Meta Muse Is Free to Start. Here’s What I’d Connect First.

Last updated September 10, 2026

Ten days ago I wrote about Hatch, Meta’s AI agent that wanted $199 a month to run errands for you. My take then was that the idea was better than the price. The capability looked real, and almost nobody was going to hand over two hundred dollars a month to have a cable bill negotiated.

On Tuesday Meta came back with Muse, and it starts at free. Price was my whole objection, so I don’t have much of one left. I still think cheap AI is a temporary condition and that the free tier exists to get people connected before the meter starts, but right now it costs nothing to find out whether an agent is actually useful to you.

So I went through Meta’s own documentation. With an agent, what it can reach matters more than what it can do.

What Meta Muse Actually Does

Muse takes an instruction and carries it out across the apps you connect, instead of handing you a paragraph about how you could do it yourself.

In Meta’s announcement, the agent opens a browser, fills out forms, and negotiates on your behalf. It keeps working after you close the app and comes back when something changes or when it needs your approval, and the examples Meta reaches for are selling a car for more than you would have, lowering a bill, and adjusting a training plan when the rest of your week moves. It connects to services including Google Workspace, Ticketmaster, OpenTable, Spotify, and Apple Health.

Checkout runs through Link, built by Stripe, which generates a one-time-use card so your real card number never reaches the merchant. Meta says Muse is the first AI agent covered by Link’s purchase protections, and Shop Pay and 1Password support are both listed as coming. It’s US-only for now, and you can talk to it in the Muse app, on the web at muse.ai, or inside a WhatsApp thread, with Meta’s AI glasses next. Axios reported that the whole thing lives in a chat interface that reads like a text thread, and that you can name your agent and give it an avatar.

What Meta Muse Costs

Muse is free to start. Power runs $20 a month and Maximum runs $100, and both buy the same thing, which is more usage.

TechCrunch reported that subscriptions kick in as your usage rises, which is why Muse asks for a payment card before you can start on the free tier. Meta says it expects most people to stay on free, and the app carries a usage meter that shows what percentage you have left and warns you before it runs out. That’s worth knowing going in, because “free” here means free until you use it a lot, not free with no card on file.

Meta went from a price only a small group would pay to a price anyone can try, and it did it in under two weeks.

The Security Setup Is the Interesting Part

Muse runs on its own dedicated computer in the cloud, and a second agent called Sentinel sits between it and the internet.

Meta calls the machine Muse Secure VM. It’s walled off so no other person’s agent can reach it, and it’s where the credentials for every service you connect are stored. Sentinel runs on that same machine but is kept apart from Muse at the system level, and nothing Muse does reaches the internet unless Sentinel clears it first. Meta also says Muse never sees your passwords or payment methods, including ones you type into the browser yourself, because they go into secure storage that Muse can use without reading.

The rest of the permission model is the part I care about. Muse checks with you before sensitive actions like sending an email or making a purchase. It keeps an audit trail of everything it has done and everything it plans to do. You pick which apps it connects to and how much access each one gets, so with email you decide whether it reads your mail or can also send as you. You can change that or disconnect a service whenever you want, and you can opt out of your interactions being used to train Meta’s models. Approve-before-spend has been my line on every agent I’ve written about, and this is that design.

What I’d Connect First

Start with the accounts where a mistake costs you an afternoon instead of your identity.

Spotify, OpenTable, and Ticketmaster are the obvious first three. If Muse books the wrong Thursday at the wrong restaurant, you call and move it. Calendar is a reasonable fourth if you set it to read your schedule rather than write to it, because most of what you want an agent to do with a calendar is understand when you’re busy.

Email is where I’d slow down and use the read-only setting first. Let it draft for a week and see what it produces before you let it send anything as you. I’d leave Apple Health, stored payment methods, and anything touching a bank for later. There’s no reason to lead with your health records when you’re still finding out whether the thing is any good. The audit trail is the feature to actually open after that first week. If you can’t tell what it did, that answers the question.

The Part Meta Can’t Engineer Away

The Sentinel design addresses the technical problem, not Meta’s record.

TechCrunch laid the history out plainly: the FTC settled with Facebook in 2011 over charges it deceived users about privacy, penalized the company $5 billion in 2019 over eight separate privacy violations, and charged Meta in 2023 with violating the order that came out of that settlement. In 2019 the company also found user passwords stored in readable format on its own systems. Reuters reported internal concern that Muse can mishandle access to sensitive personal data.

Whether it earns the access is a different question, and it’s one you get to answer one connected app at a time.

Frequently Asked Questions

Price was my objection, and Meta removed it. Now the question moves to whether you want this particular company holding the keys, and the nice part about a free tier is that you can answer that with two low-stakes apps and a week of watching the audit trail.

Related reading: Meta’s Hatch agent wanted $199 a month | The Instagram setting that stops Meta training on your photos | Gemini task automation across 40+ apps | New to AI? Start here