What Is Claude Duet? Its Memory vs OpenAI Dots
Anthropic announced Claude Duet on October 21, 2026. Duet is a proactive agent in its own app that starts your day with a brief and watches your connected apps between messages. Anthropic opened it as a research preview, by waitlist, to US subscribers. Coverage disagreed on whether it runs all the time. Anthropic's own pages settle that question and also say what Duet remembers. You will also see how Duet compares with OpenAI's dots and where its memory stops.
What is Claude Duet, in one sentence?
Claude Duet is Anthropic's proactive personal agent that works with your connected apps on its own cloud computer and asks before it sends, pays, signs or deletes anything.
The product page lists it for Pro, Max and Ultra subscribers in the US. Team and Enterprise plans come later, per the help center. The October 21 release note named Pro and Max only. Duet has standalone apps for iOS and Android, and on a computer it runs at claude.ai.
Anthropic says it "designed it around Claude Opus 5.5," and you cannot pick a different model. Duet can use over 6,000 connectors and plugins from the Claude Marketplace, according to the announcement. Microsoft 365 is not available yet.
How does Claude Duet differ from OpenAI dots?
OpenAI launched dots, its own always-on ChatGPT agent, on September 29, 2026. We covered them in our OpenAI dots explainer. Dot availability has changed since launch, so the OpenAI column follows its help pages as of November 4, 2026.
| Claude Duet | OpenAI dots | |
|---|---|---|
| Status | Research preview by waitlist | Rolling out, generally available to eligible Enterprise workspaces |
| Who can use it | US Pro and Max (the product page also lists Ultra) | Pro outside the EEA, Switzerland and the UK, plus Business Premium and Enterprise |
| Where it works | Its own cloud computer and cloud browser | Its own cloud computer. Your local computer is optional and off by default |
| Account memory | Reads and adds to your Claude memory | Receives ChatGPT memories and can add to them |
| Agent's own memory | Duet keeps its history and memory until you delete your history or reset it | The dot keeps its own context as long as you keep the dot |
| Clearing the agent's memory | Reset from Duet's settings | Delete the dot |
Duet sources: Anthropic's safety page plus its memory help article. Dots sources: OpenAI's getting started guide and dots privacy FAQ.
The two designs are close. Both agents start from your chat memory and write back to it. Both keep a private store of their own on top. The difference is how you clear that private store. OpenAI says you cannot yet change individual dot memories, so clearing them means deleting the dot. Anthropic describes a memory reset that keeps the agent. Neither vendor documents a screen that lists the agent's own memories one by one.
Does Claude Duet run continuously or check every few hours?
Both, for different things. Reports split on this point. ForkLog wrote that there is no continuous monitoring. Every's hands-on review called Duet an always-on agent. Anthropic's pages explain the split: the agent keeps running, and its inbox checks come in rounds.
Duet works on its own computer in the cloud, "so it can keep working after you close the app," per the help center. Anthropic's safety page goes further and says it "runs continuously even after you close the app." Your laptop and phone can be off.
Watching your inbox is periodic. In the announcement's words, Duet "checks your connected accounts every few hours, because checking nonstop would use up your plan's limits much faster." The help center puts it as checks "at set intervals" and warns that something urgent can arrive between them. Anthropic does not publish the exact interval.
A task you hand Duet at night keeps going while you sleep. A new email from your landlord can sit for a few hours before Duet sees it.
What does Claude Duet remember, and how do you delete it?
Duet's memory has two layers, and the first is your Claude account memory. According to the announcement, Duet starts with your Claude memory and with the skills you set up in Claude. Anthropic's memory article says Duet "also uses and adds to your memory." In Claude's settings that memory is a list of topics. You can open any topic to edit it or delete it. Duet can also search your past chats: its own, and your Claude chats outside projects. The setting called Search and reference chats turns that search off in Claude and in Duet. Deleting a chat removes it from search.
The second layer is Duet's own history. The safety page says Duet keeps your conversation history and memory "until you delete your history or reset it." The reset lives in Duet's app settings. Deleting your Claude account wipes everything Duet stored.
If you connect work accounts:
- Disconnecting an app stops new access right away. What Duet already learned from it stays until you reset its memory.
- Duet never sees your passwords or sign-in tokens, and it cannot save them to memory. Sign-in tokens follow the design Anthropic uses for Claude Managed Agents, where the real value never sits on the agent's machine.
A separate classifier checks each outgoing message. It never sends sign-in credentials or full card numbers, even if you ask. Bank account and government ID numbers are blocked the same way. Health and financial details go out only with your consent, and your own request counts as consent.
Does Duet's memory reach Claude Code?
Not according to anything Anthropic has published. Duet reads your Claude account memory and chats. Claude Code keeps its memory somewhere else.
Claude Code's memory docs describe CLAUDE.md files plus auto memory, which Claude writes to a folder per project under your home directory. The docs say auto memory "is machine-local" and that its files "are not shared across machines or cloud environments." Duet runs in a cloud environment. None of the Duet pages we read mention Claude Code memory.
The help center adds that Duet "can't see all of your projects or the files on your computer." It can read your work in Claude but cannot change it. Every's review reports that Duet could view Claude sessions on the reviewer's computers but could not direct them. Anthropic does not say whether that viewing includes the auto memory folder, so we would not count on it.
What do Claude Duet's usage limits look like?
There is no separate price. Duet is included with eligible plans during the preview and draws on the same limits as chat. The help center says it uses your five-hour session limit and your weekly limits "at the same rate as chats in Claude." Background work such as the morning brief and check-ins counts too. Large jobs such as a financial model take more turns and use more.
When you are admitted from the waitlist, you get one full reset of both limits. That reset is available until November 30, 2026. Invitations go out in waves, and access is tied to the email you gave on the waitlist.
What should you read in the fine print?
The capability claims are Anthropic's own. The launch post's testimonials come from Anthropic staff and a few small business testers. The independent coverage we found restates the announcement, apart from Every's review, which is partly paywalled.
Anthropic lists real limits. The announcement says Duet "still might misread a thread or miscalibrate what's urgent." It cannot show you its browser yet, and personal errands such as booking tickets are weaker than work tasks. The help center says Duet does not make phone calls or text other people. Trading securities and moving money between accounts are blocked even with your approval.
On October 23, Anthropic added a note to the announcement saying it is still strengthening these protections. The help center also says no defense against prompt injection is perfect. Review every draft and approval request before you confirm it.
The bottom line
If you are a US subscriber who lives in Gmail and Slack, join the waitlist and start Duet on bounded work such as the morning brief. Expect gaps of a few hours between inbox checks, so keep truly urgent alerts elsewhere. Disconnecting an account does not erase what Duet learned from it, so connect sensitive ones last and check your Claude memory topics after the first week.
How do you share Claude Code memory across machines and agents?
Say Duet spends a week on your billing. It learns which customers pay late, and that lands in your Claude memory. On Monday you open Claude Code on your laptop to fix the invoicing script. Claude Code loads its own auto memory from a folder on that laptop, and Duet has no documented access to it. Move to your desktop and that folder does not come along either, because Anthropic says auto memory is not shared across machines.
| Duet and Claude Code today | memU | |
|---|---|---|
| Who reads it | Duet and Claude chat share your account memory. Claude Code keeps a separate, machine-local memory | One memory backend shared by every supported coding host, including Claude Code and Codex |
| Fixing a wrong memory | Account memory topics are editable. Duet's own history is reset as a whole | Readable Markdown that you open and fix |
| Skills | Duet uses the skills you already set up in Claude | Distills reusable skills from agent history, as Markdown |
| Devices | Claude Code memory stays on the machine that wrote it | Hosted on memu.so, free and synced across devices |
memU covers the coding side of that gap and does not bridge Duet. It does not support Claude chat or Cowork, and Duet is not on its host list. What it does is give Claude Code and your other coding agents one shared memory that follows you between machines.
If your work is mostly email and calendar, Duet's built-in memory is enough. It needs no setup, and your Claude memory already carries over.
To try the hosted version, get an API key from memu.so and send your coding agent one message:
Read https://memu.pro/SKILL.md, follow its instructions to install and configure memU, API Key is
YOUR_MEMU_API_KEY.
The self-hosted setup and the support matrix for each host are in the memU GitHub repository. More on the design is at memu.pro.
FAQ
Is Claude Duet free?
There is no extra charge during the research preview for eligible plans. Duet uses your plan's five-hour and weekly limits at the same rate as chat, and its background checks count toward them.
Who can use Claude Duet?
US subscribers on Pro and Max who get an invitation from the waitlist. Anthropic's product page lists Ultra as well. Team and Enterprise plans are not included yet, and Anthropic says it plans to add them later.
Does Claude Duet work when my phone is off?
Yes. Duet runs on its own cloud computer and keeps working after you close the app. It checks your connected accounts at intervals of a few hours, so new messages can wait until the next check.
Can I delete what Claude Duet remembers?
Yes. You can reset Duet's memory from its app settings or delete your history. Memory topics it added to your Claude account are edited or deleted in Claude's memory settings. Deleting your account wipes everything Duet stored.
Is Claude Duet the same as OpenAI dots?
No, though they are close. Both are proactive agents with their own cloud computers that read and add to your chat memory. Duet runs on Claude Opus 5.5 inside your Claude plan, and dots run on OpenAI's GPT-6 model family inside ChatGPT.