What Muse is, and what it is not
Muse is Meta's personal AI agent. It launched on September 8, 2026 for users in the United States. It connects to apps you already use (email, calendar, payments, health, smart home, dining, shopping, music, events) and performs tasks on your behalf: sending emails, booking travel, lowering bills, filling out forms, building plans, sending invitations and making purchases. It asks for your approval before it spends, sends or shares anything, and it keeps working after you close the app.
It runs on Muse Spark, Meta's model, and is available at muse.ai, through iOS and Android apps, and inside WhatsApp. Meta says AI glasses support is next. It is free with a usage cap. A payment card is required at signup. Paid plans are Power at $20 per month and Maximum at $100 per month.
Name collision warningMeta is using the Muse name across several products. If you search for help, you will hit all of them. Know the difference before you follow instructions from the wrong one.
| Product | What it is | Who it is for | Covered here |
|---|---|---|---|
| Muse | The personal AI agent that connects to your apps and does tasks with your approval | Consumers and, by extension, solo operators | Yes |
| Muse Spark | The AI model underneath Muse, also available to developers via the Meta Model API | Developers | No |
| Muse Image / Video | Media generation models inside the Meta AI app, meta.ai and Instagram Stories | Creators | No |
| Muse Code | A multi-agent coding tool in the terminal | Developers | No |
Fit check
Do this before you create an account. The most expensive mistake with any agent is connecting everything, giving it vague instructions, and then judging it on whether it read your mind. Diagnose first. Build second.
Procedure- List the admin that repeats
Write down every task you did more than twice last week that had nothing to do with delivering for a client or getting a new one. Inbox sorting, rescheduling, follow-up reminders, receipts, subscription checks, booking anything, filling out the same forms.
- Score each task on three things
Frequency (how often), rule clarity (could you write the rule on an index card), and consequence of error (what happens if it goes wrong once). Score each 1 to 3.
- Keep only the high-frequency, clear-rule, low-consequence tasks
That is your Tier 1 delegation list. Anything with a high consequence of error stays with you until Muse has earned trust on the easy stuff.
- Estimate the hours
Add up the weekly minutes across the Tier 1 list. If it is under 60 minutes a week, Muse is a curiosity, not a system. Bookmark this page and come back when the list grows. If it is over 60 minutes, proceed.
Decision rule
If you cannot write the rule for a task on an index card, you cannot delegate it to an agent yet. Fix the rule first. That is true for a VA, a junior hire and Muse.
Setup
Estimated time: 20 minutes. Goal: one clean account, one primary surface, an agent that talks to you the way you want to be talked to.
- Create the account on the web first
Go to muse.ai and sign up there rather than in WhatsApp. The web surface gives you the fullest view of settings, connections and the usage meter. Install the mobile app second.
- Add the required card, then set your own ceiling
Muse requires a payment card even on the free tier. Use a card with a low limit or a virtual card number if your bank offers one. Muse is designed to ask before it spends, but your bank's limit is the backstop that does not depend on software behaving.
- Pick one primary surface
Web for desk work, the app for approvals on the move, WhatsApp if your business already lives there. Do not try to run it in three places during week one. Fragmented context produces fragmented results.
- Name it and set its voice
Muse lets you name the agent, pick an avatar and configure how it communicates. Name it something you will not confuse with a person in your inbox. Set the communication style to brief and direct. You want status updates, not conversation.
- Do not connect anything yet
Finish this phase with zero connected apps. Phase 2 is where connections happen, one tier at a time.
Connections, one tier at a time
Meta built the connection flow so you add apps one at a time. Use that. The permission ladder below is the core of this SOP. Every tier you add increases what Muse can do and what it can get wrong.
| Tier | Permission type | Typical connections | Gate to move on |
|---|---|---|---|
| Tier 1: Read | Muse can see, summarize and draft, but nothing leaves | Calendar, email inbox (read), notes | Three days of accurate daily summaries with no misread priorities |
| Tier 2: Write | Muse can send, book, schedule and reply, each with your approval | Email send, calendar write, reservations, reminders | Ten approved actions in a row you would have done the same way |
| Tier 3: Spend | Muse can purchase, cancel or pay, each with your approval | Link by Stripe checkout, subscription management, bill negotiation | A written spend ceiling in your operating brief and a card limit at the bank |
| Tier 4: Browser / API | Muse operates a service through its own browser or a credential you supply | Any tool without a built-in connector | Approve every single action until you have seen the pattern hold for two weeks |
- Connect one Tier 1 source. Ask Muse for a daily summary for three days. Check it against reality.
- Only when Tier 1 is clean, add one Tier 2 tool. Route every write action through approval. Do not batch-approve.
- Tier 3 waits until you have completed Phase 3 and set a spend ceiling.
- Tier 4 is the exception, not the default. Browser-based access is the highest-risk mode of any agent. Use it only for tools that matter enough to justify the review time.
