How to Use Meta Muse: A Beginner's Guide to Meta's AI Agent

Most AI apps work like a conversation: you ask, it answers, you ask again. Meta's Muse works more like an assistant you hand jobs to. You give it a goal ("book a table for four on Friday near my office"), it plans the steps, uses the apps you have connected, opens its own browser when it needs to, and checks with you before anything important.

That difference matters for how you set it up. With a chatbot, a bad answer costs you a few seconds. With an agent, a bad action can send an email you cannot unsend. This guide walks through setup in the order we recommend, so your first week with Muse is useful and not stressful.

What Muse is (in one minute)

• An agent, not a chatbot. Meta announced Muse on September 8, 2026 as "a personal agent" that does the work instead of just answering questions.
• Powered by Muse Spark, which Meta calls its most capable model so far.
• Runs on its own cloud computer. Each user gets an isolated virtual machine (Meta calls it the Muse Secure VM) with a browser. That is why Muse can keep working after you close the app.
• Asks before sensitive actions. A separate monitoring agent called Sentinel reviews what Muse sends out to the internet and pauses for your approval on things like sending email or making a purchase.
• Remembers you. It keeps memories about your preferences across conversations, and you can ask it to forget things.

Before you start: what you need

• A Meta account. You can create one with Facebook, Instagram, an email address or a phone number.
• To be 18+ and in a supported country. At launch that means the US and Canada.
• A device. iPhone or Android app, muse.ai in any browser, the Mac app (released September 17, 2026), or WhatsApp. There is no Windows app yet.
• Possibly a payment card. Several early users report that Muse asks for a card on file even on the free tier. You are not charged unless you upgrade.

Step 1: Open Muse and sign in

Download the official Muse app, go to muse.ai, or open Muse inside WhatsApp. The name "Muse" is used by other unrelated apps, so check the publisher is Meta before you install anything.

Sign in with your Meta account. New accounts go through an identity and age check. Muse then walks you through a short setup, including naming your assistant. You can rename it or change its personality later just by asking.

Step 2: Introduce yourself

Your first message should be context, not a task. Muse saves what you tell it as memories and applies them to later work. Something like:

Hi! A few things about me so you can help better:
- I'm in Chicago (Central time) and work 9 to 5 on weekdays.
- No meetings before 10am, ever.
- I'm vegetarian. My partner is gluten-free.
- I prefer short answers with bullet points.
- Always ask me before sending, posting or buying anything.

That last line is worth including even though Muse already asks before many actions. Stating your rule in plain language gives it one more reason to stop and check.

Step 3: Set permissions before you connect anything

Go to Settings > Permissions. Current guides describe two main modes:

• Always ask: Muse checks with you before any action.
• Ask for some actions: Muse asks before write actions (sending, posting, buying) and important reads, and handles routine steps on its own.

Pick Always ask for your first week. Yes, you will tap Approve more often. That is the point: you will learn how Muse breaks a task into steps, and you can loosen it once you trust what it does.

On desktop, Cmd + , (Mac) or Ctrl + , opens Settings.

Step 4: Check your data controls

Open Settings > Data controls. By default, Meta uses your conversations and tool activity to help train its models, after removing key personal information. If you do not want that, switch off the training toggle here. Meta says Muse conversations are not shared with its ad systems, but that is a separate promise from model training.

We cover every privacy setting in detail in How to set up Meta Muse safely.

Step 5: Connect one app (not ten)

Muse becomes useful when it can see your stuff: email, calendar, documents. It does this through Connectors. There are two ways to add one:

1. Ask in chat: "Connect my Google Calendar." Muse shows what the Connector can access and asks you to authorize it.
2. From Settings: go to Settings > Connectors, tap Connect next to the app, tap Continue, and finish the sign-in on the other app's page.

Where a Connector offers it, choose read-only access first. Your calendar is a great first Connector: it is useful immediately ("what does my week look like?") and read-only access cannot do any damage.

The full list of apps and how each permission level works is in our Meta Muse Connectors guide.

Step 6: Run a low-risk first task

Pick something where a mistake costs nothing. Good first tasks:

Remind me every Sunday at 6pm to plan meals for the week.

Look at my calendar for next week and tell me which days are overloaded. Don't change anything.

Here's a PDF of my apartment lease. Summarize it in 10 bullets and list every date or deadline I need to remember.

Notice what you do not need to do: wait. Muse is not turn-by-turn, so you can send a follow-up while it is still working on the first request.

Step 7: Learn the approval screen

When Muse wants to do something that needs your OK, it sends an approval request. Tap See task details to read exactly what it plans to do. Hands-on guides list these choices:

• Allow once (safest while you are learning)
• Allow for this task
• Allow for this site
• Always allow
• Deny

For tasks that use Muse's browser, you can also open the browser to watch it work, take control to do a step yourself (like typing a password), or stop the task entirely.

A good habit: read the actual email text or order summary in the approval, not just the headline. Sent emails cannot be recalled.

Where everything lives

Once you have run a few tasks, here is where to find the results:

Side chats are worth learning early. They keep a topic (say, planning a trip) out of your main thread. Muse still knows about all your chats, so you do not lose context.

What Muse is good at (and where it struggles)

Early reviews are fairly consistent. Muse shines at:

• Calendar and email triage: "What needs a reply today?" with Gmail or Outlook connected.
• Making files: PDFs, spreadsheets, documents, images and even short audio "podcasts," saved to your Library.
• Bookings with a supported Connector: OpenTable reservations and Ticketmaster tickets work far more smoothly than random websites.
• Ongoing goals: give it something like "help me run a half marathon in March" and it builds a plan and checks in.

It struggles with:

• Sites without a Connector. Muse falls back to driving a browser, which is slower and more fragile. Some sites block AI agents outright.
• Facts it has not checked. Reviewers have caught it inventing a phone number and recommending restaurants that had closed. Verify anything you will act on.
• Multiple inboxes. The Gmail and Outlook Connectors each hold one account at launch.

Your first week: a simple plan

1. Day 1: Introduce yourself, set Always ask, check Data controls, connect your calendar.
2. Day 2: Upload a long document and ask for a summary and a to-do list.
3. Day 3: Set two recurring reminders and a weekday morning briefing.
4. Day 4: Connect email read-only. Ask what needs a reply.
5. Day 5: Try one booking with a supported Connector, approving each step.
6. Day 6: Give Muse a goal with a deadline and let it make a plan.
7. Day 7: Open the Activity log, review what it did, and decide which permissions to loosen.

Want ideas for what to ask? Our list of 30 practical Meta Muse use cases has copy-ready prompts, and How to write better Muse prompts explains why some requests work much better than others.