Duet runs on iOS, Android and the web and is built on Claude Opus 5.5. Pro and Max subscribers in the US can join a waitlist, while Team and Enterprise plans are not included yet. Instead of waiting for a chat, the agent sends a morning summary of key items from connected apps and reaches out only when something needs attention.
The agent runs on its own cloud computer with a browser and connects to Gmail, Slack, Stripe, Docusign and Shopify through more than 6,000 connectors and plugins from the Claude Marketplace. Purchases go through Stripe Link with a virtual card, so the real card number is not shared. Anthropic says Duet can only read, draft and prepare on its own, and it asks for approval before sending messages, making payments, signing documents or deleting anything. Trading securities and moving money between accounts are blocked, even with approval.
Anthropic says the agent never sees passwords or login tokens directly. It has also added a classifier and special safety instructions to screen messages for sensitive data, and it checks connected accounts only every few hours to protect usage limits. The company concedes that the early version can misread threads, misjudge urgency and miss events between checks, and that it is weaker at personal errands such as booking tickets.
Why it matters if you build agents
Developers building agents can look at a consumer design where the agent starts work on its own and approval gates sit before messages, payments and deletions. The product also shows Anthropic's patterns for credentials and purchases, including keeping login tokens away from the agent and using a virtual card for payments.
Anyone relying on scheduled or background checks should note the few-hour polling interval and that background checks count against shared Claude usage limits.
Get stories like this every weekday
One short email with the day's agent tooling news.