Meta launched Muse on September 8, positioning it as a personal AI agent that can book restaurants, send emails, shop online, and handle multi-step tasks without constant supervision.
The agent runs on a dedicated virtual machine inside Meta's cloud infrastructure, using Muse Spark, the company's latest model. You can access it through a standalone app, a web interface, or WhatsApp. Meta says most features will be free, but users who want to "do more" pay $20 or $100 per month depending on usage levels.
Who holds the keys
Muse operates inside Meta's servers. It can log into your accounts, fill out forms, and make purchases—but it does so by controlling a browser environment that Meta provisions and monitors. You never download the model. You don't run it locally. You send your tasks to Meta's infrastructure and wait for results.
Meta says the agent runs in "its own isolated environment" and never sees your actual passwords or payment details. Users can opt out of having their Muse conversations used to train future models—but opting out is a choice you make after signing up, not a default.
The rent goes up
The free tier exists, but Meta has not published exactly what it covers or where the limits sit. The company confirmed that "most users" should stay inside free usage, but added that power users will hit the $20 or $100 ceilings.
That pricing structure mirrors what's happening across the AI agent space: free to try, expensive to rely on. And because the agent lives in Meta's cloud, you can't move it, audit it, or replace it with an alternative that runs on your own hardware.
What you're actually renting
Muse is a workflow manager that charges monthly rent and keeps your conversation logs unless you opt out. If it becomes essential to how you work, you're paying Meta to access your own automation.
That's not a product. It's a lease.