Meta Muse: What the Personal AI Agent Actually Does

Meta presents Muse as the first personal AI agent designed for everyone, with a one-line thesis: not a chatbot that answers questions but a helper that completes work in the background. The Tool Finder tour opens on the same distinction, arguing that the security framework could be strong enough to persuade users who hesitate to hand personal information to AI systems.

The technical backbone is Muse Secure VM, a dedicated virtual computer per person. This isolated environment hosts the agent's own browser and keeps connected-service data plus credentials inside it. Mark Zuckerberg's comparison in an interview is bold: much like end-to-end encrypted messaging, Meta has no direct interface into that space, so personal data stays better protected. The reference to Microsoft Azure's confidential virtual machine approach is meant to reinforce that claim.

On the model layer sits Muse Spark, described as Meta's most capable family to date, trained for real-world agentic work and under development since April. The app surface is deliberately simple: no technical background is needed, the user states what must be done, and the agent acts. Everyday apps such as Gmail and Google Calendar connect, and jobs from school supply lists to travel arrangements move to the agent.

The shopping and booking flow runs end to end thanks to the built-in browser. The agent can purchase through Amazon and Shopify connections, reserve cinema seats, fill in forms, and create appointments. Payments use single-use card infrastructure with per-connection approval logic: the agent keeps working after the app closes and returns for confirmation before sending mail or charging anything.

The most praised part of the tour is the goals tab. The user defines a target around health, work, or weight, and the agent reflects that context across plans while coordinating time and resources. An ideas tab seeds projects such as a family trip, a personal feed narrows the agenda to stated interests, and an artifacts area stores live-data dashboards such as chess tracking.

Interaction runs through two paths: a standalone mobile app and WhatsApp chat. A status strip at the top shows what the agent is doing right now, and tapping it opens details such as ticket booking or mail drafts. The approvals tab offers one-time or always-allow choices per action. Naming the agent and picking an icon is possible, though the presenter finds the generic AI artwork childish and openly wishes for a more characterful system.

Access is currently limited to the US across iOS, Android, and muse.ai, starting on a free tier with 20 and 100 dollar monthly plans disclosed for heavy use. The tour compares this with Google's upper-tier plans around 99 and 199 dollars and notes a similar band could be debated under intense usage. The United Kingdom remains outside coverage for now. A side story stands out: designer Alex Cornell, who appears in the walkthrough, led Cocoon, the app Meta acquired years ago, and that lineage is said to show in the polished design.