Meta Muse: the agent that keeps working after you close the app.
On September 8, 2026 Meta introduced Muse — not a new model, but a personal agent you talk to like a person, in a new Muse app or straight inside WhatsApp. Give it a goal and it plans, opens a browser, fills in forms, negotiates, and comes back only when it needs your yes. Under the hood it runs on Muse Spark, Meta's most capable model. Here is what it actually does, what it is allowed to touch, and what Meta hasn't said yet.
Assistant vs. agent — the one distinction that matters
Meta AI, the thing already in your WhatsApp and Instagram, answers. You ask, it replies, the turn ends. Muse is built for the other shape of work: you hand over a goal and walk away. Meta's own examples are deliberately mundane — sell a car for more than the first offer, get a bill lowered, keep a training plan updated as the rest of your week shifts, turn a recipe reel you saved on Instagram into a grocery list, plan a dinner party and remember which friend can't eat shellfish before the invites go out.
Two behaviours make that possible. First, it keeps going after you close the app: long tasks continue in the background and Muse messages you when something changed or when it needs approval. Second, it remembers — details you mentioned once, preferences, people — and uses them to suggest things unprompted. You can tell it to "forget" specific things at any time.
The announcement, decoded
Connecting a service is not all-or-nothing. For email, for example, you choose whether Muse can only read your mail or can also send on your behalf, and you can dial that back or disconnect the service entirely whenever you like. Meta also states that Muse conversations and VM data are not shared with its ad systems, and that you can opt out of your interactions being used to train Meta's models. Our advice on day one: connect one low-stakes service, watch the audit trail for a week, then decide what else it earns.
What's new since launch — Meta Connect 2026
Meta used its Connect 2026 event to show where Muse is headed next. Most of this is not live yet, so treat it as roadmap rather than something you can try today:
- A talking avatar. A new Muse Realtime Avatar model gives the agent a face, body and voice for live video conversations, instead of just text and audio. Now rolling out in the US — see below. See our Muse Spark lesson for where this fits in Meta's model lineup.
- Onto your desktop. A Mac "computer use" mode will let Muse operate apps on your behalf and keep working while you step away — the same background-tasking idea in the app, extended to your whole desktop.
- Its own email address. Muse is getting a dedicated email address so you can add it to a thread or forward mail for it to handle. Meta says "soon," with no date yet.
- On your glasses, by voice. A wake word will bring Muse to Meta's smart glasses for things like guided workouts, meal logging, appointment booking and shopping — Meta says "coming months." See our glasses setup lesson for what glasses can already do today.
- A lot more places to shop and connect. New retail and payment connectors include Best Buy, Gap, Sephora, Walmart, Wayfair, Shopify (full catalog), Expedia and PayPal, plus developer tools like GitHub, Notion and Granola. Meta says it received over 1,500 applications from outside developers for custom connectors in the announcement's first week.
New in November 2026: voice, avatars and spaces
Two of those Connect promises have now shipped, per Meta's newsroom:
- Talk to it. Real-time voice conversations and an expressive avatar are rolling out in the Muse app on mobile, web and Mac — tap the phone icon in the top-right corner. You choose voice only, avatar beside the chat, or full screen, and you can interrupt mid-answer. There are hundreds of voices to pick from. It is US-only to start; voice on Meta's glasses and Muse Charm is still "coming soon." (Meta, November 5, 2026.)
- Build your own mini-apps. Spaces let you describe something — a holiday gift tracker, a planning board, a daily word game — and Muse builds it inside the app. You can keep refining it by asking, pin it to your sidebar or main tabs, hide it, or schedule when it appears. Meta says spaces are available everywhere Muse is available. (Meta, November 6, 2026.)
Spaces are most useful when they pull together things Muse can already see — but some of Meta's own examples draw on Instagram saves, Marketplace messages and school emails. Every space that reads a connected account is another place that data shows up, so build your first one from information you type in yourself, and only then point one at a connected service. Sources: voice and avatars, spaces.
Try it today — a 15-minute first run
If you are in the US, this is the sequence that teaches you the most while risking the least:
1. Install the Muse app (or open muse.ai) and set up nothing except the basics. 2. Give it a goal that needs no account access: "Find me three used-car listings under $12,000 within 30 miles that were posted this week, and tell me which one you'd negotiate on and why." Watch how it plans, browses and reports back. 3. Close the app deliberately. See whether it messages you when it is done. 4. Only then connect a single service — a calendar is a good first candidate — and set it to read-only. 5. Open the audit trail and read every step it took. That trail is the product; the chat is just the front door.
The honest part
Meta Muse — quick answers
What is Meta Muse?
Is Muse free?
Where can I get it?
Can Muse spend my money?
Does Muse see my passwords?
How is Muse different from Meta AI?
Is Muse actually the most popular AI agent?
Can Muse shop on Amazon?
The model underneath it: Muse Spark — Meta's agent model, explained. And if you want the safety side of agents before you connect anything, read what Meta AI sees, and how to limit it.