Anthropic's Claude Duet is an agent that comes to you first: it watches your apps, works while you're away, and asks before it spends or sends

Anthropic on Wednesday, Oct. 21, 2026, opened a research preview of Claude Duet, a standalone app for iOS, Android and claude.ai where Claude acts as a proactive agent that monitors connected apps, sends a morning brief, runs tasks in the background on its own cloud computer, and asks for approval before it sends, pays, signs or deletes anything.

Every big lab now agrees the next AI product isn't a chatbot you visit; it's an agent that lives in your inbox and taps you on the shoulder. Anthropic's pitch is that it's the careful one: approvals before money moves, no raw passwords, a classifier guarding your ID numbers. That's the right list. But an agent that reads your email and can buy things on a virtual card is a very large target, and "checks every few hours" is a polite way of admitting this stuff still burns through usage. The waitlist and the "research preview" label are doing a lot of work. Watch whether it earns trust with small businesses before Muse earns scale with everyone else.

Claude Duet is a new app in which Claude works as a proactive agent, software that starts work instead of waiting for a person to open a chat. Each morning it sends a short brief on important items across connected apps. It checks those accounts every few hours, not nonstop, which Anthropic says is meant to save plan usage, and it messages you only when something needs attention. You hand it tasks by typed or voice message, in one ongoing thread. Examples Anthropic gives: scheduling meetings, building dashboards, financial or sales forecasts, and prototyping and building web apps. Those lines are Anthropic's.

It runs on Claude Opus 5.5, and users can't pick a different model. It has its own virtual machine, a computer in Anthropic's cloud, with a browser, so it keeps working when a laptop or phone is off. It carries over memory, skills, and connected apps from the person's Claude account. It can use the 6,000-plus connectors and plugins in the Claude Marketplace. Anthropic's examples include Gmail, Slack, QuickBooks, Stripe, Gusto, Docusign, HubSpot, Canva, and Shopify. Those lines are Anthropic's.

Before it sends a message, makes a payment, signs a document, or deletes anything, it shows exactly what it is about to do and waits for approval. Some actions stay off-limits even if the person approves, including trading securities and moving money between accounts. It can buy things using Stripe's Link wallet, paying with a virtual card, and with Shop Pay. Sign-in through 1Password is "coming soon," Anthropic says. Those lines are Anthropic's.

On safety, Anthropic says Duet never directly sees passwords or sign-in tokens. A token, here, is the key an app uses to stay signed in. Duet uses Anthropic's auto-mode classifier, a checker that sorts an action into allowed or not, to approve actions. It uses prompt-injection defenses from Anthropic's browser-use research. Prompt injection is a hidden instruction in a web page or an email, written so the agent follows the page instead of the person. A new classifier is meant to refuse sharing the most sensitive data, such as credit card numbers and government IDs. Health or financial details are sent only when the person explicitly says to. Red teams and an outside firm penetration-tested it, meaning they were hired to try to break it, and it is in Anthropic's bug bounty, which pays outsiders who report flaws. Those lines are Anthropic's.

The release is a research preview, a limited early version reached by a waitlist, not a full public launch. It is available in the United States only, for subscribers to Claude Pro and Max, the paid individual plans. Duet is included in those plans. There is no separate Duet price. It draws on the same usage limits, and the background checks count against that limit. It is not on Team or Enterprise plans yet. Microsoft 365 is not available yet. It cannot make phone calls or text other people. You cannot watch its browser or take it over. Those lines are Anthropic's.

Anthropic's own limits, in its words: it "still might misread a thread or miscalibrate what's urgent," and personal errands such as booking tickets "aren't quite as strong."

It arrives the same day Meta's Mark Zuckerberg said, on the earnings call, that 10 million people have set up Muse, Meta's always-on personal agent, The Verge reported. OpenAI's Dots is the other widely discussed product in that category. Duet is Anthropic's version: an agent that checks in, rather than a chat window a person has to go open.

In plain terms, Anthropic on Wednesday opened a United States waitlist for Claude Duet, included with Pro and Max. It checks connected apps every few hours, keeps working on its own cloud computer while your devices are off, and waits for a yes before it sends, pays, signs, or deletes. It will not trade securities or move money between accounts, even if asked. Anthropic says it can still misjudge what is urgent, and that booking tickets is a weaker skill.