An agent, not a chatbox
Muse gets its own cloud computer with a browser. You give it a goal; it browses, fills forms, and finishes multi-step tasks — even after you close the app.
Setup guide
From install to first delegation in 15 minutes
Muse launched September 2026 in the US. Here's the fastest path from zero to having an agent actually working for you.
The 30-second version
What Muse actually is
Plugged into your life
Connect email, calendar, Instagram, shopping and more. It reads your school emails, checks your calendar before booking, and turns saved recipe reels into grocery lists.
Asks before it acts
Nothing irreversible happens without your approval. Purchases go through one-time virtual cards, and a separate guard system watches every outgoing action.
The core skill
Most people use Muse wrong
They treat it like ChatGPT: ask a question, get an answer. That wastes the product. Muse's power is delegation — and delegation needs four ingredients:
❌ Weak request
Find flights to San Francisco.
No objective, no constraints, no authority limit. You get a list and do the rest yourself.
✓ Strong request
Plan my trip to San Francisco around the meetings already on my calendar. Nonstop flights only, aisle seat, total trip under $700. Prepare the full itinerary, but ask me before booking anything.
Objective + context (calendar) + constraints (nonstop, aisle, $700) + authority limit (ask before booking).
Every case and prompt in this playbook follows this pattern. Steal the templates →
FAQ
Questions, answered
Is this site affiliated with Meta?
No. usemuse.lol is an independent, unofficial guide. Muse is a product of Meta; this site is made by people learning to use it well.
How is Muse different from ChatGPT?
ChatGPT answers questions. Muse completes tasks: it has its own browser, keeps working after you close the app, connects to your email and calendar, and asks approval before spending money or sending messages.
Is it safe to let it buy things?
Purchases need your explicit approval, go through single-use virtual cards (via Stripe Link) so merchants never see your real card, and a separate guard system reviews outgoing actions. You set the boundaries.
Do the prompts on this site really work?
They're built from Meta's documented capabilities and launch examples, structured around the delegation pattern (objective + context + constraints + authority limit). Results vary by task — start small, verify, then delegate bigger.
1
Install it
Get Muse on iOS, Android, or the web at muse.ai — it's also reachable on WhatsApp. The free tier (100M tokens/week) covers everything below. Note: launch reporting says Meta requires a payment card on file even for the free tier, for abuse prevention.
2
Connect two accounts
Go to settings and connect email and calendar first — they unlock 80% of the useful cases (triage, digests, trip planning). Add shopping, Instagram or health data later, only as you need them. You control exactly what Muse can access and can revoke anytime.
3
Understand approvals
Before Muse sends a message, makes a purchase, or changes anything on an account, it shows you an approval card. Read them for the first week. You'll quickly learn what to auto-approve and what deserves a look. Purchases use single-use virtual cards, so merchants never see your real details.
4
Run your first delegation
Don't start with your taxes. Pick something small, visible, and reversible — the morning inbox digest is perfect. Paste this:
Summarize the emails I received overnight. Format: 🔴 needs a reply today (with a one-line draft), 🟡 worth knowing, ⚪ everything else in one line each. Skip newsletters and promotions. Keep it under 300 words.
5
Graduate to standing goals
One-off tasks are the warmup. The real power is goals that run while you sleep: a price watch, a bill-due guard, a weekly digest. Set one this week — the library has templates.
6
Teach it your constants
Muse remembers what you tell it. Give it the facts it will reuse forever: your home airport, dietary restrictions, kids' school details, budget thresholds. Every future delegation gets smarter without you repeating yourself. And you can ask it to forget anything, anytime.
Learned the hard way
Guardrails from real users
Things early Muse users reported in September 2026 — worth knowing before you hand over the keys.
🔑 Use a dedicated account for risky access
One user granted Muse full Google account access as a test, then switched to a dedicated account with shared edit access. Start there — don't hand over your main inbox on day one.
📧 Give your agent its own email address
One user set up a dedicated inbox just for his agent — signups, receipts, and newsletters flow there instead of his personal email. Cleaner separation, easier to audit what the agent did, and nothing personal leaks into its context. Suggested by @bartclaeys.
✋ Prepare, don't submit
Claim forms, carts, complaint letters, regulator filings: let Muse prepare everything and file nothing. The approval gate is the product — one user had Muse draft a 50-state unclaimed-property sweep without submitting a single form.
🧩 Expect manual takeover on logins
CAPTCHAs and bank logins can still stall the agent mid-task — one reviewer's browser session dropped during a Chase login. Budget a minute of your own time for the tricky auth steps.
👁️ Check what's paired and connected
A journalist found Muse's paired Mac app syncing iMessage notification previews without clear disclosure; Muse apologized and unpaired it. Review connected apps and permissions regularly, and revoke what you don't use.
🐢 Start small or it'll feel slower
Two early critics uninstalled within a day: setup took longer than doing the task themselves, and simple tasks were slower via the agent. They're right — Muse pays off on recurring and multi-step work, not on one-off two-minute jobs.
🔒 Remember who builds it
It's a Meta product with broad access to your accounts. If that bothers you, use the dedicated-account pattern, keep approvals strict, and skip connecting anything you wouldn't hand to a human assistant.
Rules of thumb
Delegate well
Start small, verify, scale
Test with reversible tasks before handing over anything with money or other people involved.
Always set the authority limit
"Ask before booking", "report only, don't cancel" — the one line that keeps delegation safe.
Format the output
Tables, word caps, emoji flags. An agent with a format brief beats a clever agent without one.
Give it memory
Constants (airports, allergies, budgets) turn good one-off prompts into a system that compounds.