Claude Duet runs on its own virtual machine so it keeps working when your laptop is off, and Anthropic has opened a waitlist for US Pro and Max subscribers.

On 21 October 2026, Anthropic announced Claude Duet, a new app that can carry out tasks for users. It runs on a virtual computer in the cloud, with its own web browser. This means users do not need to keep their laptop or phone turned on for it to work.

The company has opened a research preview for people in the US who subscribe to Claude Pro or Max. They can join a waitlist to get access.

Anthropic said Claude Duet is built around Claude Opus 5.5. Every morning, it sends users a short report about things happening across the apps they have connected.

Users describe a task in a short message. The agent then decides which apps to use and what steps to take, working in the background. It sends a notification when the task is finished or when it needs the user to check something.

Its own computer

According to the product page, approvals, sign-ins and safety checks happen on the user's own computer in the cloud. Anthropic says this helps keep private information safe. The agent handles login details in the same way as Claude Managed Agents. According to the company, it never directly sees passwords or sign-in tokens.

Claude Duet checks connected accounts every few hours. Anthropic said checking them all the time would use up a user's plan limits much faster.

Anthropic is also introducing a new system that checks messages for safety. It uses a classifier and a special safety prompt. The company says this system is designed to refuse requests to share the most sensitive types of information, such as credit card numbers and government ID details.

Other private information, including health details and personal financial information, can be sent only if the user has clearly told Claude Duet to send it, Anthropic said.

Before sending a message, making a payment, signing a document or deleting anything, the agent shows the user what it plans to do and waits for permission. Anthropic said the agent cannot trade securities or move money between accounts, even if the user approves. Users can disconnect any app that the agent can access.

Anthropic said security teams that test systems by trying to break them, known as red teams, and an outside company have tested Claude Duet for security weaknesses. The app is also covered by Anthropic's public bug bounty program, which rewards people for finding and reporting bugs.

The product page also warns that Claude Duet has limits. It may misunderstand a conversation or get the level of urgency wrong. It might also miss something that arrives between its checks with the user. Anthropic said the agent cannot yet show users its browser, and it is less capable at personal tasks, such as booking tickets, than it is at work tasks.

Who can use it?

The research preview is available through a waitlist to Claude Pro and Max subscribers in the US. Claude Duet is a separate app for iOS and Android. On a computer, users can open it on the web at claude.ai. It is separate from the main Claude app.

During setup, Claude Duet brings over the user's saved memory, skills and connected apps from their Claude account.

During the preview, Claude Duet uses the same usage limits as the rest of Claude. Its background checks also count toward those limits. Anthropic said larger tasks, such as making a financial model or a presentation deck, need more model turns and more work in the background.

Team and Enterprise plans are not included yet.

Sources:

• Anthropic, Introducing Claude Duet: https://claude.com/resources/articles/claude-duet
• Anthropic, Claude Duet product page: https://claude.com/product/claude-duet