Mark Zuckerberg describes personal use cases for Muse, including baking help, permit monitoring, and MMA footage analysis
Following·Day 6·first covered 3 Sep 2026·6 sources
Meta Launches Muse Personal AI Agent on Dedicated Secure VM
The gist
Muse gives a major social platform operator persistent background access to personal data across email, payments, and third-party services. The VM-plus-Sentinel architecture is intended to limit what the agent itself can read directly, and Meta has stated plans for a Confidential VM mode that would extend that isolation to Meta's own infrastructure.
The full picture
Meta launched Muse on September 8, 2026, a personal AI agent built on its Muse Spark 1.3 model that runs on a dedicated Muse Secure VM per user. A separate Sentinel component runs on the same VM but stays isolated from Muse at the system level, approving, blocking, or escalating every outbound action; Meta states Muse never directly sees passwords or payment details. Mark Zuckerberg described using Muse to help his daughter with baking, monitor mountain climbing permits, and analyze MMA training footage from gym cameras, calling it "something much closer to a persistent personal worker than a chatbot." Muse integrates with Gmail, Spotify, Ticketmaster, Shopify, OpenTable, Instagram, Facebook, and WhatsApp, processes payments via Link by Stripe using one-time virtual cards, and continues working after the app closes. It is available in the US for adults on iOS, Android, web, and WhatsApp, with a free tier at up to 100 million tokens per week and paid tiers at $20 and $100 per month. Meta plans to later offer a Confidential VM mode in which users hold their own access keys locally, preventing even Meta from accessing the VM. Benchmark tests showed Muse Spark 1.3 used roughly 20% fewer tool calls and 25% fewer tokens than its predecessor, with independent results placing it close to Google's Gemini 3.8 Flash overall, Muse ahead on some intelligence benchmarks and Gemini faster.
How it developed
Meta launches Muse personal AI agent in the US, powered by Muse Spark 1.3, with dedicated Secure VM and Sentinel architecture
Benchmark comparison of Muse Spark 1.3 and Google's Gemini 3.8 Flash published, showing models close with tradeoffs on speed, cost, and task accuracy
Sources
Meta AI Twitter
Rohan Paul Twitter
Milk Road AI Twitter
1 more source
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free