Anthropic opens Claude Duet preview for US Pro and Max users

Anthropic has opened a research preview of Claude Duet, a proactive agent that finds work that needs attention, carries out assigned tasks in the background, and asks for permission before taking action for the user. People in the US who subscribe to Claude Pro or Max can join the waitlist to get access. Duet runs in a separate app for iOS and Android and on the Claude website. During the preview, it remains separate from the main Claude app. It is included in Pro and Max plans and uses the same usage limits as those plans.

Duet is built around Claude Opus 5.5 and is designed for professionals and small business owners who have a lot of work to manage. Every morning, it sends a short report based on information from connected apps. It checks those accounts every few hours and sends a message when it decides something needs attention. Tasks can include scheduling meetings, building dashboards, making financial or sales forecasts, creating prototypes, and building web apps. Users can manage quick requests and projects that last a week in one ongoing text or voice conversation. A virtual machine with its own browser continues working even when the user's devices are turned off.

Duet brings over the memory, skills, and connected apps from existing Claude accounts. It can also use more than 6,000 connectors and plugins available through the Claude Marketplace. Examples of supported services include Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, DocuSign, HubSpot, Canva, and Shopify. Duet can make business purchases using Stripe's Link or Shopify's Shop Pay. These purchases include approval steps, and Link can use virtual cards. Support for signing in with 1Password and more advanced visual editing for Shopify is planned.

Anthropic says human approval is a key safety rule for Duet. The agent can read information, write drafts, and prepare work on its own. But before it sends a message, makes a payment, signs a document, or deletes anything, it must show the user what it plans to do and wait for approval. Trading securities and moving money between accounts are not allowed. Duet's system for handling login details follows the design used in Claude Managed Agents. Its controls for taking actions use the same classifier as Claude and Claude Code. A separate classifier and safety prompt prevent it from sharing credit card numbers and government ID numbers. For other sensitive information, such as health or personal finance details, users must give clear instructions. Anthropic says its red teams and an outside security company tested the system. Duet is also covered by Anthropic's bug bounty program.

The preview still has some limits. Duet may misunderstand conversations or fail to rank tasks by urgency correctly. It is less capable at handling personal errands, and users cannot yet see its browser. Background checks count toward each plan's usage limits, and larger projects need more interactions with the model. Anthropic says feedback from early users will help it decide how Duet and the main Claude app should work together.

Earlier, we also reported that Anthropic was working on its own proactive agent. Now, that agent has arrived.

Sources and related context

• Claude Duet product page and waitlist: The product page confirms that the preview is for Pro and Max users in the US. It also confirms the use of Opus 5.5, morning reports, and the transfer of memory, skills, and connected apps from existing accounts.
• Scaling Managed Agents: Decoupling the brain from the hands: This article explains how Anthropic separates login credentials from other parts of its Managed Agents system. The same design is used to handle Duet's sign-in details.