Anthropic on October 21, 2026 opened a research preview of Claude Duet, a separate app in which Claude acts as a proactive agent: rather than waiting for a prompt, it surfaces what it judges needs a user’s attention, takes on assigned tasks, and keeps working after the user closes the app.
Access is limited for now. Anthropic says the preview is available by waitlist to U.S. subscribers on its Pro and Max plans, with access opening in waves. Its help center says Team and Enterprise plans are not included yet, though the company plans to add them later. Duet is offered as a standalone iOS and Android app and on claude.ai, and Anthropic has not announced a separate charge for it during the preview.
Anthropic says Duet is built around Claude Opus 5.5, and users cannot choose a different model. It inherits the memory, skills, and connected apps from a person’s existing Claude account, and users talk to it by text or voice in one continuing thread. According to the company, it runs on its own virtual machine with a browser, so tasks continue even when a user’s laptop or phone is off.
The pitch centers on judgment about when to interrupt. Anthropic says Duet sends a short morning brief drawn from connected apps and then checks those accounts every few hours, not continuously, because constant monitoring would burn through plan usage limits faster. Those background checks count against the same limits as the rest of a subscriber’s Claude use, as do larger jobs like building a financial model or a slide deck.
Anthropic is aiming the product at professionals and small business owners. The company lists examples drawn from its marketplace of more than 6,000 connectors and plugins, including finding unpaid invoices in QuickBooks or Stripe and drafting reminders, preparing Gusto payroll for review, readying Docusign agreements, and managing a Shopify store. It can also make purchases through Stripe’s Link wallet using a virtual card, Anthropic says, and support for signing in with 1Password-saved logins is listed as coming soon.
On safety, Anthropic says Duet will show users exactly what it intends to do and wait for approval before it sends a message, makes a payment, signs a document, or deletes anything. Some actions, such as trading securities or moving money between accounts, are blocked even with approval. The company also describes a new classifier meant to block sharing of its most sensitive information tier, such as card numbers and government IDs, along with prompt-injection defenses and outside penetration testing. Those are the company’s descriptions of its safeguards, not independent findings.
Anthropic is candid about the preview’s limits: it says Duet may misread a thread or misjudge what is urgent, is weaker at personal errands such as booking tickets, and cannot yet show users its browser. CNBC and VentureBeat both frame the launch as Anthropic’s answer to Meta’s Muse and OpenAI’s Dots, personal agents that have drawn early user traction.
DigiEditorial treats Anthropic’s October 21 announcement, product page, and help-center article as the primary sources, and corroborated the launch date, plan eligibility, Opus 5.5 basis, and approval model against reporting from CNBC and VentureBeat.