Anthropic’s proactive personal agent app, in research preview since October 21, 2026. Duet runs on Claude Opus 5.5 on its own cloud computer, sends a morning brief, checks your connected apps every few hours, and asks before it sends, pays, signs or deletes anything. Waitlist only: US Pro and Max subscribers, iOS, Android and claude.ai.
Section created 2026-10-28. The product overview moved here from claude-ai/.
Start here
- Claude Duet — Product overview: how it works, integrations, safety summary, pricing, the product-page demo, and the launch-post testers.
- Getting Started with Claude Duet — Accepting the invite, the app layout, setting up connectors and skills in Claude first, notifications, and a seven-day trial plan.
Reference
- Limits and Limitations — Access rules, how it spends plan limits, the help center’s limitations list, the Microsoft 365 gap, and gaps reported by reviewers.
- Safety, Privacy, and Security — Approvals and the out-of-model intent check, the never-share and needs-consent tiers, memory after disconnecting, credential and browser isolation, and testing.
- Payments with Link and Shop Pay — Connecting a wallet in Duet, the approval flow, one-time cards, limits, and who to contact when a purchase goes wrong.
Field reports
- Early Reviews — Dan Shipper (Every), Peter Yang, Claire Vo, Brock Mesarich and the Hacker News launch thread, with where they agree.
Elsewhere in the wiki
- Personal Agents Compared (October 2026) — Duet against Muse, Dots, Grok Bot and others.
- Two Shapes of the Personal Agent — the synthesis Duet fits into.
- Claude Cowork and Claude Tag — Anthropic’s on-demand and team agents.
Sources
First-party: launch post (2026-10-21), safety post (2026-10-23), product page, and three help-center articles, all saved in ai-research/ on 2026-10-27 and 2026-10-28.