How to Use Muse: The Complete Guide

The mental model: delegate, don't chat

The single most useful thing to understand about Muse is that it is not a better search box. A chatbot answers a question and stops. Muse takes an outcome — "get my inbox under control," "plan the trip," "find me a cheaper flight and book it" — makes a plan, uses the services you've connected, and keeps going until the job is done or it needs you.

That difference changes how you talk to it. With a chatbot, you iterate on prompts. With Muse, you write briefs: what the finished result looks like, what boundaries apply, and what it should check with you about first. Vague briefs get vague work. Precise briefs get surprisingly good work.

Everything below follows from that idea: set it up right, give it one clear job first, connect services deliberately, and keep the approval loop tight until you trust the pattern.

Getting in the door

Muse launched on September 8, 2026, and as of this writing it is available in the United States only, for users 18 and older. You can reach it four ways:

• The iOS app and Android app
• The web app at muse.ai
• Inside WhatsApp, where you message Muse like a contact

Meta says support for its AI glasses is coming. There is no desktop app; the web app covers that.

One thing that surprises people: Muse asks for a payment card during setup, even for the free tier. The reason is that subscriptions kick in as usage rises — the card is there so a heavy month doesn't interrupt your work. Meta says most people will stay on the free tier, and the app includes a usage meter showing what percentage of your allowance remains, warning you before free usage runs out.

Once you're in, resist the urge to connect everything at once. The next two sections explain why.

Your first delegation: the one-job rule

Pick one job you already understand well — something where you'd instantly recognize a wrong answer. Email triage, summarizing a long thread, researching options for a purchase you were going to make anyway. Familiarity is the point: you're calibrating the tool, not just getting the task done.

Then write the brief with three parts:

1. The outcome. Not "help with email" but "go through today's unread email and give me a prioritized summary: what needs a reply from me today, what can wait, what's junk."
2. The boundaries. "Don't send or delete anything. Don't open attachments from senders I don't recognize. If something looks like a bill or a legal notice, flag it separately."
3. The check-in. "Show me the summary before you do anything else."

Run it, read the result critically, and adjust. The first run teaches you more about how Muse reasons than any tutorial — including this one.

Connectors: giving Muse hands

Without connections, Muse is a smart conversationalist. With them, it's an agent. Meta calls these integrations Connectors — links to email, calendars, payments, health, shopping, and smart-home services, with more being added over time.

The design worth appreciating: you connect services one at a time, and each connection asks what access it gets. To connect one, either just ask Muse ("connect my Gmail") or go to Settings → Connectors, tap Connect next to the service, read what's shown, and authorize it. If a service you want isn't listed but has a public API, Muse can set up the connection with credentials you provide; when there's no API at all, it can fall back to operating the service through its browser.

The discipline that matters

Connect a service only when a real task needs it. Every connection is surface area — not just for mistakes, but for your own attention, because each one is something Muse can now act through. Start with one (email is the highest-leverage for most people), get a full task cycle working, then add the next.

You can disconnect a Connector at any time under Settings → Connectors. Note the fine print: disconnecting stops future data exchange, but information Muse already used may remain in its memory and your conversation history. You can also ask Muse to forget specific interactions.

Approvals: staying in the loop

This is the control system, and it's the part to actually learn rather than skim. Before sensitive actions — sending an email, making a purchase — Muse asks for approval. The options you'll typically see:

• Allow once — proceeds this one time
• Allow for this task — this type of action, for the whole task
• Allow for this site — this type of action on this website going forward
• Always allow — for this Connector going forward
• Deny — not this time

The ladder matters: start narrow ("allow once") and widen only for patterns you've watched work. "Always allow" is a convenience you grant after evidence, not during setup.

Three more controls round it out:

• Permission settings (Settings → Permissions) let you set defaults per connector and for web access — including an "always ask" posture if you want maximum oversight.
• The Activity log (tap your assistant avatar) shows a chronological record of what Muse did and what permissions you granted. Check it after the first few tasks; it's how you learn what "normal" looks like.
• You can ask Muse to undo or stop some actions mid-flight. Reversible things (reading a site, comparing prices) can be stopped; sent emails can't — same as doing it yourself.

Goals and background work

Here's where Muse stops feeling like software and starts feeling like staff. You can give it goals — ongoing objectives it works toward over days or weeks — and it will keep working after you close the app, coming back when something changes or it needs a decision.

Practical examples: "watch flight prices for the Chicago trip and tell me when they drop below $300," "every Friday, summarize what shipped in my projects this week," "keep an eye on this bill negotiation and update me." The Goals tab tracks these; the Upcoming view shows scheduled tasks and reminders.

The rule from the earlier sections applies doubly here: don't make anything recurring until you've watched the one-off version work. A background task with a bad brief doesn't just fail once — it fails on a schedule.

Memory and personalization

Muse remembers things across conversations — preferences you state, facts that come up in the course of work — and uses them to make unprompted suggestions. You can also customize the surface: give it a name, pick an avatar, and set how proactive it should be.

This is genuinely useful (it stops asking you the same questions) and mildly unnerving (it knows things you didn't deliberately teach it). The controls: you can ask it to forget specific things, review what's in the Activity log and System Files, and — the nuclear option — reset it, which permanently deletes all information, chat history, files, and active tasks. Reset can't be undone.

One setting worth finding early: the proactivity level. If unprompted suggestions feel like interruptions, turn it down. If you want an agent that surfaces things, turn it up. There's no universally right answer — it depends on how you work.

Usage and pricing

The pricing structure is simple; the details are still filling in.

The honest caveat: Meta hasn't published exact task or message quotas per tier. The usage meter in the app shows your remaining allowance as a percentage, and the app warns you before free usage runs out and offers subscription options. Until the quotas are documented, treat "will I hit the limit?" as a question only your meter can answer — check it after your first heavy week.

Who should pay? If you're delegating a few tasks a week, probably nobody — stay free and watch the meter. If Muse is doing daily background work for you (monitoring, triage, research), Power is the natural first step. Maximum is for people running serious volume through the agent.

What it's good at — and where it stumbles

Genuinely strong: multi-step errands across services (research + book + confirm), inbox and calendar triage, drafting from your context, monitoring things over time, and building small artifacts — documents, trackers, web pages — from a brief.

Where to stay skeptical: anything where a confident wrong answer is expensive — legal, medical, financial commitments. Tasks on unfamiliar websites, where its browser fallback can misread a page. Long chains of dependent steps without check-ins, where one early misread compounds. And taste: it will produce competent, average work unless your brief demands otherwise.

The throughline: Muse is a leverage tool, not an oracle. It multiplies the judgment you put into the brief and the review — it doesn't replace either.

Guide FAQ