Meta debuts Muse, its long-planned personal AI agent

Driving the news: Muse, as the agent is known, exists in a chat interface, similar to a text thread.

• It's designed to be more proactive and long-running than typical chatbots. Users can name their agent, create an avatar and customize how it communicates.
• The Muse agent runs on a dedicated virtual machine in Meta's cloud, using a built-in browser that's visible to the user.
• Meta is offering a free tier of Muse, as well as two subscription options, at $20 per month and $100 per month.
• "For the vast majority of users, they should be able to do what they need to within the free tier," Wang told Axios. "But for real power users, you know, those subscription tiers help us cover the the compute costs,"
• There is no advertising within Muse, but Wang said the company is exploring commerce opportunities that could generate additional revenue.
• Initially Muse will be available in the U.S. and works on iOS, Android and the web, with support coming soon for Meta's AI glasses.

The big picture: Wang described Muse as an early but big step toward Meta's broader goal of "personal superintelligence."

• "The full vision in the future is we want to develop personal superintelligence that helps people accomplish their goals, pursue their passions, build things that they never would have built if they didn't have the technology," he told Axios.

Between the lines: Meta is offering more privacy options than with its prior AI products, as Zuckerberg suggested in his manifesto.

• The virtual machine itself is designed to be private to each person, although it runs on Meta's infrastructure.
• By default queries made to one of Meta's models are able to be used by the company, but there is an option to turn that off.
• Meta is also working on a confidential version where Meta can't see what is taking place inside the user's virtual workspace. It's aiming to release that before the end of the year.

Zoom in: A separate system called Sentinel controls Muse's access to the internet and connected services. Muse can suggest an action, but Sentinel determines whether it is allowed, blocked or sent to the user for approval.

• Meta says people can distinguish between read and write access, including setting more granular permissions within a service. Permissions can be narrowly scoped to a particular task, service, transaction or period of time.
• At the same time, executives said they were mindful that asking people to approve every small action can lead to them approving requests reflexively, which can undermine safety. Muse is designed to require approval for more sensitive actions while allowing previously authorized, lower-risk tasks to proceed.

Striking this balance is important given the very real risks involved. In recent months there have been examples having data inadvertently deleted.

• Meta's own Summer Yue went viral after posting her account on X of having an OpenClaw agent delete files from her Mac after she gave it access to her email.
• Other services are also trying to prepare for a flood of agents all persistently trying to get tickets or services or gear on behalf of users. Resy, for example, has made it clear it will delete accounts of those using automated agents.

What we're watching: Whether mainstream consumers are willing to grant an AI agent progressively more access to their personal data — and if Muse can prove useful enough to make the risk a worthwhile one.