Meta has officially launched Muse, a groundbreaking personal AI agent that goes beyond mere question answering. Muse is capable of executing various tasks such as sending emails, booking travel, negotiating bills, and working towards long-term objectives. Notably, it continues its operations even after the app is closed, only seeking user approval when necessary. A significant aspect of this release is its architecture; each user is provided with a dedicated virtual machine, known as Muse Secure VM, ensuring that the agent operates in a secure and isolated environment. Currently, Muse is available as a consumer service in the US on platforms like iOS, Android, and muse.ai, with options for both free and paid subscriptions. While developers cannot self-host Muse, they can access its foundational model, Muse Spark 1.3, through the Meta Model API and Muse Code.
Muse is designed around user interaction, allowing individuals to describe their tasks or goals, which the agent then plans and executes. It can browse the internet, fill out forms, and negotiate on behalf of users. For instance, it can assist in selling a car or adjusting a training regimen. Muse retains context from previous conversations, enabling it to create grocery lists from saved recipes while considering dietary preferences. Importantly, sensitive actions require user consent, and a comprehensive audit trail is maintained for transparency.