Anthropic Releases Research Preview of Personal AI Assistant Claude Duet, Available First in the US

On October 21, 2026, Anthropic announced the research preview of Claude Duet, a personal AI assistant. It operates as a proactive agent that lets users know on its own what needs their attention. It continues work on requested tasks even while the user is away, but for things like sending messages or making payments, it checks with the user before executing.

This research preview is initially being offered to Claude Pro/Max subscribers in the United States, and registrations for the waitlist to use it are being accepted.

We’re opening a research preview of Claude Duet, 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 when you step away, and checks with you before it acts on your behalf. https://t.co/kt8gAbWuX8— Claude (@claudeai) October 21, 2026

From morning reports to completion of requested tasks

Claude Duet was developed for people who handle many decisions and tasks in their daily work, as well as for small business owners. Users can specify what they want to be notified about and how often. For example, they can receive a short morning report summarizing important items in connected apps, and notifications also arrive when Claude Duet determines that action is needed.

The AI model used is Claude Opus 5.5, which is said to be designed to distinguish between what requires the user's attention, what can be deferred, and what should be confirmed first. It also carries over memory, skills, and connected apps from existing Claude accounts, so little setup is required when starting to use it.

Users can request things such as scheduling meetings, creating documents, dashboards, and web apps. Interactions with Claude Duet can be continued in a single thread, whether by text input or voice.

For integration with external services, users can use the more than 6,000 connectors and plugins in Claude Marketplace. Anthropic cites use cases such as finding overlooked messages in Gmail or Slack and preparing replies, or checking unpaid invoices in QuickBooks or Stripe and drafting reminder messages.

Upon receiving a request, Claude Duet determines which apps and procedures to use and works on a cloud-based virtual machine equipped with a dedicated browser. It operates even when the user's PC or smartphone is not powered on, and notifies the user when work is complete or when confirmation is needed.

Approval for sending and payments, and mechanisms to protect confidential information

It autonomously proceeds with reading information, drafting, and preparing work, but before sending messages, making payments, signing documents, or deleting anything, it presents the content to be executed and waits for the user's approval. It cannot execute securities trading or transfers between accounts even with approval. Users can also manage app access and disconnect apps they do not want Claude Duet to use.

For purchasing products needed for work, users can use Stripe's digital wallet Link. Claude Duet confirms the product to be purchased with the user, and Link requests approval for the expenditure. Since a virtual card is used for payment, the actual card number is not shared.

For safety measures, it utilizes mechanisms Anthropic has developed for existing agents. The company says it does not show passwords or sign-in tokens directly to Claude Duet, and it also incorporates countermeasures against malicious instructions hidden in external content.

Additionally, to prevent leakage of confidential information, each message is inspected [1]. It is said to be designed to refuse requests to share credit card numbers or government-issued identification information. On the other hand, information related to health or personal finances is allowed to be shared only when the user explicitly instructs it to be sent.

Availability conditions and research preview limitations

Claude Duet is used through a separate iOS/Android app from the regular Claude app, as well as a web version on claude.ai. During the research preview period, Claude Duet is available on Pro/Max plans and uses the same usage allowance as other Claude features. According to the FAQ on the product page, Team/Enterprise plans are not yet included.

The process of checking connected apps in the background also counts toward usage, and checking constantly will use up the allowance faster, so checks are performed every few hours. For larger tasks such as building financial models or creating slides, the number of AI model processing runs and the amount of background work also increase.

Claude Duet is designed primarily for work, and is not as good at personal errands such as booking tickets as it is at work tasks. Also, at present, the browser screen that Claude Duet itself is operating cannot be displayed to the user.

At this stage, it may misread conversations or misjudge urgency. The FAQ also notes the possibility of overlooking information that arrives between checks. Anthropic says it will adjust notification timing and urgency judgments based on feedback from early users.