avday Anthropic opened a research preview of Claude Duet on 21 October. It’s a separate app where Claude works as an always-on agent across your email, calendar and work tools. It’s US-only and invite-only.
Duet is Anthropic’s answer to Meta’s Muse, OpenAI’s Dots and Grok Bot, which all launched since August. The four promise similar things, so the details matter if you’re choosing one: who can get it, how its usage is counted and what it will do without asking.
What Claude Duet does
Anthropic describes Duet as “a proactive agent that understands what needs your attention.” It “keeps working while you’re away, and waits for your ok when it matters.”
In practice, according to Anthropic’s announcement and help centre, it works like this:
- A morning brief. Each morning it looks across your email and calendar, picks what needs attention and can start drafting replies for you to review.
- Check-ins every few hours. It checks your connected accounts at set intervals, not continuously. Anthropic says checking nonstop “would use up your plan’s limits much faster”. The help centre warns that “something urgent can arrive between checks.”
- Its own computer. Duet runs on “a virtual machine with its own browser, so it doesn’t require your laptop or phone to be on.”
- Your existing Claude setup. It carries over your memory, skills and connectors. Anthropic names Gmail, Slack, Google Calendar, Google Drive, QuickBooks, Stripe, Gusto, HubSpot, Docusign, Canva and Shopify among them.
It runs on Claude Opus 5.5, and you can’t pick another model. Anthropic says it built Duet for “professionals and small business owners with more on their plates than they can track.”
What it asks before doing
Duet reads, drafts and prepares work on its own. The line is at actions with consequences. Anthropic says “before it sends a message, makes a payment, signs a document, or deletes anything, it shows you exactly what it’s about to do and waits for your approval.”
Some things are blocked even if you approve them: “trading securities or moving money between accounts”, plus accessing someone else’s account.
Payments go through Stripe’s Link and Shopify’s Shop Pay, using one-time cards. According to the help centre, “Claude isn’t given your full credit card number (PAN) or bank account number.” Passwords go into a separate form on your device, which is designed to keep them out of the chat.
Two limits are worth knowing before you connect anything:
- Disconnecting doesn’t erase what it learned. Anthropic’s help centre says disconnecting an app stops future access, but “memory of past information from earlier conversations is kept.”
- Prompt injection isn’t solved. The help centre says “no defense is perfect, which is one reason consequential actions need your approval.” We covered what can go wrong when agents get write access in OpenAI’s report on agents that weakened their own monitoring.
Anthropic hasn’t said anything specific to Duet about training on your data or how long it keeps it. Its help centre says Duet “uses your Claude account settings”. Anthropic says it “will share more details about how Claude Duet is designed for safety in the coming days.” We hadn’t seen that follow-up by 25 October.
Who can get Claude Duet, and what it costs
Duet is “available in the US only during the research preview”. You need a Pro or Max plan and an invitation from the waitlist, which goes out in waves. It isn’t on Team or Enterprise plans yet. Anthropic hasn’t given a date for the UK or anywhere else.
There’s no extra charge during the preview. But Duet draws on your normal plan limits, and its background work counts too. The help centre says “you may reach your limits sooner than with chat alone.” Claude Pro is $20 a month in the US if billed monthly. Testers get one full reset of their limits, usable until 30 November.
How it compares with Muse, Dots and Grok Bot
| Claude Duet | Meta Muse | OpenAI Dots | Grok Bot | |
|---|---|---|---|---|
| Launched | 21 Oct, preview | 8 Sep | 29 Sep | 11 Aug, beta |
| Who can get it | US Pro and Max, waitlist | US and Canada, free tier and paid plans | ChatGPT Pro and Business Premium | SuperGrok and Cursor plans |
| In the UK | No | Not stated | Business Premium only | Not stated |
| Model | Claude Opus 5.5 | Muse Spark | GPT-6 Astra | Not stated |
| Counts against your plan | Yes, including background work | Not stated | Chats no, Codex and Work tasks yes | No, its own usage |
Sources: Anthropic and Meta’s announcements, CNBC for Muse in Canada, x.ai’s Grok Bot post, and The Next Web and Engadget for Dots. We couldn’t read OpenAI’s own Dots post. The Next Web and Engadget both report that chats with a dot don’t count towards ChatGPT usage limits, but tasks it starts in Codex or Work do.
The usage row is where they differ most for heavy users. Grok Bot “comes with its own usage, separate from your Grok and Cursor plans”, according to x.ai.
Duet shares the limits you already use for Claude chats and, if you use it, Claude Code. One commenter on the Hacker News thread about the launch said that was a reason they had no interest in it.
Meta reported scale too. CNBC reported on 21 October that Mark Zuckerberg told analysts more than 10 million people in the US and Canada have set up Muse agents.
What developers can do with it
Nothing directly, for now. Anthropic hasn’t announced a Duet API or SDK. Duet uses the same connectors and plugins as the rest of Claude, though Anthropic says some, such as Slack, “work with limits in Claude Duet”.
Stripe says Duet also uses Stripe’s MCP server for invoices, payment links and refunds. If you’re building agents yourself, our guide to how developers code with AI agents compares how 15 engineers work with them.
Anthropic says it plans to bring Duet to Team and Enterprise plans “later”. It hasn’t given a date, or said when the UK will.
