Meta's Muse Agent Gives Every User a Cloud-Hosted Ubuntu Linux Machine

Every user of Meta's Muse agent now gets a free, full-featured computer in the cloud, according to David Singleton, VP of Engineering at Meta Superintelligence Labs and former CTO of Stripe. The machine runs a complete Ubuntu Linux image where users can install software, write and compile code, or browse the web.

Singleton says the cloud computer was designed so users can do almost anything they would do on a physical machine at their desk. The feature places Muse in direct comparison with OpenAI's ChatGPT, whose "Work" mode was recently found by users to run its own virtual machine with 15 GB of RAM and more than nine CPU cores.

Sentinel monitors while the Runtime Cell stays open

According to Singleton, the underlying architecture — called "Muse Secure VM" — separates the user workspace from sensitive system components. Users and their Muse agent operate inside what Meta calls a "Runtime Cell," where activity is largely unrestricted. A separate process called "Sentinel" runs outside the cell and monitors sensitive actions.

Passwords and credentials are stored outside the Runtime Cell entirely. The cell also gets its own root filesystem, isolated from the host system. Meta claims this prevents anyone who breaches the sandbox from gaining privileged access to Meta's broader infrastructure or to other users' data. The design is also intended to guard against prompt injection attacks, according to Singleton.

Full file-system transparency

Singleton says Meta deliberately built transparency into the Runtime Cell. Users can inspect every file inside it, from Debian system files to the binaries that run Muse itself. A file explorer is built directly into the app's Library tab, and the Markdown files Muse generates while reasoning through a task are visible to users as well.

The inclusion of a classic file explorer view is notable for a consumer AI app — it exposes low-level system detail that most mainstream products abstract away entirely. Less technical users have a simpler path: Settings > Data controls > Download your agent data, which exports their files without requiring them to navigate the filesystem directly.

Muse's early traction

Muse pulled in more than 500,000 users in its first week and reached number one in the Apple App Store, according to the report. At Meta Connect 2026, the company added real-time video chats, dedicated email addresses for the agent, and the ability to control Mac applications directly.

Meta was an early mover in the generative AI race with its Llama model family but has fallen behind OpenAI and Anthropic on frontier model capability. With Muse, the company appears to be betting that product breadth and distribution — rather than raw model performance — will determine which AI assistant wins mainstream adoption.

What this means

Giving every user a persistent, full Linux environment is a meaningfully different bet than shipping a better chatbot. It signals Meta is chasing the same "agent that can actually do things on a computer" category that OpenAI's Work mode and various coding-agent startups are targeting, rather than competing purely on model benchmarks. The security architecture — isolating credentials and root access outside the sandboxed cell — addresses a real risk with computer-use agents: a single prompt injection or malicious webpage could otherwise expose a user's entire system. Whether Meta's approach holds up under adversarial testing, and whether running a cloud VM per user is economically sustainable at scale, are the open questions that will determine if this becomes a durable product or an expensive experiment.