Meta has unveiled Muse, a personal AI agent designed to take proactive actions rather than merely answer queries. It can compose emails, arrange travel, negotiate bills, and work toward long-term objectives. Muse continues operating even after you close the app, pausing only when user approval is necessary. For developers, the key architectural innovation lies in its deployment: each user is allocated a dedicated cloud virtual machine, known as the Muse Secure VM, which isolates the agent, its browser, and all credentials in a secure environment.
Deployment and Availability
Muse is a consumer-oriented service, now rolling out in the United States across iOS, Android, and via muse.ai. It offers both free and paid tiers. While developers cannot self-host Muse, its underlying model, Muse Spark 1.3, is accessible through the Meta Model API and Muse Code, with an open-weights release planned per Meta's roadmap.
What Muse Does
Muse is conversation-centric. Users describe tasks or goals, and the agent autonomously plans and executes them. It can open a browser, complete forms, and negotiate on the user's behalf. Meta's examples include optimizing a car sale price, reducing utility bills, or customizing a workout plan. The agent retains context across sessions, converting a saved Instagram recipe reel into a grocery list while remembering friends' dietary restrictions. Sensitive actions, such as sending emails or making purchases, always require explicit user consent. A comprehensive audit trail documents all completed and planned activities.
The Model Behind Muse
Muse operates on Muse Spark 1.3, which Meta Superintelligence Labs released last week. This model likely enhances reasoning and tool-use capabilities, enabling more complex, multi-step tasks. As of 2026, such dedicated-VM architectures are gaining traction in enterprise AI, but Muse's consumer-focused implementation highlights Meta's push toward reliable, secure, and personalized AI assistance. This development reflects a broader industry shift toward proactive agents that balance autonomy with robust safety protocols, marking a significant step in human-AI collaboration.
via MarkTechPost