Meta Platforms has unveiled Muse, a new AI agent designed to carry out tasks on a user's behalf.
Meta said Muse will assist with online shopping, buying movie tickets and scheduling appointments.
What Muse can actually do
Chief executive Mark Zuckerberg described Muse as a personal agent designed to understand a user's goals and work continuously toward accomplishing them.
The system is built to act across apps, connecting to email, calendars, payments, shopping, health, smart home, dining, music and events services.
Once given a task, Muse can open a browser, fill in forms, make purchases and handle bookings without further input.
Where it fits into Meta's AI push
The launch marks the latest step in Meta's year-long shift from chatbots that answer questions toward agentic AI that takes actions on its own.
Each Muse agent runs on its own dedicated cloud computer, known as a Muse Secure VM, which allows it to keep working in the background even when a person is not actively using it.
How Meta says it keeps things safe
Meta said it has layered monitoring and sandboxed execution around Muse to constrain how the assistant accesses accounts and payments.
For checkout, the company said Muse can use a virtual card system that generates a one-time-use number, meaning the agent never sees a person's actual card details.
A separate system called Sentinel reviews the agent's actions before they reach the internet, and can require explicit user approval before sensitive steps, such as sending an email or completing a purchase.
A cautious rollout
Muse is initially available only to US residents aged 18 and over, accessible through a dedicated app on iOS and Android, a website, or directly through WhatsApp.
Meta said the assistant will come to its Ray-Ban smart glasses "soon", without giving a firm date.
The service follows a freemium model, with a free tier expected to cover most everyday use, and two paid tiers priced at $20 and $100 a month for heavier users.
Trust remains the open question
Meta's own internal testing reportedly produced mixed results, with staff praising the agent's usefulness while also flagging disconnections, silent failures and at least one security lapse involving unauthorised exposure of user data.
That tension sits at the heart of the launch, since Muse's usefulness depends on how deeply it can be trusted with real emails, real payments and real personal information, precisely the kind of access that also carries the greatest risk if something goes wrong.
