Claude Duet Explained: What It Is, Who Gets It, and What It Can Do Without Asking You

Anthropic has shipped the thing every AI company has been promising: an assistant that contacts you first. Claude Duet opened as a research preview on October 21, 2026, and it is a separate app from the Claude you may already use. It reads your connected accounts every few hours, sends you a brief each morning, works on tasks while your phone is in your pocket, and stops to ask when it is about to do something it cannot undo. The practical questions are the usual ones, and the answers are narrow: you need a paid plan, you need to be in the US, and you need to be let off a waitlist.

Key Takeaways

• It is waitlist-only and US-only. Anthropic's own words: "We're starting today with Pro and Max users in the US." Team and Enterprise plans are "not included yet", and the free plan is not named anywhere.
• It costs nothing extra. Claude Duet is included in Pro and Max during the research preview. Pro is $17 a month on the annual plan ($200 up front) or $20 monthly; Max starts at $100 a month.
• It spends your existing usage limits, and faster than chatting does. Anthropic says Duet "shares your plan's usage limits with the rest of Claude", its background checks count toward them, and big jobs like building a deck "require more model turns".
• It is a separate app, not a mode. Claude Duet is a standalone app for iOS and Android, plus claude.ai on a computer. Anthropic says it "lives in its own app, separate from the main Claude app" for now.
• It runs on its own cloud computer. It has "a virtual machine with its own browser", so it keeps working after you close the app. It checks your connected accounts "every few hours", not constantly, specifically to avoid burning your limits.
• Four things are blocked even with your approval: trading securities, banking actions such as moving money between accounts, accessing someone else's account, and one more category it will always re-ask about rather than remember (paying, deleting an order, creating an account).
• It never sees your passwords. Anthropic says credentials pass through separate machines the model cannot read, and Duet "can't save them to its memory."
• Anthropic says plainly that it is not finished. It "might misread a thread or miscalibrate what's urgent", personal errands are weaker than work tasks, and it cannot yet show you its browser.
• It is powered by Claude Opus 5.5, the model Anthropic launched on September 22, 2026.

What Claude Duet Actually Is

Claude Duet is a Claude that starts the conversation. Anthropic describes it as "a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working even when you step away, and checks with you before it acts on your behalf." The difference from the Claude app is the direction of traffic: the chat app waits for you to open it, and Duet messages you.

Three things happen without you doing anything. Each morning it sends "a short brief on the most important items across your connected apps." Through the day it watches those apps and messages you "only when something needs your attention." And when you hand it a job, it does the work on a cloud computer of its own and notifies you when it is done or stuck.

Anthropic is explicit about who it built this for, and it is not everybody: "We built Claude Duet for professionals and small business owners with more on their plates than they can track."

Can You Get It, and What Does It Cost?

Not unless you already pay, and not unless you are in the US. Here is the full gate, from Anthropic's announcement and the Claude Duet product page.

And the prices behind those plan names, from claude.com/pricing read on October 24, 2026:

One thing that is not on the pricing page: Claude Duet. On October 24, 2026 the word "Duet" appears on claude.com/pricing three times, and all three are in the site navigation and footer. It is in no plan card and in no row of the plan comparison table, which has rows for Claude Code, Claude Science, Claude in Chrome and Claude Dashboards. The Claude help centre has no Claude Duet section either: its collections list runs from "Claude Cowork" (14 articles) to "Claude for Education", and Duet is not among them. For a research preview behind a waitlist that is the expected state, and it is also the reason the announcement and the product page are the only documents to go by right now.

The Usage-Limit Catch, Which Is the Real Price

Claude Duet does not cost money, it costs your limits, and it burns them in a way chatting does not. This is the detail most likely to bite someone who upgrades for it, and Anthropic states it twice.

From the announcement: "During the research preview, Claude Duet shares your plan's usage limits with the rest of Claude, and it uses them differently than a typical conversation does. Its background checks count toward those limits, and larger tasks, like building a financial model or a deck, require more model turns and background work."

Read that as three separate costs. The morning brief and the periodic checks of your inbox and calendar consume limits whether or not you asked for anything that day. A single large task consumes far more than a question would. And all of it comes out of the same allowance as your normal Claude chats, so a busy Duet means less Claude everywhere else.

There is one piece of good news in the design. Anthropic says Duet "checks your connected accounts every few hours, because checking nonstop would use up your plan's limits much faster." So it is deliberately not real-time, which is also why Anthropic warns it "can miss something that arrives between checking in with you."

What It Will Do, What It Asks First, and What It Refuses

This is the part worth reading before you join the waitlist, because an agent that can send email in your name is a different proposition from a chatbot. Anthropic published a separate safety, privacy and security article on October 23, 2026, and between the two posts the boundaries are specific.

