Meta just dropped Muse for Mac, marking the first time its AI agent can actually do things on your computer rather than just chatting in a box. Available as a free download for US users today, this desktop extension lets the agent interact directly with native apps like Mail, Calendar, Notes, and Finder. Mark Zuckerberg highlighted the speed of development on X, noting that the team is “shipping fast,” while Chief AI Officer Alexandr Wang confirmed the rollout. This isn’t a new model from scratch; it builds on the mobile and web version that hit number one in the App Store shortly after its September 8 debut.
The real shift here is context gathering without the copy-paste headache. Instead of manually feeding data into a prompt, Muse collates information across your ecosystem. You can ask it to build an end-of-day summary by pulling from emails, Slack messages, and documents simultaneously. It even runs asynchronously, meaning you can start a task on your iPhone, close the window on your Mac, and check the progress later via WhatsApp. The agent stays active in the background, keeping the thread alive across devices. This cross-platform continuity suggests Meta is betting heavily on ecosystem lock-in, not just raw model performance.
Security controls are tight, likely to preempt privacy backlash. Access is strictly opt-in, with Full Disk Access remaining optional. Crucially, destructive actions like deleting files or sending messages require explicit user approval. Behind the scenes, Muse runs on Muse Spark, Meta’s most capable model for agentic work, housed in a Secure VM isolated from other users. A separate Sentinel agent monitors outbound traffic, ensuring nothing leaves the machine without approval. However, Meta admits it can still access data for service operations, with fully user-encrypted Confidential VMs not arriving until 2026.
While the tech is impressive, the geographic restriction limits its immediate impact. Being US-only means millions of potential users are left waiting, and the reliance on cloud processing rather than local inference raises latency and privacy questions for enterprise adopters. The promise of 100 million free tokens per week is generous, but the real test will be whether users trust an AI with keys to their digital kingdom. For now, it’s a powerful convenience tool with significant guardrails, but the long-term privacy trade-offs remain unclear.
