Meta Releases Muse, an AI Agent That Does Tasks Instead of Answering Questions
Meta began rolling out Muse, a personal AI agent, in the United States on Sept. 8. The company says it runs in its own app on iOS and Android and at muse.ai, that it can be used inside WhatsApp, and that it is coming soon to AI glasses.
Meta describes Muse as an agent that does work rather than answering questions. In its announcement, the company says that Muse can send an email, book travel, open a browser, fill in forms, and complete a purchase; that it keeps working after the app is closed; and that it comes back when something changes or when it needs approval. No independent evaluation of Muse has been published.
The company says each user's agent gets a dedicated computer in the cloud, called Muse Secure VM, holding the agent and that person's data, where no other user's agent can reach it. A separate program on the same machine, named Sentinel, has to approve anything Muse sends to the internet. Credentials go into secure storage, Meta says, so the agent can use them without seeing them, and Muse asks before it sends an email or buys something. The company says conversations and the data on that machine are not shared with its advertising systems.
Purchases run through Link, built by Stripe, which Meta says issues a one-time-use card so a user's real card details stay hidden. Meta calls Muse the first AI agent covered by Link's purchase protections.
Meta's announcement is headed "Introducing Muse: The World's First Personal AI Agent Built for Everyone," and describes the model behind it, Muse Spark, as "Meta's most capable model to date." Neither claim is independently established in the announcement. Muse Spark was introduced in April.
The company says Muse is free for most uses, with subscription plans for people who want more, and that a version called Muse Confidential VM, encrypting the whole virtual machine with a key only the user holds, will follow later this year.