Two details in there are better than the industry norm. The approval check "runs outside the model doing the work so instructions hidden in a page cannot affect the outcome" -- which is the defence against a web page telling your agent to do something you did not ask for. And purchases go through a one-time card: Anthropic says Duet pays with a card from a wallet like Link from Stripe "limited to that merchant and amount", so "your real card number isn't shared with Claude Duet or the merchant." It can also check out with Shop Pay. Signing in with 1Password is listed as coming, not shipped.

Anthropic also names the problem with asking permission a lot, which most vendors do not: "Approval fatigue is a serious issue, because too many requests teach people to tap through them, sometimes without adequate consideration."

What It Does With What It Reads

An agent in your inbox is deciding all day where your information should go, and Anthropic built a separate classifier for exactly that. It sorts your information into two tiers, and the wording is worth having in front of you.

Never share: "sign-in credentials, full card and bank account numbers, and government ID numbers." Anthropic says Duet "is instructed never to share these with anyone, even if you ask."

Needs consent to share: "other private information, such as health or finances", which it will send "only if you've explicitly approved it." Your own request counts as consent, so Anthropic's example is that asking it to email your dentist about tooth pain does not need a second approval.

The classifier runs separately from the model that does the work and checks each message before it goes out. Anthropic is candid that it is new: the safeguards "will occasionally miss something or flag a harmless message", which is why the approval screens sit behind them.

Three practical facts about its memory, from the same article:

• Disconnecting an app does not unlearn it. "Disconnecting an app stops new access from that moment, but what Claude Duet already learned from it stays until you reset its memory", which you can do in the app's settings.
• Switching an app off switches it off in both places. The apps you connected to Claude carry over to Duet, and turning one off in either product turns it off in the other.
• Deleting your account clears everything. "Deleting your account wipes everything Claude Duet has stored."

What It Connects To

Whatever you already connected to Claude, plus the marketplace. Anthropic says Duet "can use the over 6,000 connectors and plugins in the Claude Marketplace", and the examples it gives are almost entirely small-business admin rather than personal life:

If your day does not contain payroll or invoices, note what is missing from that list. Anthropic says so itself: "We built its core functionality around work, so personal errands, such as booking tickets, aren't quite as strong."

How It Compares to the Agents Already Shipping

Claude Duet is the second mainstream agent to get its own cloud computer and its own browser, and the closest thing to it is not another Claude product. It is Meta's Muse, which launched on September 8, 2026 and also runs on its own machine, keeps working after you close the app, and pays with a one-time card through Link by Stripe. The difference is who can have it: Muse has a free tier, and Duet does not.

The one that is cheapest to try is Muse, because it has a free tier. The one with the clearest published boundaries is Claude Duet: Anthropic names the four things it will not do even with your approval, and Meta has published nothing equivalent.

Our evergreen guide to what an AI agent actually is, in plain terms keeps the full table of shipped agents and what each one is allowed to do with your accounts.

Should You Pay for Claude Pro to Get It?

Join the waitlist if you already pay for Pro or Max, live in the US, and the morning-brief idea sounds like it solves a real problem for you. It costs nothing extra and you can stop using it. Watch your usage limits in the first week, because the background checks spend them whether you asked for anything or not.

Do not upgrade from Free for this. You would be paying $20 a month to join a waitlist for a research preview, in a product Anthropic says may misjudge what is urgent, with no published date for anyone else getting in. If you want a reason to pay for Claude, it should be the things that are actually shipping on Pro today, which we compare in which AI chatbot should you pay for.

Skip it entirely if you are outside the US, on Team or Enterprise, or on the free plan. None of those three is in the preview, and Anthropic has not said when any of them will be.

And think twice if your inbox is other people's business. Duet reads the threads you connect it to, including what other people wrote to you. The safeguards Anthropic describes are detailed and it says plainly they will sometimes miss. A research preview is an odd place to find out.

The bottom line: Claude Duet is the first version of the assistant these companies have been describing for two years, and the price of the idea is visible for once. It is free on the plan you already pay for, it is US-only and waitlist-only, it will quietly spend the usage limits you were using for something else, and it asks before anything it cannot take back. If you run a small business on Gmail and QuickBooks, get in the queue. If you wanted something to book your holiday, Anthropic has told you it is not that yet.

Sources

• Anthropic, Introducing Claude Duet (October 21, 2026)
• Anthropic, Claude Duet Safety, Privacy, and Security (October 23, 2026)
• Anthropic, Claude Duet product page (read October 24, 2026)
• Anthropic, Claude release notes, October 21, 2026 entry (read October 24, 2026)
• Anthropic, Claude pricing (read October 24, 2026)