Source: ai-research/claude-blog-introducing-claude-duet-full-2026-10-27.md (launch post), ai-research/claude-duet-product-page-2026-10-28.md, ai-research/claude-help-get-started-with-claude-duet-2026-10-28.md, ai-research/claude-blog-claude-duet-safety-privacy-security-2026-10-23.md; earlier summary ai-research/claude-blog-claude-duet-2026-10-21.md; creator reviews in raw/ (see Early Reviews). Official launch video: raw/Introducing_Claude_Duet.md (see Launch video below).

Product: Claude Duet (research preview) URL: https://claude.com/resources/articles/claude-duet · product page https://claude.com/product/claude-duet Date: October 21, 2026 Availability: Waitlist, US Pro and Max subscribers; not on Team or Enterprise yet. iOS, Android, and claude.ai on a computer.

Claude Duet is Anthropic’s proactive agent app: it watches your connected accounts, sends a morning brief of what matters, and runs tasks on its own cloud computer, waiting for your approval before it sends, pays, signs, or deletes anything. Anthropic built it “for professionals and small business owners with more on their plates than they can track.” It runs on Claude Opus 5.5 and uses the memory, skills and connectors already on your Claude account.

Moved 2026-10-28 from claude-ai/ into its own section, Claude Duet. The hands-on reviews that were on this page now live in Early Reviews.


How it works

  • Built around Opus 5.5. Anthropic “designed it around Claude Opus 5.5” and credits its judgment: deciding “what needs you, what can wait, and when to check with you first.” You can’t choose a model; Duet “decides how much effort a job needs” (help center).
  • Your Claude context carries over. Memory, skills and connected apps from your Claude account are there “from your first message”, so setup takes “minimal work”. Skills that exist only on your computer do not carry over (Claire Vo: “Duet does not ingest all your local skills”).
  • One thread, plus channels. You message it by typing or by voice “in one ongoing thread, so a quick reply and a weeklong project both start in the same place.” You can also start separate channels for different topics, and “they share the same memory” (help center).
  • Morning brief. Each morning it “looks across your email and calendar, picks what most needs your attention, and can start the first steps, like drafting replies for you to review.” Through the day it sends updates and “follows up on threads that have gone quiet.”
  • Background checks. It checks connected accounts “every few hours”, because “checking nonstop would use up your plan’s limits much faster.” You can tell it what you want to hear about and how often. Something urgent can arrive between checks.
  • Tasks tab. Shows what it is working on and highlights tasks that need your input; each task has its own page and you can steer it mid-task. Duet may suggest a task looks finished, “but only you can mark a task as completed.”
  • Make tab. Bigger results come back as files you can review and edit: a research brief, financial model, slide deck or dashboard.
  • Plugins. Supports plugins, “such as the small business plugin” (see Claude for Small Business, now 44 skills), and can suggest ones that fit your work.
  • Its own computer and browser. Duet works on a cloud virtual machine with its own browser, so your phone or laptop does not need to stay on. For sites with no connector it uses that browser. You can’t watch or take over the browser yet.
  • Sources on numbers. “When it gives you a number or a summary, it tells you which email, file, or spreadsheet it came from” (help center).
  • Past-chat search. It searches your Duet chats and your Claude chats outside projects, controlled by the same Search and reference chats setting in Settings > Memory.
  • Feedback. Tell Duet in plain words what went wrong and it files a report to Anthropic.

Integrations

  • Can use the “over 6,000 connectors and plugins in the Claude Marketplace”. The help center names Gmail, Google Calendar, Google Drive, Slack, QuickBooks, HubSpot, Docusign and Calendly.
  • You connect apps in Claude, not in Duet; switching an app off in one place switches it off in both. Slack “works with limits”; Microsoft 365 “isn’t available yet”. See Limits and Limitations.
  • Launch-post examples: find missed Gmail and Slack messages and help reply; check QuickBooks or Stripe for unpaid invoices and draft reminders; prepare and run Gusto payroll; draft a LinkedIn job post and post it once you approve; prepare agreements and send them through Docusign; draft follow-ups from HubSpot deal history; turn a launch into a Canva social campaign; set up and manage a Shopify store (a richer storefront editor is “coming soon”).
  • Payments: Link (Stripe’s wallet) and Shop Pay (Shopify’s), connected inside Duet, each purchase on a one-time card. 1Password sign-in is “coming soon”. Update 2026-11-01: 1Password has announced “1Password for agents”, reported on The Artificial Intelligence Show Ep. 248 with integrations rolling out in Duet (garbled caption; see Safety, Privacy, and Security). See Payments with Link and Shop Pay.

Safety approach (summary)

Full detail: Safety, Privacy, and Security.