The operating brief
Muse learns from your conversations and will start making unprompted suggestions. What it learns depends on what you tell it. An operating brief is the standing set of instructions you would give a new assistant on day one. Write it once, paste it as your first real conversation, and refer back to it when Muse drifts.
Copy the template below and fill in the brackets. Keep it under 400 words. Longer briefs get ignored, by agents and by people.
OPERATING BRIEF FOR [AGENT NAME] WHO I AM I am [name], [role] at [business]. We [one sentence on what the business does] for [who the clients are]. I work [days/hours] in [time zone]. WHAT YOU HANDLE - Inbox: sort by [client / vendor / admin / noise]. Draft replies to admin and vendor mail. Never send client replies without my approval. - Calendar: protect [deep work block] on [days]. Offer [meeting windows] only. Confirm and reschedule with approval. - Follow-ups: remind me [X] days after any unanswered email I sent. - Subscriptions and expenses: track, flag anything over [$ amount], flag anything unused for 60 days. - Logistics: book travel and reservations within the limits below. WHAT YOU NEVER DO WITHOUT ASKING - Spend more than [$ amount] on any single action. - Spend more than [$ amount] in a week, total. - Send anything to a client or prospect. - Cancel any service, subscription or booking. - Share my data or files with any third party. HOW TO REPORT One update per [morning / evening], in this order: actions needing approval, actions completed, anything you are unsure about. Bullets, no preamble. IF YOU ARE UNSURE Ask. A question costs me 10 seconds. A wrong action costs me an hour.
Every rejected approval in Phase 5 is a line that belongs in this brief. Update it weekly. A brief that never changes is a brief that was never tested.
Delegation by task class
Muse's launch capabilities map well onto the admin load of a solo operator. The table sorts them by what to hand off now, what to hand off after trust is earned, and what stays with you.
| Hand off now | Hand off after Tier 2 trust | Keep |
|---|---|---|
| Inbox triage and daily summary | Sending vendor and admin replies | Client-facing replies, proposals, quotes |
| Reminders and follow-up nudges | Scheduling and rescheduling meetings | Pricing decisions and negotiation |
| Subscription and expense audit | Cancelling unused subscriptions | Contract language and signatures |
| Form filling with data you supply | Travel and reservation booking | Anything touching client data without consent |
| Turning a saved reel or post into a checklist | Bill negotiation with your ceiling set | Content strategy, ad decisions, brand voice |
Vague requests produce vague work. Give Muse the same structure you would give a contractor: context, the exact outcome, the boundary, and how to report. Three starters:
INBOX TRIAGE Each morning at [time], read my inbox. Sort into: client, prospect, vendor, admin, noise. Draft replies for vendor and admin. List client and prospect emails with a one-line summary each. Do not send anything. Report as bullets. FOLLOW-UP TRACKER Track every email I send that asks a question or requests a decision. If no reply after [3] business days, remind me with the original subject line and who it went to. Do not follow up on my behalf. SUBSCRIPTION AUDIT Review my connected accounts for recurring charges. List each with amount, billing date and last use. Flag anything unused for 60 days or over [$ amount]. Do not cancel anything. I will approve cancellations individually.
Approval discipline
Approval before spend, send or share is the single design choice that makes Muse usable for a business. It is also the first thing people stop reading once the requests become routine. The moment you start tapping approve without reading is the moment the agent starts running you.
- Read the full request, every time
Recipient, amount, content, timing. If the request is a draft email, read it as the person receiving it.
- Apply one test
Would I have done exactly this myself, right now? If not, reject. Do not edit and approve in the moment. Reject, note why, and let the next attempt come through the brief.
- Log every rejection
Keep a running note: date, task, what was wrong. This is the raw material for your operating brief updates and your weekly scorecard.
- Never batch-approve during week one
Approving ten actions at once is how a wrong assumption gets executed ten times.
- Set an approval window
Handle approvals at two fixed times a day rather than as they arrive. Muse keeps working in the background; your attention does not need to.
Usage and cost control
Muse is free with a usage cap, shown as a percentage meter inside the app. It warns you before free usage runs out and offers the paid plans. Meta expects most people to stay on the free tier. Operators should treat that as the default and make the upgrade earn its place.
The upgrade mathDo not upgrade because the meter ran out. Upgrade because the meter ran out on work that was worth more than the plan. Use the scorecard in Phase 7 to get the number, then run this:
Weekly minutes saved (from scorecard) x 4 = monthly minutes saved Monthly minutes saved / 60 x your hourly rate = monthly value Power ($20/mo) is justified when monthly value is at least $100. Maximum ($100/mo) is justified when monthly value is at least $500. Anything under a 5x return is a subscription you will forget about. Muse will find it in your next audit, which is a strange kind of irony.
If the meter runs out on low-value tasks (chit-chat, re-asking the same thing, tasks that should have been rules), the fix is a tighter brief, not a bigger plan.
Security hygiene and the weekly scorecard
Meta's stated architecture: Muse runs inside a dedicated secure virtual machine with its own browser, a separate Sentinel agent monitors it at the system level, Muse cannot see your passwords or payment methods, and conversations are not shared with Meta's ads systems. Those are Meta's claims, published two days ago, and they have not yet been through independent security review. Meta's own regulatory history is public. This SOP does not tell you whether to trust it. It tells you how to run it so the question matters less.
