Claude Duet vs OpenAI Dots vs Meta Muse: Which One Should You Use?

Claude Duet, OpenAI Dots and Meta Muse are all 2026-era "proactive" AI agents that work in the background and message you only when something needs a decision, but they aren't interchangeable. Claude Duet, announced October 21, 2026, is a waitlisted research preview for US Claude Pro and Max subscribers built on Claude Opus 5.5, with its own standalone iOS/Android app. OpenAI Dots, launched at DevDay on September 29, 2026, lives inside ChatGPT for Pro and Business Premium users and can only read your connected apps until you hand it a task. Meta Muse, live since September 8, 2026, is the most consumer-facing of the three, with a free tier, its own app, and a direct line inside WhatsApp. Below is how each one actually works, what it costs, how much it can do without asking first, and which one fits your situation.

What Is Claude Duet?

Claude Duet is Anthropic's first standalone "always-on" agent app, unveiled as a research preview on October 21, 2026. Instead of waiting in a chat window, Duet runs on its own cloud virtual machine with its own browser, so it keeps working whether or not your phone or laptop is on. Each morning it sends a short brief on what needs your attention across your connected apps, then checks in every few hours, messaging you only when something actually requires a decision. You can hand it ongoing work — scheduling meetings, building a dashboard, running a forecast, putting together a prototype or a simple web app — through one continuing thread, by text or by voice.

Duet is built on Claude Opus 5.5 and carries over the memory, skills and connected apps already tied to your Claude account. Anthropic says it designed Duet specifically for professionals and small-business owners rather than casual chat users, and the product page frames it as "a partner you can trust."

What Is OpenAI Dots?

OpenAI introduced Dots at its September 29, 2026 DevDay as always-on agents that live inside ChatGPT. Rather than a separate app, a "dot" is a persistent agent you spin up inside your existing ChatGPT account; it gets its own cloud computer and browser that keeps running between your conversations. When you haven't given it an explicit task, a dot goes looking for ways to help on its own — what OpenAI calls proactive research — but during that unprompted phase its access to your connected apps is strictly read-only. It cannot send a message, change content, or touch a browser or computer until you give it a specific job or approve a specific action.

Dots reportedly connect to more than 4,000 third-party apps through OpenAI's plugin ecosystem. Our earlier explainer on OpenAI Dots access, price and how it works covers the rollout mechanics — including the desktop-only setup requirement and the EEA/UK/Switzerland exclusion at launch — in more depth.

What Is Meta Muse?

Meta Muse is a personal AI agent that Meta describes as doing the work rather than only answering questions. Announced September 8, 2026 and framed by Meta as "the world's first personal AI agent built for everyone," Muse runs on a dedicated cloud computer Meta calls the Muse Secure VM, which holds both the agent and your data in an isolated environment. It's built on a model Meta calls Muse Spark — a separate product from the Muse Wonder model family Meta's research lab ships through its developer API, which we cover in our look at Meta's Muse Wonder models.

Muse can send emails, book travel, fill out forms, open a browser and negotiate on your behalf, and it keeps working after you close the app, resurfacing when something changes. Meta also pitches it as a planning tool that remembers details mentioned only once — turning a saved Instagram recipe into a grocery list, for instance, or recalling a friend's dietary restriction weeks later.

Access, Waitlists and Pricing Compared

None of the three is available everywhere yet, and none is a universal free-for-all. Claude Duet is the most restricted: it's a waitlisted research preview, open only to US-based Claude Pro and Max subscribers, with Team and Enterprise plans not yet included — the Claude Duet help-center guide walks through joining that waitlist. There's no separate Duet fee — it shares your existing Pro or Max plan's usage limits, and Duet's background checks count against that same allowance.

OpenAI Dots ships to ChatGPT Pro and Business Premium users in eligible markets; Pro access currently excludes the EEA, Switzerland and the UK, and Standard Business workspaces are not included (only Business Premium is). Enterprise, Edu and Healthcare workspaces can try a beta version, but it's off until an administrator turns it on. There's no separate Dots subscription at launch, either.

Meta Muse is the broadest of the three on paper: it's rolling out in the US on iOS, Android and the muse.ai website, plus WhatsApp, with no waitlist and no technical setup required. Meta says Muse is "free for most of what people need," with paid subscription plans for people who want to do more — though Meta's own announcement does not publish the exact price of those paid tiers.