  • Approvals. Before it sends a message, makes a payment, signs a document or deletes anything, it shows the action, destination and content, including every recipient, and waits. If you can’t approve right away, it finishes the rest of the task and brings the approval back later.
  • Separate intent check. A check built on the auto mode classifier runs outside the working model and judges whether each action matches your intent.
  • Off-limits even with approval: trading securities, moving money between accounts, accessing someone else’s account.
  • Sensitive-information classifier with two tiers: “never share” (credentials, full card and bank numbers, government IDs) and “needs consent” (health, finances).
  • Credentials never reach the model. Passwords go through an encrypted in-app form; sign-in tokens follow the Claude Managed Agents proxy design.
  • Tested by Anthropic’s Safeguards team, red teams and an external penetration-testing firm; in scope of the HackerOne bug bounty.

Pricing and limits

  • Included in Pro and Max at no extra cost during the research preview.
  • Uses your plan’s five-hour session and weekly limits “at the same rate as chats in Claude”; background work such as the morning brief and check-ins also counts. Larger tasks, like a financial model or a deck, take more model turns.
  • Detail and the unconfirmed “one free limit reset” claim: Limits and Limitations.

Current limitations (launch post)

  • May misread a thread or misjudge urgency.
  • Personal errands (booking tickets, etc.) are weaker than work tasks.
  • Cannot yet show you its browser.
  • Lives in its own app, separate from the main Claude app; early-user feedback will shape how the two work together.

The product-page demo

The product page shows a fictional café chain, “Marlo”, and owner “Maya”. It is a scripted demo, not a customer result. What it shows Duet doing:

  • A five-item morning brief: a finalist’s competing offer expiring at 6 p.m., a redlined 640 health-permit fee needing approval, a Series A deck with two slides to review, and 27 wrong delivery orders with refunds drafted.
  • An approval card for the $640 permit fee that expires in 8 minutes.
  • Pulling missing data-room documents, reading investor bios, and drafting meeting-prep answers.
  • An email approval card (“Send from you?”) for sending the deck.
  • “Marlo Pulse”, a one-page business-health dashboard built from point of sale, Shopify, CRM, support inbox, payroll and bank data, refreshing at 5:30 every morning before the brief, followed by a churn finding and a proposed two-week free-shipping test.

The pattern worth copying: ask for a standing dashboard that refreshes before your brief, then let Duet point at what changed.

Early testers named in the launch post

  • Inside Anthropic. Boris Cherny, who leads the Claude Code team, talks through his mental to-do list with Duet on walks. Chelsea Larsson, who leads model and product behavior design, hands it her to-do list before leaving to be with her kids and gets an evening report. Karan Ahuja, who leads the Duet team, uses it to schedule user interviews, prototype and ship code.
  • Jess Provencher (broker and owner, Pro Homes). Unasked, the morning rundown flagged a property showing on her calendar for the wrong day.
  • Amin Ebrahimi (owner, Shahs of Kabob, about 110 employees in South Florida). Uses it for accounting, customer support and marketing, and has worked through invoices store by store: “I don’t need to open up 100 tabs.”
  • Mya Dalton (brand strategist). Duet prepared a brief for a call with a potential client, noticed the call fell just before her flight and offered to move it.

Launch video (2026-10-21)

Source: raw/Introducing_Claude_Duet.md — Anthropic, “Introducing Claude Duet”, youtube.com/watch?v=Fb-68zqcSVs, 1:39, uploaded 2026-10-21 (launch day). The captions hold about 125 words of sound bites from speakers the transcript does not name.

  • One speaker credits Opus 5.5 with “the discernment and judgment around when to be proactive, how to act on your behalf”. This matches the launch post’s claim above.
  • Another values its context awareness: “understanding that this is the type of stuff I need versus what I don’t need”, and that “it just brings what I need to know from Slack to me proactively”. The help center says Slack “works with limits” (see Limits and Limitations); the video does not show what those limits are.
  • The video closes on phone use: “This is just like really the way that I use Claude on my phone these days.”

Key Takeaways

  • Claude Duet is Anthropic’s first proactive personal agent: it acts on a schedule and messages you first, rather than waiting for a prompt.
  • It runs on Opus 5.5 (no model choice) on its own cloud computer, and inherits your Claude memory, skills and connectors.
  • Every consequential action (send, pay, sign, delete) waits for your approval, and a separate intent check runs outside the model.
  • It is included in Pro and Max at no extra cost, but its background checks draw on the same limits as your chats and Claude Code.
  • It is US-only, waitlisted, and not on Team or Enterprise; Microsoft 365 is not supported yet.
  • Duet differs from Claude Cowork, which runs task workflows when you start them; Duet watches in the background and comes to you.

Open Questions

  • Proactivity in practice. Anthropic designed Duet to stay quiet until something needs attention; Claire Vo found it too quiet to re-engage her. Whether notification settings change that is not established. See Early Reviews.
  • Desktop app and team sharing. Neither exists in the research preview, and Anthropic has not given a timeline. Team and Enterprise plans are promised “later”.
  • Adding connectors inside Duet. The help center says you manage connections in Claude settings, which explains why Brock Mesarich found no way to add one in the app. Whether in-app connection will come is not stated.
  • Safety detail. Resolved 2026-10-28: the follow-up post “Claude Duet Safety, Privacy, and Security” (2026-10-23) is saved and summarised in Safety, Privacy, and Security.