Monthly security checklist
- Review every connected app. Revoke anything you have not used in 30 days.
- Confirm the spend ceiling in your brief still matches the card limit at your bank.
- Check whether any connected account now contains client data it did not contain last month.
- Re-read Meta's Muse terms for changes. Set a calendar reminder; do not rely on remembering.
- When 1Password integration ships, move any credential you gave Muse for a Tier 4 connection into it, and rotate the old one.
- Keep Muse off any account that is legally yours to protect: client CRMs, client ad accounts, anything under an NDA.
Four numbers. Track them in a note, a sheet, or ask Muse to keep the log itself (that is a Tier 1 task). If you cannot fill in all four, you are not running a system, you are running an experiment.
Clean completions divided by handed off is your trust rate. Under 70 percent: tighten the brief and stay at the current tier. Over 90 percent for two weeks: you have earned the next tier.
The 30-day rollout
| Week | Tier | Focus | Exit criteria |
|---|---|---|---|
| Week 1 | Read only | Fit check, setup, one connection, operating brief written, daily summaries | Three accurate summaries in a row. Brief pasted and acknowledged. |
| Week 2 | Write | One write-capable connection. Reminders, follow-up tracker, drafted vendor replies | Ten consecutive approvals with no edits. Rejection log started. |
| Week 3 | Write, expanded | Scheduling, rescheduling, one booking task. First scorecard. | Trust rate above 80 percent. Brief updated from rejections. |
| Week 4 | Spend, limited | Subscription audit, one supervised cancellation, one purchase under ceiling | Upgrade math run. Decision made on paid plan with a number behind it. |
At day 30 you will have one of two outcomes. Either Muse is saving you a measurable block of admin time each week, in which case you now have that time to put into the marketing system that gets you clients. Or it is not, in which case you have a clean rejection log that tells you exactly why, and you can turn it off without having exposed anything you cared about. Both are wins. Only the first one grows the business.
Questions operators ask
What is Muse by Meta?
Muse is Meta's personal AI agent, launched September 8, 2026 for users in the United States. It connects to your email, calendar, payments and other apps and performs tasks on your behalf, such as sending emails, booking travel, filling forms and making purchases, with your approval before it spends, sends or shares anything. It is powered by Meta's Muse Spark model and available at muse.ai, on iOS and Android, and inside WhatsApp.
Is Muse by Meta free?
Muse is free to use with a usage limit. A payment card is required at signup. Two paid plans were announced at launch: Power at $20 per month and Maximum at $100 per month, which increase how much work you can hand off. The app shows a usage meter and warns you before free usage runs out.
What is the difference between Muse, Muse Spark, Muse Image and Muse Code?
Muse is the consumer agent that does tasks for you. Muse Spark is the underlying AI model that powers it. Muse Image and Muse Video are Meta's media generation models available through the Meta AI app and Instagram. Muse Code is a developer coding agent. This SOP covers Muse, the agent.
Does Muse see my passwords or payment details?
Meta states that Muse runs inside a dedicated secure virtual machine with its own browser, that a separate Sentinel agent monitors it, and that Muse does not have visibility into passwords or payment methods. Meta also states that Muse conversations are not shared with its ads systems. These are Meta's claims and have not yet been independently audited, so this SOP treats every connection as a decision you own.
Can I use Muse for my business?
Muse is positioned as a personal agent, but most of its capabilities map directly onto the admin load of a solo operator or small business: inbox triage, scheduling, follow-ups, subscription audits, expense tracking, form filling and logistics. The limit is client data. Do not connect client systems or share client information with Muse without written consent and a review of your contracts.
What should I never let Muse do?
Do not let Muse send client-facing proposals, quotes or contract language without your edit. Do not let it spend above a limit you set in your operating brief. Do not connect systems that hold client personal data without consent. Do not approve requests you have not read in full.
How do I connect an app that Muse does not have a built-in connector for?
If the service offers a public API, Muse can set up a connection using credentials you provide. If there is no API, Muse can operate the service through its own browser. Treat browser-based access as the highest-risk tier and approve each action individually until you trust the pattern.
Does Muse keep working when I close the app?
Yes. Muse continues working on tasks after you leave the app and learns from your conversations to make unprompted suggestions over time. This is why the operating brief and approval rules matter: the agent is acting when you are not watching.
Is Muse a marketing tool?
No. Muse is an operations and admin agent. It is not Meta Ads Manager, it does not run campaigns, and it does not replace a content system. Its value to an operator is time recovered from admin, which you then reinvest in the marketing system that actually produces clients.
Sources and currency
Product facts on this page are drawn from Meta's Muse announcement and product pages and launch coverage dated September 8, 2026. Muse is two days old at publication. Pricing, connectors and security claims will change. This page is reviewed monthly; the date at the top of the page is the last review.