Claude Duet 2026: Anthropic’s Proactive AI Agent, Hands-On

Claude Duet is Anthropic’s answer to a question it had dodged for most of 2026: what happens when your AI doesn’t wait for you to type something? Announced today as a research preview, Duet is a standalone personal agent. It checks your connected apps on its own, sends you a morning brief, and keeps working on tasks in a cloud machine after you close the app. Meta’s Muse launched in September and topped the US iPhone charts. OpenAI shipped Dots a few weeks later. All three big labs are now in the proactive-agent race. If you pay for Claude, here’s what Duet changes, whether you can get in, and which jobs to hand it first.

What’s actually new in Claude Duet

Regular Claude chat is reactive. You open a window, type a prompt and get an answer. Close the tab and the work stops. Claude Duet reverses that. According to Anthropic’s announcement and early coverage from CNBC and 9to5Mac, Duet does three things chat never did. It sends you a morning brief built from your connected apps. It checks those apps every few hours and alerts you when something needs attention. And it runs longer tasks in a cloud virtual machine that keeps going after you close the app.

Under the hood it runs on Claude Opus 5.5, the same model Anthropic sells through its API. There is no special Duet-only model, so Duet’s reasoning should match what you already get from Opus in chat. What’s new is the harness around the model: scheduling, persistent memory, app connections and an approval layer. Duet is available on the web and as standalone iOS and Android apps, separate from the main Claude app.

The approval layer is how Anthropic handles safety. Duet asks before it sends messages, makes payments, signs documents or deletes files. Reports also say Duet doesn’t see your passwords or login credentials when you grant it account access. For now access is limited to Pro and Max subscribers in the US through a waitlist. Team and Enterprise plans are not yet included in the Anthropic Claude Duet preview.

Why Claude Duet matters

• The default becomes “AI comes to you.” A proactive AI agent brings you the information, so you stop having to remember to ask. If Duet’s checks work, inbox and calendar triage moves from your job to its job.
• Anthropic is going after consumers, not just developers. Claude built its reputation on coding and enterprise work. A standalone personal agent app moves directly into the territory Meta took with Muse.
• The approval model is the product. All three agents can technically act for you. What separates them is how they decide when to stop and ask. Anthropic’s list (messages, payments, signatures, deletions) is the strictest default set so far.
• Pricing pressure is coming. Muse is free to start. Dots sits behind OpenAI’s most expensive tiers. Duet lands in between at Pro and Max, splitting the personal AI agent app 2026 market into three price bands.
• The cloud VM changes what “a task” means. Work that once took one long chat session, like researching vendors, compiling a report or monitoring a thread, can now run unattended and report back.
• Your connected apps become the attack surface. An agent that reads your email every few hours also reads every prompt-injection attempt sent to your inbox. Review its permissions before you hand over everything.

How to use Claude Duet today

Here’s how to use Claude Duet in the first week, assuming you’re a US Pro or Max subscriber. If you aren’t, step 1 still applies, and the prompts below work as templates in regular Claude chat.

1. Join the waitlist. Go to Anthropic’s Duet product page (reported as claude.com/product/claude-duet) and sign in with the account that holds your Pro or Max subscription. Access is tied to the plan, so pick the right account if you have more than one.

2. Connect two apps, not ten. Start with email and calendar. Proactive checks pay off fastest there, and both are easy to audit. Add document storage and messaging once you trust Duet’s judgment.

3. Write standing instructions for the morning brief. A vague brief is just noise. Tell Duet exactly what counts as signal:

   Every weekday morning, send me a brief with:
   1. Emails from clients or anyone I've replied to in the last 30 days
      that are still unanswered. One line each, most urgent first.
   2. Today's meetings, plus any doc or thread I should read beforehand.
   3. Anything with a deadline in the next 72 hours.
   Skip newsletters, promotions, and automated notifications entirely.
   Keep the whole brief under 200 words.

4. Set alert thresholds for the periodic checks. Duet checks your apps every few hours. Decide in advance what’s worth an interruption:

   Only alert me between checks if:
   - A client email mentions "invoice", "refund", "cancel", or "urgent"
   - A meeting on my calendar in the next 24 hours is moved or canceled
   - Someone asks me a direct question and I haven't replied in 6 hours
   Everything else can wait for tomorrow's brief.

5. Give it one long-running background task. The cloud VM has been the most underused feature in every agent launch so far. Pick something bounded that you’d normally put off:

   Research the top 8 email marketing platforms for a small
   WooCommerce store (under 5,000 subscribers). For each one:
   current pricing at 2,500 and 5,000 contacts, WooCommerce
   integration quality, automation features, and deliverability
   reputation. Put it in a comparison table with sources.
   Work on this in the background and send it to me when done.
   Do not sign up for anything or contact any vendor.

