs hosted agents.
Which is safer?
On what each company has published, Muse. Meta has documented a separate permission service, surrogate credentials, read and write access split per connector, and a planned VM encrypted with a key only you hold. OpenAI describes rules, read-only background tools and a monitor that can stop a dot. Both ask before sending or buying. Neither can undo an action you approved.
Sources
- Introducing Muse, Meta: about.fb.com · How We Built Safety Into Muse: security.muse.ai
- Muse for Small Business: about.fb.com · CNBC on pricing: cnbc.com
- Muse downloads, TechCrunch: techcrunch.com · Amazon block, GeekWire: geekwire.com
- Introducing dots, OpenAI: openai.com · dots safety FAQ: help.openai.com
- dots launch, TechCrunch: techcrunch.com · Engadget: engadget.com
A concierge for your own life is one kind of agent. If yours is the coding agent on your laptop and other people need it, give it a room.
公開日