Anthropic opened a research preview of Claude Duet on Wednesday, October 21, a standalone app in which Claude works as a proactive agent rather than waiting to be asked. Duet sends users a morning brief, keeps watch over their connected apps during the day, completes tasks in the background on its own cloud computer, and checks with the user before it sends a message, makes a payment, signs a document or deletes anything.
The preview is waitlisted and starts with Pro and Max subscribers in the United States, Anthropic said. Duet is available as a standalone app on iOS and Android and on claude.ai, is included in those plans at no extra charge, and shares the same usage limits as the rest of Claude — with the caveat that its background checks count toward those limits.
How Claude Duet Works: A Brief In The Morning, Checks Every Few Hours
Duet is built around Claude Opus 5.5 and designed, in Anthropic’s words, to be resourceful and to decide what needs the user, what can wait, and when to check first. Each morning it sends a short brief on the most important items across the user’s connected apps. Through the day it checks those accounts every few hours — Anthropic says checking nonstop would burn through plan limits much faster — and messages the user only when something needs attention. Users can also set what they want to hear about, and how often.
Tasks arrive in one ongoing thread, by typing or voice. They range from scheduling meetings to building dashboards with dozens of tools, creating financial or sales forecasts, prototyping and building web apps. Duet carries over the user’s memory, skills and connected apps from their Claude account, and it runs on a virtual machine with its own browser, so neither the user’s laptop nor phone needs to stay on while it works. It notifies the user when a task is complete or needs review.
The app can draw on the more than 6,000 connectors and plugins in the Claude Marketplace. Anthropic’s examples include surfacing missed Gmail and Slack messages, checking QuickBooks or Stripe for unpaid invoices and drafting reminders, preparing payroll in Gusto, drafting a job posting for LinkedIn, sending agreements through Docusign, and managing a Shopify store. It can also buy things for the business using Link, Stripe’s digital wallet — Duet confirms the item, Link asks the user to approve the spend, and payment goes through a virtual card — or check out with Shop Pay.
The Guardrails: Approval First, And A Classifier For Sensitive Data
Approval is the product’s central design rule. Duet reads, drafts and prepares work on its own, but before any consequential step it shows the user exactly what it is about to do and waits. A few actions are off-limits even with approval, including trading securities and moving money between accounts, and any connected app can be disconnected at any time.
Anthropic says Duet’s credential handling follows the design used for Claude Managed Agents, so it never directly sees passwords or sign-in tokens, and its actions are screened by the same auto-mode classifier that guards Claude Code. The preview also ships with a new dedicated classifier and safety system prompt that checks each message for sensitive information: it is designed to reject requests to share the strictest tier of data, such as card numbers and government identification, and to pass other sensitive material, such as health or personal finance details, only when the user has explicitly instructed it. Anthropic says red teams and an external firm penetration-tested Duet, and that it is covered by the company’s public bug bounty programme.
Anthropic is candid about the limits. Duet is an early version that can still misread a thread or miscalibrate what counts as urgent; it is built around work, so personal errands such as booking tickets are weaker; and it cannot yet show the user its browser.
Why It Matters
Claude Duet matters because it moves Claude from a tool users open to a service that comes to them — the same shift Meta is making with Muse and OpenAI with its Dots agent, and the defining contest among the agent platforms to know in 2026. The design choices are the differentiators: work happens on Anthropic’s virtual machine rather than the user’s device, every consequential action waits for approval, and the meter is the user’s existing plan limits, which background work will consume faster than chat. For the professionals and small business owners Anthropic built it for, the test is whether the morning brief and the few-hours checks save more attention than the approvals demand. How Duet’s judgment holds up in the preview will shape how it merges with the main Claude app, where Anthropic’s coding agent, covered in our explainer on Claude Code, already works autonomously inside tighter boundaries.
Background
Duet is Anthropic’s consumer-facing step into proactive agents, following a year in which Claude gained connectors, memory and agentic coding tools while rivals shipped always-on assistants of their own — Meta’s Muse agent reached 10 million US downloads in its first 40 days this autumn. Anthropic has framed its agent work around a framework for trustworthy agents that also guides Claude Code and Claude in Chrome; Duet is the first product to put that framework in front of non-developers, inside its own app, separate for now from the main Claude app while the company learns from early users.
0 Comments