Meta launches Muse personal AI agent across US services
The software runs independent virtual machines to handle user tasks and charges up to $100 monthly for heavy compute tiers.
Meta rolled out Muse on September 8, 2026, introducing a personal artificial intelligence agent designed to carry out digital tasks without requiring technical background from consumers.12 The rollout covers users across the United States through web browsers, mobile operating systems, and messaging software.13 The deployment marks an operational shift from standard conversational chat windows toward autonomous background execution.13
Alexandr Wang stated that the product succeeded by making automated systems accessible to non-technical users.4 Wang noted that watching people outside technology adopt agents without having to learn concepts like chain of thought, model context protocol, or command line interfaces reflected long-term development goals.4
The system operates inside an isolated Linux environment termed the Muse Secure VM, which houses a dedicated browser, storage volume, and processing capacity.13 A separate security process named Sentinel governs outbound network traffic and enforces user permissions.12 System developers Mona Sarantakos and Christine Awad stated that Muse possesses its own computer interface, enabling the agent to search websites, fill out web forms, and execute consumer bookings directly.52
Platform onboarding pairs each profile with an avatar, customizable tone, and integrated task storage known as Library.16 Software reviewer Claire Vo tested the agent on household scheduling, noting that the software generated a structured family schedule document from calendar entries and inbound correspondence.6 The agent functions autonomously after users exit the application, continuing multi-step research until tasks complete or require manual confirmation.23
Meta paired the agent with payment protections through Link by Stripe, generating single-use virtual card numbers during digital transactions so that raw billing data remains hidden from external merchants.12 Sarah Perez reported that the service connects directly to user mailboxes, calendars, and payment services, while keeping credentials sequestered within an encrypted vault that the operational model cannot inspect.32
Commercial access relies on a tiered allocation model.13 Mark Zuckerberg established a baseline free compute ceiling of approximately 100 million tokens per week.1 When consumers exceed standard allowances, paid tiers charge $20 monthly for Power access and $100 monthly for Maximum usage.31 Eric Hal Schwartz tested the automated workflows on personal errands, noting that while the delegation delivered practical assistance, handing over access to personal services created initial hesitation.
Meta announced plans to introduce an upgraded Confidential VM infrastructure later in 2026, incorporating client-held encryption keys to isolate user records from administrative access.21 The company also disclosed upcoming hardware integrations that will extend the software from personal screens to wearable smart glasses.23
What this rests on
18 sentences trace to 6 sources.
-
1
QA
QWE AI Academy Third-party reporting · qwe.edu.pl · 9 Sep 2026 How to Use Muse – Meta’s Personal AI Agent [2026] See the source
-
2
C
CNBCTV18 Third-party reporting · cnbctv18.com · 9 Sep 2026 What’s Muse, world’s first personal AI agent built for everyone - CNBC TV18 See the source
-
3
T
techcrunch.com Third-party reporting · techcrunch.com · 8 Sep 2026 Meta debuts its Muse AI agent. Will consumers trust it? | TechCrunch See the source
-
4
H
https://x.com/alexandr_wang Document · x.com · 21 Sep 2026 Alexandr Wang Says Muse Can Handle Everyday Appointment Research for Users See the source
-
5
M
Muse Third-party reporting · introducing.muse.ai · 8 Sep 2026 How We Designed Muse See the source
-
6
LN
Lenny's Newsletter Third-party reporting · lennysnewsletter.com · 16 Sep 2026 Muse review: The personal AI agent that gets consumer UX right See the source