Meta introduces Muse, a personal AI agent that works in the background

Meta has introduced Muse, a personal AI agent it describes as taking everyday busywork off users' plates. The product page presents it as an agentic assistant that works across devices: a mobile app for Android and iOS, a Mac app, and conversations directly inside WhatsApp.

What Muse does

According to Meta, Muse can answer questions, complete tasks, browse the web, make purchases, generate images and create documents. Through connectors it links to services such as email, calendar and Instagram, and if a task needs a tool that does not exist, Muse builds it. It can also set reminders, track goals and monitor things a user cares about, continuing to work in the background after the app is closed — for example, watching daily weather reports and alerting about rain.

A dedicated virtual machine and an approval gate

The agent runs on what Meta calls the Muse Secure VM: a persistent, dedicated virtual machine with its own browser, which books appointments, fills in forms and handles customer service enquiries. Critical actions, such as sending emails or making purchases, can be reviewed and approved before they happen, and users get an audit trail of everything the agent has done and what it plans to do. A permission prompt can be allowed once, always allowed, or denied.

Security and data handling

Meta says logins are kept in a secure credential store the agent cannot read, with 1Password integration coming soon. When shopping, a one-time card number is generated at checkout so neither the merchant nor the agent sees the real card, and eligible purchases are covered by Link's purchase protections, which Meta calls a first for AI agents. Conversations are not shared with Meta's ad systems.

Availability and limits

Muse is available for free with a usage limit. Users who reach the cap can upgrade to a paid subscription or wait for the limit to refresh. Meta says the agent is built with automated protections and agentic AI security standards, and that it asks for permission before acting on a user's behalf.