Claude Duet Checks Your Apps Every Few Hours and Spends Your Plan Limits Doing It

Anthropic opened a research preview of Claude Duet on October 21, a standalone agent app that watches a user's connected accounts, sends a morning brief and messages them when something needs attention. Access is by waitlist, starting with US subscribers on the Pro and Max plans, and the model underneath is Opus 5.5.

The product page says the plan is the price: Duet is "Available in Pro and Max plans in the US" and shares usage limits with the rest of Claude. Team and Enterprise plans are not included.

The cost is in the plan limits

Anthropic's help article says Duet checks connected accounts about every few hours, an interval it chose to avoid using up plan limits too fast. Those background checks count against the limits anyway, and larger jobs such as financial models or decks need more model turns.

Neither page gives a number of checks a day or a share of a Pro allowance. A subscriber who connects Gmail, Slack and Stripe cannot tell from the documentation how much chat capacity the agent will take before they ask it for anything.

What it will and will not do without asking

The approval rules are specific. Anthropic says Duet "shows you exactly what it's about to do and waits for your approval" before sending messages, making payments, signing documents or deleting anything. Trading securities and moving money between accounts are off-limits even with approval.

Action	Rule in Anthropic's help article
Send a message, make a payment	Approval first
Sign a document, delete anything	Approval first
Trade securities, move money between accounts	Not allowed
Buy something	Stripe's Link virtual card, or Shop Pay
Sign in with 1Password logins	Announced as coming

Duet runs on a virtual machine with its own browser, so a laptop or phone need not stay on. It reaches more than 6,000 connectors and plugins in the Claude Marketplace. Anthropic says credentials are handled so the agent never directly sees passwords or sign-in tokens, that a new classifier and system prompt check each message for sensitive information, and that red teams, an external firm and a public bug bounty cover it. All of that is Anthropic's description. MacRumors, which reported on October 21, relayed the claims without testing them.

Where it is thin

Anthropic lists its own limits: Duet can misread a thread or misjudge urgency, it can miss something that arrives between check-ins, personal tasks such as booking tickets are less developed than work tasks, it cannot yet show the user its browser, and it lives in a separate app from the main Claude app. The gap between check-ins is the practical one. A message that lands an hour after a check waits for the next one.

The availability story also varies by source. Anthropic's pages say the app is on iOS and Android and on the web at claude.ai, while MacRumors describes a standalone iOS app and the web.

MacRumors says Anthropic positions Duet against Meta's Muse agent and OpenAI's Dots agent. The Terminal has covered the agent tooling around it, including Stack Overflow's plugin for agents and Claude's work on OpenSSH.

The preview has no end date. Anthropic's page gives no launch date beyond the October 21 announcement and no date for opening the waitlist to Team and Enterprise customers.