Set up Meta Muse in one afternoon

Muse is Meta's new personal AI agent, and it's different from the chatbots you've used so far. It can read your email, send messages for you, book things and pay for things. That's exactly why the setup matters more than the first prompt.

Meta launched Muse in the US on September 8, 2026. This guide is how I'd set it up on day one so it's useful from the start and only does what you've approved. Muse is brand new and Meta is still adding features, so the menus may not match exactly. Use the labels you see in your app. Last checked 27 Sep 2026.

What you need to know before you sign up

• Who can use it. It's rolling out in the US for adults 18 and over, on iOS, Android and the web at muse.ai. Meta has said glasses and a Mac app are coming.
• What it costs. There's a free tier with a weekly usage limit, plus paid plans at $20 and $100 a month for heavier use. Meta has said it expects most people to stay on free, and that over time it plans to make money by taking a small fee from transactions Muse makes for you.
• The card. Launch coverage reports Meta asks for a payment card on file before you can use Muse at all, including the free tier. Use a card with a low limit or a virtual card if your bank offers one.
• How it works. Each person's Muse runs on its own computer in Meta's cloud (Meta calls it the Muse Secure VM), with its own browser. That's why it can keep working on a task after you close the app.

The five steps, in order

The order matters here. You lock down the privacy settings before you give it anything, and you give it access one app at a time, starting read-only.

A real one, set up end to end

Here's the whole thing for one person: a freelance designer who suspects she's paying for software she forgot about. It's a good first job for Muse because it's mostly reading, the payoff is real money, and she stays in control of every send.

• Settings first. She turns off training, names her Muse, and writes in Soul that she's a freelancer, wants answers as short lists, and wants nothing sent or bought without her approval.
• One connector. She connects Gmail read-only and tests it by asking for the last invoice from her biggest client. It finds it.
• The read-only task. She asks Muse to go through the last six months of receipt and renewal emails and list every subscription with the price, how often it bills and the next renewal date. It comes back with eleven, three of which she forgot existed.
• The check. She compares the list to her card statement herself. One "subscription" was a one-time purchase, so she tells Muse to drop it and to mark anything it isn't sure about.
• The first action. She picks two to cancel, switches Gmail to allow sending, and asks Muse to draft the cancellation emails. It asks for approval before each one, she reads them, and taps Allow once.

By the end she's got a full list of what she pays for, two cancellations sent, and a job she can ask Muse to rerun every quarter. Muse never touched her bank or card to get there.

Copyable starters

Paste this into Soul (assistant icon, then Identity, then Soul) and fill in the brackets. It covers how you like to work and, more importantly, what it has to check with you first.

I'm [what you do]. Right now I want help with [one or two goals].
Give me the answer first, as a short list, then anything I need to act.
If you aren't sure about a fact, a price or a date, say so. Don't guess.
Always show me a draft and wait for my approval before you send,
post, buy, book, cancel or change anything in any of my accounts.
Never share my information with a person or site I haven't named.
When a task is done, tell me what you did and what you didn't finish.

Then give it the first job. This is the subscription audit from the example above. It only needs read access to your email.

Look through my email from the last 6 months for receipts, renewals
and billing notices. List every subscription I'm paying for with:
the service, the price, how often it bills, the next renewal date,
and the email you found it in. Mark anything you're not sure is a
subscription. Don't cancel, reply to or change anything. Just the list.

And here's the checklist I'd run before calling day one done.

[ ] "Help improve our AI models" is off in Data controls
[ ] Soul says what it has to check with me before it acts
[ ] One app is connected read-only, and I tested it on a specific email
[ ] Approvals are on, and I'm using "Allow once" while I learn
[ ] I ran one read-only task and checked its answer against the source

What I'd be careful about

Muse is less than a month old and it's getting access to a lot. A few things I'd keep in mind before you connect more.

Deleting isn't the same as forgetting. Meta's help center says that after you delete something or disconnect an app, Muse may still remember what it learned from it. You can ask it to forget a topic or edit the Memory file directly, but the safest move is not to hand it things you'd want back.

Some sites don't want it there. Amazon started blocking Muse from shopping on Amazon.com on September 21, saying the agent doesn't identify itself and violates its terms. Expect more of this, and don't build a routine around a store that hasn't agreed to it.

Custom connectors are on you. Muse can build a connector for an app that isn't on its list, but Meta says it doesn't review custom connectors or how they use your information. Stick to the listed ones until you know what you're doing.

People who do security for a living are split. Some have said publicly they wouldn't give any company this much access. Meta says passwords and cards sit in a separate credential store Muse can't read, and it's promised an encrypted option later this year. Both can be true. Decide what you're comfortable with one connector at a time.

Once Muse is handling one job well, the same setup applies to the rest. You can see how the steps compare in setting up Claude in a day and setting up Grok in a day. If you want to go further with it, the next step is building your first AI agent around one recurring job.