6. Treat approval prompts as your training loop. When Duet asks to send a message, read the draft instead of approving it reflexively. Each edit shows you whether its tone and judgment are good enough to trust with more.

7. Audit after seven days. Ask Duet to summarize what it did:

   List every action you took or proposed this week, which ones
   I approved or rejected, and which alerts I ignored. Then suggest
   changes to my standing instructions to cut the noise in half.

For developers

Duet has no public API in the preview. To build a similar scheduled agent into your own stack, use the underlying model on the Claude API now. Pair it with a cron job and your own tools:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-5-5",
    "max_tokens": 1024,
    "messages": [{
      "role": "user",
      "content": "Here are my unread emails as JSON: ... Write a 150-word morning brief, most urgent first."
    }]
  }'

How it compares: Claude Duet vs OpenAI Dots vs Meta Muse

All three run on their own cloud computers, keep memory and work on a schedule. The real differences are who they’re for, what they cost and how they hand control back to you.

Feature	Claude Duet	OpenAI Dots	Meta Muse
Status	Research preview (Oct 21, 2026)	Launched Sept 29, 2026	Launched Sept 8, 2026
Access	US Pro and Max, waitlist	Pro, Business Premium, Enterprise in select markets	Free to start
Product home	Standalone app (web, iOS, Android)	Inside ChatGPT	Standalone app
Model	Claude Opus 5.5	GPT-6 Astra	Meta in-house
Proactive cadence	Morning brief, checks every few hours	Pursues goals around the clock, sends updates	“Heartbeat” check roughly every 30 minutes
Integrations focus	Personal connected apps	Workplace tools like Slack and Teams	Email, Google Docs
Approval gates	Messages, payments, signatures, deletions	Sensitive actions such as password changes	Dedicated approvals tab
Best for	Individuals who already pay for Claude	Teams and professionals	Consumers and small businesses on a budget

Claude Duet vs OpenAI Dots

Dots targets the workplace, with Slack and Teams hooks and enterprise pricing. Duet targets the individual. If your work happens in team chat, Dots fits better. If it happens in your own inbox and calendar, Duet does.

Claude Duet vs Meta Muse

Muse wins on price and polish today, and its 30-minute heartbeat is more aggressive. Duet’s stricter approval gates and the Opus 5.5 model are the reasons to pay. Its slower cadence may also help if Muse’s frequent pings feel like a second inbox.

What’s next

Wider access comes first. A US-only waitlist for Pro and Max is a capacity-limited preview, not a full launch. Watch for international rollout and for Team and Enterprise availability, which would put Duet head to head with Dots for workplace customers. Pricing is the other open question. Duet is bundled into existing subscriptions for now, but always-on cloud VMs cost real money. Expect usage caps or a premium tier once the preview ends.

Integration breadth will decide whether Duet becomes a daily habit or a novelty. Anthropic already has a large connector ecosystem through the Model Context Protocol. If Duet can draw on it, Duet could quickly move past Muse’s Google-centric setup. Also watch how Duet relates to Anthropic’s other agent products, including Cowork and the Managed Agents features released in May. A consumer agent that shares memory and skills with your work agent would be a strong differentiator.

Finally, watch the trust story. The first serious prompt-injection incident with a proactive agent will happen, and it will hit whichever lab has the loosest approval defaults. Anthropic is betting that strictness costs some convenience now and wins on trust later. The preview will show whether users accept that trade.

Frequently Asked Questions

What is Claude Duet?

Claude Duet is Anthropic’s standalone personal AI agent app, announced on October 21, 2026, as a research preview. Unlike regular Claude chat, it works proactively. It sends a morning brief, checks your connected apps every few hours and runs tasks in a cloud virtual machine even when the app is closed.

Who can get the Claude Duet preview?

Only Claude Pro and Max subscribers in the US, through a waitlist. Team and Enterprise plans aren’t included yet, and Anthropic hasn’t announced an international timeline.

Does Claude Duet cost extra?

Not during the preview. It’s included with your existing Pro or Max subscription. Anthropic hasn’t said what pricing will look like afterward, so don’t build a workflow that assumes it stays included.

Can Claude Duet send emails or make payments without asking?

No. Duet asks for approval before sending messages, making payments, signing documents or deleting files. Reports also say it doesn’t see your passwords or login details when you grant it account access.

Is Claude Duet better than OpenAI Dots or Meta Muse?

It depends on where you work. Muse is free and the most consumer-friendly. Dots is built for teams that live in Slack or Teams. Duet suits individuals who already pay for Claude and want Opus 5.5’s reasoning with stricter approval controls.

Is the “claude-duet” project on GitHub the same thing?

No. The claude-duet repository on GitHub is an unofficial open-source tool for sharing a Claude Code session between two developers, and duet.so is an unrelated third-party platform. Anthropic’s official product lives at claude.com and in its own iOS and Android apps.