The short version. Muse is Meta's personal AI agent, launched on September 8, 2026. What sets it apart from a chatbot is simple: Muse doesn't just answer questions — it carries out multi-step tasks on its own. It sorts your inbox, books trips, fills in forms, compares prices. And here's the part that matters most for a business owner: the agent can run your DMs in your tone of voice. I connected it to my Instagram and tested it — it answers the way I do, and I like the result.
Muse is officially available in the US and Canada. Signing up takes 15–30 minutes: register with an email address, confirm you're 18+ with a card on file, pick a name for your assistant, and connect your email and calendar so the agent can sort mail, draft replies, and manage scheduling.
One caveat if you're signing up from outside the US: there's a VPN nuance that kills access even after a successful registration — it's where half of all newcomers get stuck. I've broken it down in the free PDF at the end of this article, in the section "The geo trap."
What makes Muse different from a chatbot
A regular chatbot lives in a chat window: you ask, it answers. Muse lives on its own isolated virtual machine — the agent's "cloud computer." It keeps working in the background even when the app is closed: in the morning you get a finished result, not a to-do list. A separate safety module oversees every outward-facing action. Purchases go through single-use virtual cards — the agent never sees your real card number.
The launch was loud: 2.8 million installs in the first 12 days, hitting #1 among free apps in the US App Store and overtaking ChatGPT. For comparison, ChatGPT managed 1.3 million in its first 12 days of mobile launch back in 2023. The numbers come from analytics firm Apptopia, via TechCrunch.
Meta's head of AI, Alexandr Wang, pitches the product directly at business owners: in his words, you can have the agent run your small business — sorting email, managing the calendar, working with customers (Inc.com).
Your DMs — its job
Let me tell you how this looks in practice, because I've already walked this path myself.
I connected Muse to my Instagram DMs. First, the agent studied how I communicate: my tone of voice, my manner, my phrasing. Then I gave it a simple job — answer my customers. The result satisfied me: the agent sounds like me. Messages no longer sit unanswered for hours while I'm busy working — the customer gets a reply right away, in my tone.
And this isn't a one-person experiment. On June 3, 2026, Meta rolled out Meta Business Agent — an agent living inside the messengers where customers already talk to businesses: WhatsApp Business, Instagram, Messenger. According to Meta, over a million businesses already use it: the agent answers questions, recommends products from the catalog, books appointments, qualifies leads — speaking in the brand's tone of voice, which the business sets up during onboarding (Meta).
A typical real-life story: a nail salon owner kept losing clients — while she served one customer, DMs sat unanswered for hours, and people booked with competitors. With the agent, that "lead leak" closed: it replies instantly while the owner is busy doing the actual work (Tech in Asia).
What you can realistically delegate right now: inbox triage and draft replies, calendar management and client bookings, answers to routine DM questions, quoting your price list and terms.
How to sign up
The flow is straightforward. Register with an email address — use a dedicated one rather than tying the account to your personal social profiles. You'll need just an address, a name, a date of birth, and a password of 8+ characters.
Then confirm your age and add a card. Note: the card is required even on the free tier — it's part of registration and proof you're 18+, not a purchase. After that, pick a name for your assistant — what you'll call it. And connect the services it needs: email, calendar. That's how the agent gets to sort your mail, write drafts, and run your calendar.
Two classic mistakes that break everything: registering through a personal Instagram account — your personal and business context get mixed up; and going through signup on shaky geo conditions if you're outside the launch countries.
What it costs
Muse has three tiers. Free — up to 100 million tokens per week. A token is the unit of AI work: the harder the agent's task, the more tokens it "eats." Then Power at $20/month (up to 500 million tokens a week) and Maximum at $100 (up to 3 billion). Pricing data via Tech Insider.
One nuance worth knowing upfront: even the free tier requires a bank card on file and confirmation that you're 18+ and located in the US. That's why the guide above mentions a separate email and — for those outside the US — a VPN switched on before you start.
Gizmodo's journalists tested in practice whether anyone can actually burn through 100 million tokens in a week, trying as hard as they could. Their verdict — and what it means for your work — is in the PDF, in the section "Muse is not a chatbot."
Honestly about the risks
Let me be straight with you so there are no surprises later.
First, the agent is your tool, and the decisions stay yours. Keep it in "shadow mode" for the first week: let it prepare drafts only, and approve every action yourself. Your edits become its training data. Autopilot is for low-risk tasks only; anything customer-facing stays under your approval.
Second, double-check what matters and never trust it blindly. Research shows people forgive the same mistake in a human but not in an algorithm — and learning a message was written by AI makes people judge the sender more harshly (Phys.org). So honesty here is your ally, not your enemy.
Third, if your customers are in Europe — note Article 50 of the EU AI Act: since August 2, 2026, the person you're talking to must know they're talking to AI. Fines run up to €15 million (Regulation AI). Set up a disclosure that an assistant is replying — it's honest and it settles the question.
When it's not for you
Muse isn't for everyone, and saying so matters.
If most of your customers are in Europe and you're not ready to disclose that an AI is replying — wait. Transparency requirements already apply there, and the fines are serious.
If it matters to you that every reply to a customer is written personally by you, and the very idea of delegating correspondence feels wrong — don't force it. A tool should remove load, not add anxiety.
If you're not ready to spend the first week reviewing drafts and correcting the agent — better not to start. An unsupervised agent will sound like "an average robot from the internet," not like you. The tone magic only works when you invest in the setup.
Get the PDF: the guide + a 1,000,000,000-token promo code
I've packed everything concrete into a free PDF guide. Inside:
- a 5-step sign-up walkthrough with common mistakes broken down — print it and follow along;
- a promo code for 1,000,000,000 tokens — inside the PDF only, not in this article or the funnel. Just 60 activations total; codes can run out fast, so act right after signing up;
- the "From a billion to thirty billion" block: how the referral mechanic gets you to 30,000,000,000 tokens — invite a friend, you both get tokens.
To get the PDF, leave your name and contact in the form below — the file opens on the thank-you page right after you submit.