Meta Launches Muse Personal AI Agent with Cloud Computer

Meta launched Muse, a personal AI agent that runs on its own cloud Linux virtual machine with file system, terminal, and Chromium browser. It can write code, browse the web, and perform tasks persistently in the background. Muse can handle parallel tasks, discover deadlines and conflicts, execute multi-step operations across apps (form filling, booking, calendaring), and deliver outputs like PDFs, Markdown, and interactive Artifacts. It sends notifications when human input is needed. Muse is powered by Muse Spark 1.3. Meta published a deep dive on safety and privacy measures, noting that the agent holds extensive user context.
Key insights
Muse Is a Persistent, Autonomous Agent with Its Own Cloud Computer
Meta's Muse provides each user a dedicated cloud Linux VM including a file system, terminal, and Chromium browser. It can run code, execute command-line tools, and use web browsers to gather information and perform actions. Tasks run in the background without requiring the user to stay in the chat. It can parallel process multiple tasks simultaneously.
Muse Proactively Discovers and Notifies Users
With appropriate permissions, Muse can discover deadlines, schedule conflicts, and important updates. It alerts the user when attention is needed, rather than requiring constant prompting. It can execute chains of operations across different applications, such as information gathering, form filling, cart additions, reservations, and scheduling.
Safety and Privacy Built into the Design
Meta stated that because Muse holds extensive personal context over time, it was built with security, safety, and privacy as core principles. A deep dive blog post details the safety system. The agent's ability to store and use personal information is framed as both the source of its usefulness and the reason for rigorous safeguards.
Outputs Go Beyond Text: Documents and Interactive Artifacts
Muse can generate checklists, PDFs, Markdown files, and interactive Artifacts (e.g., travel plans, shopping tables, chess analysis dashboards). This moves beyond simple text responses to deliver usable, structured deliverables.