Anthropic launches Claude Duet, a proactive AI agent, as a research preview
In short
Anthropic has launched Claude Duet as a research preview, an AI agent that works in the background, sends a daily brief and asks for approval before key actions. It is open by waitlist to Claude Pro and Max subscribers in the United States, according to Chinese tech outlets.
Read the full story 2 min read
Anthropic has launched Claude Duet as a research preview, presenting it as an AI agent that can work proactively. IT Home reported that the company announced it on 21 October, while 36Kr, citing Sina Finance, said it was announced on Wednesday. Yicai also carried a headline on the preview launch. [ 3 , 4 , 5 ]
According to cnBeta, Duet compiles a short brief each morning of important items from connected apps. During the day it checks relevant accounts every few hours and sends a notification only when something needs the user's attention. Users can tell it what to watch and how often to alert them. All interaction takes place in one continuing conversation, by text or voice, and Duet can use the memory, skills and connected apps already in the user's Claude account. [ 2 ]
The tasks users can hand over range from scheduling meetings and sorting email to building dashboards, forecasting sales or finances, making prototypes and building web apps, cnBeta and IT Home said. Duet has its own virtual machine and browser and does not need the user's computer or phone to stay on. It notifies the user when a task is done or needs review. Both outlets said it can use more than 6,000 connectors and plugins through the Claude Marketplace, including Gmail, Slack, QuickBooks and Stripe. IT Home added that it can complete user-confirmed online payments through Link or Shop Pay, using virtual cards. [ 2 , 3 ]
On safeguards, both outlets said Duet shows the user what it plans to do and waits for approval before sending messages, making payments, signing documents or deleting content. cnBeta said some actions, such as securities trades and transfers between accounts, will not be carried out even with approval. Anthropic said Duet does not directly see passwords or login tokens. Both outlets said it blocks leaks of credit card numbers and government ID, and cnBeta added that health or personal finance information is sent only when the user explicitly asks. [ 2 , 3 ]
Duet is available by waitlist, first for Claude Pro and Max subscribers in the United States, with standalone iOS and Android apps and a web version. Usage is shared with the user's existing Claude plan, and cnBeta said background checks also count toward the limit. IT Home said it is built on Claude Opus 5.5. Both outlets reported that Anthropic acknowledged it is an early version that may misread the urgency of items, is less mature on personal matters, and cannot yet show users its browser actions. [ 2 , 3 ]
IT Home said Duet joins a contest over always-on proactive AI assistants led by Meta's Muse and OpenAI's Dots. 36Kr said Anthropic described Duet as designed for small businesses and professionals, and that it would likely challenge Dots and Muse. [ 3 , 4 ]
Why it matters
Chinese outlets frame Duet as Anthropic's entry in a contest over always-on AI assistants with Meta's Muse and OpenAI's Dots. The documents describe an early preview with stated limitations, so its wider impact is not yet clear.
Key facts
- Anthropic announced Claude Duet as a research preview, an AI agent that can work proactively. [ 2 , 3 , 4 , 5 ]
- IT Home reported the announcement was made on 21 October. [ 3 ]
- Duet has its own virtual machine and browser, so it can keep working when the user's device is off. [ 2 , 3 ]
- It can use more than 6,000 connectors and plugins from the Claude Marketplace. [ 2 , 3 ]
- Duet must get user approval before sending messages, making payments, signing documents or deleting content. [ 2 , 3 ]
- Access is by waitlist, first for Claude Pro and Max subscribers in the US, with iOS, Android and web versions. [ 2 , 3 ]
- Usage counts against the user's existing Claude plan limits. [ 2 , 3 ]
Confirmed by several sources
- Anthropic launched Claude Duet as a research preview of a proactive AI agent. [ 2 , 3 , 4 , 5 ]
- Duet is offered by waitlist, first to Claude Pro and Max subscribers in the US, and shares usage limits with the user's Claude plan. [ 2 , 3 ]
- Duet runs on its own virtual machine and browser and can use more than 6,000 connectors and plugins via the Claude Marketplace. [ 2 , 3 ]
- Duet asks for user approval before sending messages, making payments, signing documents or deleting content. [ 2 , 3 ]
- Anthropic said Duet does not directly see passwords or login tokens, and blocks leaks of credit card numbers and government ID. [ 2 , 3 ]
- Anthropic acknowledged Duet may misjudge how urgent items are, is weaker on personal tasks, and cannot yet show users its browser actions. [ 2 , 3 ]
Still unclear
- Which model Duet is built on. Only IT Home says it is built on Claude Opus 5.5; the other documents do not mention it.
- Who Duet is designed for. Only 36Kr, citing Sina Finance, says Anthropic designed it for small businesses and professionals.
- Whether Duet will challenge Meta's Muse and OpenAI's Dots. 36Kr says it may well do so, and IT Home says it joins them; no document gives evidence of comparative performance or market effect.
- How Duet's online payments work. Only IT Home says it can complete user-confirmed payments through Link or Shop Pay using virtual cards; cnBeta says some actions, such as securities trades and transfers between accounts, will not be executed even if approved.