Personal AI agent that executes user-authorized tasks across connected services through a dedicated secure virtual machine and can be used through the Muse app or WhatsApp.
Free for most needs; subscription plans are offered for higher usage and additional capabilities, with exact subscription pricing not verified in this research unit.
Muse is Meta's personal AI agent for long-running consumer tasks rather than only question answering. Meta says it can work across connected apps, keep working after a user leaves the app, remember context, suggest follow-up actions and perform tasks such as email, scheduling, shopping, travel planning and purchases.12 It launches in the U.S. through iOS, Android, the web and WhatsApp.3
Its architecture gives each Muse a dedicated cloud virtual machine and browser. A separate Sentinel system mediates network and connector actions, while users can scope access and approvals.13 This is a more explicit control model than a generic browser agent, although the practical reliability of those controls matters as much as the design.
Pricing & access
Meta describes Muse as free for most ordinary use. Multiple independent launch reports identify two paid tiers: Power at $20/month and Maximum at $100/month, providing higher usage allowances.342 The U.S. launch limits geographic accessibility for now.
The dedicated Muse app, web access and WhatsApp lower friction, but using the product meaningfully requires connecting personal services and granting permissions. The App Store itself warns that Muse may be inaccurate or take unexpected actions and tells users to monitor and intervene as needed.5
Evidence & trust
Muse has unusually detailed first-party security framing. Meta says credentials are stored in the Secure VM, Muse itself does not see stored passwords or payment details, Sentinel controls network/service access, and sensitive actions can require explicit approval.1 Wired and TechCrunch independently describe the architecture while emphasizing that these claims still need real-world scrutiny.62
The main caution is not theoretical. Reuters reports that internal testing uncovered security flaws, reliability issues and instances involving private-data exposure; The Information separately reports employee tests with unauthorized actions and sensitive information exposure.47 Those reports concern prelaunch testing and Meta says it spent months hardening the system, but they materially reduce confidence in treating the launch architecture as solved security. No hands-on test by AiToolMap is claimed.
Who it's for
Muse is best suited to users who want one agent to manage multi-step personal workflows across email, calendar, shopping, travel, finances and related services while retaining granular permission controls.15 The free tier makes experimentation easier than with premium-only agents.
It is a weaker fit for people unwilling to expose personal-service context to a cloud agent, users outside the initial U.S. rollout, or anyone who needs a mature independent reliability record before allowing automated actions involving communications, money or sensitive personal data.
Strengths & weaknesses
Strengths
broad task coverage; background execution; free entry tier; iOS/Android/web/WhatsApp distribution; dedicated per-user VM; separate Sentinel control layer; granular permissions; approval gates for sensitive actions; activity/audit visibility; and unusually strong independent launch coverage.132
Weaknesses
U.S.-only initial rollout; high trust and data-access requirements; newly launched product; significant prelaunch security/reliability failures reported independently; no established verified-user track record yet; and premium tiers can reach $100/month for heavier use.47