Meta Releases Muse Personal AI Agent for Web Tasks and App Workflows

Meta has launched a personal AI agent designed to execute multi-step tasks across apps and the web.

According to Meta's announcement of Muse, the agent operates in the background to handle user requests, track long-term goals, and complete tasks such as booking appointments, managing customer service, and filling out web forms. Users can interact with Muse through a dedicated app available on Google Play and the App Store, or directly inside WhatsApp by sending standard messages.

The agent operates inside Muse Secure VM, a persistent and dedicated virtual machine equipped with a web browser. A built-in audit trail displays all past actions alongside upcoming planned steps. Users maintain control over agent operations by reviewing and approving specific actions before execution, such as sending emails, making purchases, or sharing personal data with connected applications. Default permission settings can be configured to allow actions once, always, or not at all.

To handle security, login details are stored in a secure credential store that the agent cannot read, with 1Password integration scheduled for a future update. For transactions, the agent generates a one-time card number at checkout so merchants and the agent never see real payment details. Purchases eligible for protection are covered under Link's purchase protections. Conversations with the agent are kept separate from Meta's ad systems.

Muse connects with daily applications like email, calendar, and Instagram through Connectors. When a specific task requires a non-existent utility, the agent constructs its own tools to complete the work.

Muse is available for free with a usage limit. When users hit their limit, they can wait for the free usage limit to refresh or upgrade to a paid subscription. The system maintains background operations after the app is closed, allowing it to monitor items like daily weather reports without requiring manual check-ins.