Anthropic opened a research preview of Claude Duet on Oct. 21, introducing a standalone AI agent for professionals and small-business owners. The service is available in the United States through a waitlist to subscribers on the company’s Pro and Max plans, with access being issued in waves. It runs as a separate app on iOS and Android and is also available through claude.
Claude Duet is designed to work proactively rather than wait for every instruction. Each morning, it reviews connected applications and sends a brief highlighting items it considers important, then checks those accounts at intervals during the day. Users can assign it administrative work such as scheduling meetings, drafting replies, preparing research briefs, building dashboards, creating financial or sales forecasts, and developing web applications.
The agent carries over a user’s Claude memory, skills and connected applications. Available integrations include services such as Gmail, Google Calendar, Google Drive, Slack, QuickBooks, HubSpot, DocuSign and Calendly, while the company says the broader Claude Marketplace offers more than 6,000 connectors and plugins. Duet can continue working after a user closes the app because it operates through a cloud-based computer with its own browser. Background checks and larger assignments count against the usage limits of the user’s existing plan.
Anthropic says Duet must ask for approval before sending messages, making payments, signing documents or deleting information. Some actions remain prohibited even with approval, including trading securities, moving money between accounts and accessing another person’s account. Purchases approved by users can be completed through Stripe’s Link wallet or Shopify’s Shop Pay using a one-time card for the transaction.
The preview also comes with acknowledged limitations. Duet may misread a conversation or misjudge what is urgent, and the company says it is less effective at personal errands such as booking tickets. Users cannot watch its browser activity in real time, and the service does not yet support Team or Enterprise plans. Anthropic says it tested the system with red teams and an external penetration-testing firm, added safeguards against prompt injection and sensitive-data leaks, and placed the product under its public bug-bounty program. The company says feedback from early users will guide changes to the agent and how it works with the main Claude service.