Meta Unveils Muse for Mac: A Personal AI Assistant That Operates Across Your Files, Email, Messages, Calendar and Notes

Meta has rolled out Muse for Mac, the first edition of its AI agent capable of taking action directly on a user's computer. Rather than living only in a chat window, the agent operates on local files and native applications—the places where user data already sits. This desktop release extends an agent that arrived earlier this month on phones, the web and WhatsApp, marking a significant step in Meta's push to make AI assistants genuinely useful across everyday computing.

What Meta Shipped

Mark Zuckerberg announced the Mac app on X, saying it works across apps, files, calendar, notes and messages, and adding, "The team is shipping fast." Meta's Chief AI Officer Alexandr Wang announced the launch the same day.

Muse itself debuted on September 8th in the US across iOS, Android, the web and WhatsApp. According to TechCrunch, it quickly climbed to the top of the U.S. App Store charts after launch. The Mac version is a free macOS download, currently limited to the US, and is a hosted consumer product rather than an open model users can self-host.

What Muse for Mac Can Do

On the Mac, Muse interacts with files, messages, calendar, notes and mail inside their native applications. Meta positions it around everyday tasks that span multiple apps: organizing folders, completing a form using details stored in your documents, or compiling an end-of-day summary from emails, messages and notes.

The real change is context gathering. A standard chatbot requires users to copy and paste information into a prompt, while Muse can potentially pull together relevant pieces of context across the machine—a single task might draw on a Calendar event, a Mail thread, a document in a folder and a Messages conversation.

Muse also runs asynchronously. Meta says the agent keeps working in the background even when the desktop window is closed. Users can start a task on their phone, check progress from a laptop, or nudge it along through WhatsApp, with Muse maintaining the thread across devices.

How Permissions Work

Because local access raises the stakes, the permission model is central. Access to the computer is opt-in, and Full Disk Access is optional. Users control what Muse can reach and can adjust permissions anytime through Settings.

Sensitive or outbound actions are gated. Meta says destructive operations—deleting files or sending messages, for example—require explicit approval. In practice, Muse can sort a folder on its own but must pause before deleting anything, and it can draft a summary but must ask before sending it.

The Stack Behind Muse

Muse runs on Muse Spark, Meta's most capable model to date, built for real-world agentic work according to the company's launch post. The same model also powers Muse Code, a coding agent available for MacOS and Windows.

On the cloud side, the agent operates inside Muse Secure VM, a dedicated, contained cloud computer that other users' agents cannot reach. A separate Sentinel agent runs on the same machine, isolated from Muse at the system level—nothing Muse does reaches the internet unless Sentinel approves it.

One caveat deserves plain language: Secure VM isolates user data from other users, but it does not stop Meta itself from accessing that data when necessary to run the service. Meta plans to roll out Muse Confidential VM later in 2026, which would encrypt the entire virtual machine with a key held only by the user. Meta is also running a public bug bounty program for Muse.

What to Watch

Muse for Mac remains US-only, and free usage is capped at 100 million tokens per week. The rollout of Muse Confidential VM later in 2026 will be the key test of whether Meta can address privacy concerns and compete with rivals pushing their own desktop agents. Broader international availability will likely follow once the company refines the permission and security model.