Claude Duet: Anthropic's proactive agent app

What changed?

On October 21, Anthropic opened a research preview of Claude Duet (Anthropic). It is not a regular chat: Duet comes to you when something needs attention. Each morning it sends a short brief on the most important items across your connected apps, and during the day it only messages you when it matters.

It works on its own virtual machine with its own browser, so your laptop or phone does not need to stay on. From a short message it works out which apps to use and which steps to take, then notifies you when the task is done or needs review.

How it works

• It runs on Claude Opus 5.5 with no model picker, and decides how much effort each job needs.
• It carries over your Claude memory, skills and connected apps, so setup is minimal.
• It checks your accounts every few hours rather than nonstop, because constant checks would burn through plan limits.
• It asks before sending messages, making payments, signing documents or deleting anything. Trading securities and moving money between accounts are blocked even with approval.
• It never sees passwords or sign-in tokens directly, and a dedicated classifier blocks sharing card numbers or government IDs.

Who can use it

For now, only US Pro and Max subscribers via a waitlist (Claude Help Center). It runs as a separate app on iOS, Android and claude.ai. There is no extra charge, but Duet draws on your existing plan limits, background checks included. Team and Enterprise plans come later, there is no Microsoft 365 connector yet and Slack works with limits.

Why it matters

The agent race is moving from the chat window to always-on assistants, with OpenAI and Meta shipping similar products. Duet stands out by putting approval gates and leak protection at the center of the product. It is not available outside the US yet, but if you build agents, its approval model and throttled background checks are a useful pattern. Anthropic itself warns that Duet can still misread a thread or misjudge what is urgent.