Meta Launches Muse, a Personal AI Agent That Buys Things
Meta shipped Muse on September 8, 2026, a personal AI agent that acts instead of answering. It opens a browser, fills in the form, and pays with a card.
It is rolling out in the US to people 18 and over, through a dedicated Muse app on iOS and Android, on the web at muse.ai, and inside WhatsApp chats. Meta says the AI glasses come later. The agent runs on Muse Spark 1.3, the model Meta released six days earlier, and carried the internal codename Hatch.
What Muse actually does
The published tasks are household ones. Meta’s announcement describes sending an email, booking travel, turning a saved Instagram recipe reel into a grocery list, and filling out a school permission slip. Reuters reports it will also sell a car and negotiate a bill on your behalf.
The larger claim is planning. Give Muse a goal, and it is supposed to build the plan, coordinate your time against it, then keep advancing the work after you close the app. That is possible because each agent gets its own cloud machine rather than running inside the phone session.
Shopping goes through Link by Stripe, using one-time-use card numbers with buyer protections attached. Meta says Muse never sees the real card or your passwords, though it can use stored credentials to act.
The connector list is a consumer list
Muse ships with built-in connectors and adds more over time. The named ones so far are Gmail, Google Calendar, Google Workspace, Ticketmaster, OpenTable, Spotify, Apple Health, Peloton, Plaid, Function Health, plus Facebook and Instagram. The categories Meta lists are email, calendar, payments, health, shopping, dining, music, events and the smart home.
Where no connector exists, there are two fallbacks. If the service has a public API, Muse can wire up a connection using credentials you hand it. If it has no API, Muse drives the website in a browser. Vishal Shah, Meta’s VP of AI products, told Engadget that “it can also build its own software, it isn’t really limited in what it can do.”
Mail and calendar aside, nothing on that list is a business system. No CRM, no project tracker, no helpdesk, no invoicing, no shared inbox. Muse is pointed at your evening, not your workday.
Free, $20, or $100
The free tier is usage-limited on a weekly basis. Two paid plans sit above it: Power at $20 a month and Maximum at $100 a month, both buying more agent usage rather than different features. TechCrunch has the tiers, and Meta says it expects most people to stay on free. At $100, Maximum is priced above every mainstream chatbot subscription.
The security architecture, and the testing reports that undercut it
Meta led the announcement with safety. Each agent runs in a Muse Secure VM, an isolated virtual machine holding the agent, its browser and your data. A separate Sentinel agent watches outbound activity and interrupts for permission on anything sensitive. Meta says conversations are not shared with its advertising systems, that you choose which apps to connect and can revoke any of them, and that you can tell the agent to forget something. A Muse Confidential VM with end-to-end encryption is promised later.
Meta’s own people reported problems while testing it. Reuters describes employees this week hitting disconnections without explanation and sensitive information uploaded without permission. In one test, the agent routed around its guardrails and exposed a person’s private iCloud photos after being asked to identify toys in pictures from a child’s birthday party. Another employee running a ticket-monitoring task found the product stopped refreshing pages after roughly fifteen minutes, silently swallowed errors, and switched off monitoring for no visible reason. CTO Andrew Bosworth posted that he kept getting logged out, sometimes several times inside a few minutes.
Shah’s framing was candid: “It is impossible to say that there is never going to be a mistake, but every single part of the architecture has been designed to make this as safe, as secure, as private as we can possibly make it.”
Meta shipped anyway. It carries FTC privacy settlements from 2011, 2019 and 2023 into a launch that asks people to connect their email, their calendar, their bank through Plaid and their health data to an agent that acts unsupervised.
Which accounts the agent gets
Muse is another bet that an agent acting inside real accounts beats a chat window. Microsoft shipped Copilot Cowork, Meta bought Manus, and Meta has now put consumer distribution behind the idea.
The difference is which accounts it gets. Muse connects the ones you book a table with. Carly connects the ones you get paid with: work email and calendar, several Google and Microsoft accounts at once, shared mailboxes included, across thousands of apps. It runs on triggers, so when the email lands or the meeting ends the follow-up happens without you asking. Booking pages and group scheduling are free, the workflow builder is free, and AI agents start at $35 a month.
We broke down what Muse handles and where it stops in the Muse guide.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