Autonomy and Approval Gates: How Much Control Do You Keep?

All three products are built around the same basic promise: the agent can act on its own, but it stops and asks before anything consequential. The differences are in where that line sits.

Claude Duet shows you what it plans to do and waits for approval before sending a message, making a payment, signing a document or deleting anything — and some actions, like trading securities or moving money between accounts, are off-limits even with your approval. OpenAI Dots draws its line earlier: unprompted "proactive research" is read-only by design, and you can set rules per action — run automatically, run only if pre-approved, always ask first, or hand the decision back to you — with sensitive actions such as changing a password always staying with you. Meta Muse uses a dedicated internal reviewer it calls Sentinel, which must approve anything Muse sends out to the internet, and it keeps a running audit trail of what it has done and what it's about to do; you can also set per-connection permissions, such as allowing Muse to read your email but not send from it.

Privacy and Security: Who Sees Your Passwords?

Credential handling is one of the more carefully worded parts of each announcement, because all three agents need to log into other services on your behalf. Anthropic says Claude Duet "never directly sees your passwords or sign-in tokens," following the credential design it already uses for its managed agents, and backs the launch with red-team testing, an external penetration test, and a public bug-bounty program; a dedicated classifier and safety system prompt are also meant to screen for sensitive data, rejecting requests for the most sensitive categories like credit card or government ID numbers outright.

Meta makes a near-identical claim for Muse: credentials sit in secure storage that the agent can use without seeing, Muse has no visibility into passwords or payment methods, and it cannot see passwords typed into its own browser. Meta also says Muse doesn't share your conversations or VM data with its ad systems, and it has a fully encrypted "Muse Confidential VM" planned for later this year that Meta says even it won't be able to access. OpenAI's framing for Dots is similar in spirit — saved credentials can be used for supported sites without exposing them to the model — paired with the hard read-only limit on unprompted activity and the ability for OpenAI to pause or stop a dot's work if it flags a safety concern.

Connected Apps and Integrations

Reach matters as much as caution here, since an agent is only as useful as the accounts it can touch. Claude Duet draws on Anthropic's Claude Marketplace, with more than 6,000 connectors and plugins and named integrations including Gmail, Slack, Intuit QuickBooks, Stripe, Gusto, LinkedIn, Docusign, HubSpot, Canva and Shopify; purchases run through Stripe's Link wallet or Shop Pay, and 1Password sign-in support is listed as coming soon. OpenAI Dots is reported to reach more than 4,000 apps through its plugin ecosystem, with Slack and Microsoft Teams also able to surface a dot directly. Meta Muse's named connections are narrower at launch — email, Instagram, WhatsApp and Link by Stripe for payments — with 1Password support also listed as coming soon.

Claude Duet vs OpenAI Dots vs Meta Muse at a Glance

Which Proactive AI Agent Should You Use?

If you're already paying for Claude Pro or Max in the US and want an agent tuned toward work tasks — invoices, scheduling, dashboards, documents — Claude Duet's approval-gated workflow and 6,000-connector marketplace make it the strongest pick once your waitlist invite lands, with the caveat that it's still a research preview that can misjudge urgency. If you live in ChatGPT already and want the most conservative default — an agent that can only look, not touch, until you explicitly let it act — OpenAI Dots' read-only proactive mode is the more cautious design, especially useful if you're easing into letting an AI agent near your accounts. If you want something consumer-simple with no subscription required to start, and you're comfortable with Meta handling your email, calendar and payment flows through WhatsApp or a dedicated app, Meta Muse's free tier is the lowest-friction entry point of the three — though it's also the one asking for the broadest everyday access to your life, from inboxes to payment methods.

What Happens Next

All three companies describe their current releases as early. Anthropic says it will publish more safety design detail on Claude Duet "in the coming days" and plans to widen access beyond the initial US Pro/Max waitlist. OpenAI has signaled it wants to let multiple dots work together as a team and expand Dots into the markets currently excluded. Meta has already committed to a more locked-down "Muse Confidential VM" later in 2026 and continues to add connected services beyond its initial email/Instagram/WhatsApp set. Expect the gap between "read-only until approved" and "acts first, approve when it matters" to keep narrowing as each company collects more real-world usage data — and expect pricing and regional availability, still unsettled for all three, to be the next thing to watch.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage