Claude Duet: What Anthropic's Proactive AI Agent Does and Why Europe Has to Wait
Claude Duet is an AI agent from Anthropic that has been in research preview since October 21, 2026. It checks your connected apps on a schedule, sends a morning briefing and completes tasks after you approve them. Access is via waitlist for US Pro and Max subscribers only, and the OpenAI and Google equivalents also exclude the EEA for individual users, so building the core yourself is the better move in Europe today.
What is Claude Duet?
Claude Duet is a proactive agent: instead of waiting for your question, it reaches out on its own when something important happens in your connected apps. Anthropic introduced it on October 21, 2026 as a research preview, meaning an early test version. Under the hood it runs Claude Opus 5.5.
Here is how Duet works, according to Anthropic:
- Morning briefing: a short summary of the most important items from mail, calendar and tools, every morning
- Scheduled check-ins: a look at your connected accounts every few hours, not continuously, to save usage
- Its own computer in the cloud: Duet runs on its own virtual machine with its own browser, so your laptop can stay off
- Lots of connections: more than 6,000 connectors and plugins, including Gmail, Slack, HubSpot, QuickBooks, Stripe, DocuSign and Shopify
- One thread: you type or talk to Duet in a single conversation, via the iOS and Android app or on claude.ai
Anthropic explicitly targets professionals and small business owners.
Can you use Claude Duet in Germany or the EU?
No, not yet. The research preview opens via a waitlist for Pro and Max subscribers in the US. There is no date for Europe, which German tech outlet heise also confirmed on October 23. Claude itself works normally in Germany, which is on the list of supported regions. So the restriction only applies to the new agent. On top of that, the help article says Microsoft 365 is not available yet and Slack works with limits. For companies running on Outlook and Teams, Duet would be of limited use even in the US today.
Pricing: Duet is included in Pro and Max and shares their usage limits. According to the pricing page, Pro costs $20 per month (about $17 with annual billing), Max starts at $100. Anthropic warns that background checks and large tasks such as financial models eat noticeably into your allowance. If you already use Claude heavily, Duet will get you to the limit faster.
How do OpenAI and Google compare?
For individual users in Germany the answer is the same everywhere: locked out. Only OpenAI is rolling out its agent worldwide for business plans.
| Claude Duet | OpenAI Dots | Google Gemini Tasks (formerly Spark) | |
|---|---|---|---|
| Model | Claude Opus 5.5 | GPT-6 Astra | Gemini |
| Individual plans | Pro, Max, US only, waitlist | Pro 100/200/500, excluding EEA, Switzerland, UK | AI Pro, Ultra, excluding EEA, Switzerland, UK |
| Business plans | not yet (Team and Enterprise excluded, expansion planned) | Business Premium and Enterprise, rolling out worldwide | work accounts excluded |
| Runs without your computer | yes, its own cloud VM | yes, its own cloud computer | remote browser, mobile only |
Sources: OpenAI docs on Dots, Google help on Tasks. On OpenAI Enterprise, Dots are off by default and a workspace admin has to turn them on. The docs do not say explicitly whether the EEA restriction on Pro plans also applies to Business Premium. Check in your own workspace before you plan around it.
How safe is an agent with access to your mail and accounts?
Anthropic followed up on October 23 with a dedicated security article. The safeguards are serious, but they have clear limits.
What Duet protects:
- Approval before sensitive steps: before sending a message, making a payment, signing a document or deleting something, Duet shows you the action, recipient and content and waits for your okay
- Hard bans: trading securities, moving money between accounts and accessing someone else's account are blocked, even with approval
- No passwords in the model: the model never sees credentials or tokens; a proxy outside inserts them only at the moment of the call
- A second check outside the model: a separate classifier checks whether an action matches your request. This is meant to help against prompt injection, hidden instructions in emails or web pages that try to redirect the agent
- Purchases with a single-use card: via Link from Stripe, limited to one merchant and one amount
What remains open:
- Anthropic itself says the protection will occasionally miss something or flag something harmless
- If you disconnect an app, what Duet learned from it stays in memory until you reset it
- The article does not describe admin controls, and says nothing about whether your data is used for training
- Checks happen only every few hours, so an urgent email can sit for hours before Duet sees it
- Anthropic says Duet can misjudge urgency. A morning briefing that leaves out the one important client email is worse than none, because you rely on it
There is no independent security audit so far, only internal red teams, an external penetration test and a bug bounty program.
What does this mean for you?
For companies in Germany, Duet is a preview of what is coming, not a tool you can use today. Three things are still worth doing now:
-
Build the core yourself. A morning briefing and scheduled check-ins are not rocket science. In Make or n8n the workflow looks like this:
- Schedule trigger: starts the workflow on weekdays at 7:30 a.m., plus at midday if you need it
- Mail and calendar nodes: pull unread emails from Gmail or Outlook, today's meetings and, for example, open invoices from your accounting tool
- Model call: a prompt to a model such as Claude Haiku 5.5 sorts items into urgent, today and later and sums up each one in a sentence
- Delivery: the result goes out as a message to Slack, Teams or by email
You decide which data the model sees, where it runs and what it is allowed to do. Unlike Duet, this also works with Outlook and Teams. We break down the monthly cost of workflow and model in our Make vs n8n comparison and the Haiku 5.5 article. If you would rather have us build the workflow for you, take a look at our n8n automation agency.
-
Build approvals in from the start. Copy Duet's principle: the AI may read and suggest, but sending, paying and deleting only happen after a human clicks. An approval step like that is quick to build in Make and n8n.
-
Sort out permissions now. When these agents reach Europe, employees will want to connect them to their inbox right away. Decide beforehand who may connect which agent to which accounts, and review the data protection question with your data protection officer.
Our take: Google announced its agent Spark (now Tasks) in May, OpenAI has offered Dots since September 2026 according to heise, and Anthropic is now following. When three major providers ship the same product within a few months, it will come to Europe too, only the date is open. The head start belongs to those who already know which routines they want to hand off and which approvals they need. Whether your own tool can serve as a connector for agents like these is covered in our article on the MCP server for your own app. If you want to set up a briefing like this for your team, get in touch.
Frequently asked questions
Do I need a separate subscription for Claude Duet?
No. Duet is included in the Pro and Max plans and uses their regular usage allowance. No separate price has been announced.
Can I get Claude Duet with a US account or a VPN?
We advise against it. According to Anthropic's list of regions, users must be physically located in a supported region, and the Duet research preview is limited to the US. For company data, a workaround like that is also a data protection risk.
Is Claude Duet just a new feature in the Claude app?
No. Duet is a separate app, distinct from the regular Claude app, and it reaches out on its own. It does take over memory, skills and connected apps from your Claude account.
What happens to my data if I stop using Duet?
According to Anthropic, deleting your account wipes everything Duet has stored. If you only disconnect an app, what Duet learned from it stays until you reset its memory.
Sources
Tell us what you're planning. You'll get an honest assessment within 24 hours.
More articles