Meta launched Muse AI on September 8, 2026, and the company is keen to stress it isn’t just another chatbot. Meta describes Muse as a personal AI agent: software that completes tasks on your behalf rather than answering questions in a chat window. It runs on Meta’s Muse Spark model, inside what’s called a ‘Muse Secure VM’ — a dedicated cloud computer with its own browser. This guide covers what Muse AI does, how it differs from chatbots, what it costs, where it’s available, and how to get access in practice.
What is Muse AI by Meta?
Muse is a personal AI agent — and that distinction matters. A chatbot responds to prompts with text. An agent, as Meta defines it, takes actions: shopping online, booking travel, sending emails, making payments, filling in forms, booking appointments. It all runs through the Muse Secure VM, a dedicated cloud computer with its own browser. So instead of generating text replies, Muse can actually interact with websites and services on your behalf. That shift, from answering to acting, is what separates Muse from the chatbots you’ve used before.
Close the app and Muse keeps working. Ask it to complete a task, put your phone away, and it carries on in the background on its Secure VM. When it needs your input (before sending an email or completing a purchase, say), it asks for approval first. That approval step is what keeps you in control of the sensitive stuff.
How Muse differs from a chatbot
The key difference is agency. A traditional chatbot, including Meta’s own Meta AI assistant, waits for your message and replies with information. Muse is built to do things. Meta’s example is travel booking: rather than suggesting flights, Muse goes through the booking process itself. Rather than drafting an email for you to send, it sends it (with your approval). The Secure VM makes that possible: a persistent cloud environment where Muse works independently, using its own browser to get around the web.
Then there’s continuity. A chatbot session ends when you leave; Muse keeps going. Running on its Secure VM, it can carry a multi-step task: comparing prices across stores, filling in a long form, watching for a booking confirmation — without you looking over its shoulder. You get a notification when it needs approval or when the job’s done.
Muse AI features: the complete list
Here’s what Muse AI can do at launch, based on Meta’s official documentation.

Personal AI agent (not just a chatbot)
One agent, not a collection of bots. Muse handles your tasks through a main conversation, with side chats, goals, and subagents keeping different threads tidy. Side chats handle parallel threads (a trip, a purchase, an email draft) without cluttering your main conversation. The pitch is that everything lives in one place instead of being scattered across multiple assistants.
Muse Secure VM: a dedicated cloud computer
Every Muse user gets a Secure VM: a dedicated cloud computer with its own browser, and this is where Muse does its work. Because it’s a full browser environment, it can handle websites that need logins, forms, and multi-step processes. The VM is dedicated per user, so your agent’s workspace isn’t shared with anyone else.
Task completion: shopping, booking, email, and more
Meta’s task list covers online shopping, travel booking, emailing, payments, filling in forms, and booking appointments. In practice: ask it to find and order a product, book a restaurant table, or draft and send an email. For the sensitive actions: sending email and making purchases are the examples Meta gives — it asks your approval before proceeding.
Works after you close the app
Muse keeps working on your Secure VM after you close the app on your phone. Handy for slow-burn tasks: waiting for a price drop, monitoring availability, grinding through a long multi-step process. You’ll get a notification when it needs your approval or when it’s finished. Effectively a personal assistant who works while you’re in meetings.
Approval before sensitive actions
Approval before sensitive actions is a big part of Meta’s pitch. The examples given: sending emails and making purchases. In other words, Muse won’t send a message or spend money without you confirming. You stay in control of the consequential decisions while it handles the routine steps.
Real-time avatar for video chat
There’s also a real-time avatar for video chat: instead of just text or voice, you interact with a visual version of Muse. Meta says it’s meant to make conversations feel more natural, though the core of the product is still the agent getting tasks done. Treat the avatar as a nice-to-have rather than the reason to sign up.
Connectors: Shopify, QuickBooks, Stripe, and more
Connectors let Muse work inside third-party services. Meta’s list includes Shopify, QuickBooks, Stripe, Canva, Asana, Notion, Slack, Zoom, Figma, Dropbox, Walmart, Best Buy, Sephora, Expedia, Instacart, GitHub, and Spotify. The point of a connector is action, not information: creating a task in Asana, checking a document in Notion, ordering from Walmart, rather than just telling you about those services. The practical effect is less tab-switching: Muse moves between your tools so you don’t have to.
Muse for Small Business
On September 29, 2026, Meta announced ‘Muse for Small Business’, a version aimed at small business use cases. Full details on how it differs from the consumer version haven’t been published, but the announcement signals that Meta sees small businesses (booking, ordering, customer comms) as a key audience for AI agents. For a one-person business, an agent that completes admin tasks in the background is easy to see the appeal of.
Muse AI pricing
Muse is free for most uses, with paid plans from $20 a month. Meta hasn’t published a detailed free-vs-paid breakdown, but $20/month sits in line with other premium AI subscriptions. If you’re eyeing the paid plan, check Meta’s official pricing page for the current comparison — details will shift.
Availability: where you can use Muse
At launch: US and Canada only, on iOS, Android, the web at muse.ai, and WhatsApp. Meta AI glasses support and Mac desktop control are coming, which would eventually let you talk to Muse through smart glasses or have it operate your Mac. No timeline yet for the UK or wider international rollout.
According to Reuters, citing a Sensor Tower estimate, Muse was downloaded approximately 2.8 million times in its first two weeks. This suggests strong initial interest, though download numbers do not necessarily reflect active long-term usage.
How to get access to Muse AI
In the US or Canada, you get Muse by downloading the iOS or Android app, visiting muse.ai, or messaging it on WhatsApp. You’ll need a Meta account to sign in. Once set up, Muse provisions your Secure VM, the dedicated cloud computer your agent runs on. Setup is described as straightforward, though you’ll need to grant permissions for the connectors you want: email, shopping accounts, and the like. Expect to spend a few minutes authorising those before the agent is genuinely useful; an agent without access to your tools is just a chatbot with ambition.
Pros and limitations
Here’s our honest assessment, based on Meta’s official documentation.
Potential advantages
The agent model is a genuine step beyond chatbots: Muse completes tasks rather than just answering questions. The Secure VM solves the persistence problem: tasks continue without you watching. The approval step for sensitive actions is a sensible safeguard. And the connector range, from Shopify to Spotify, suggests Meta is serious about usefulness across work and personal life. The free tier makes it easy to try before spending anything, and $20 a month is in line with what other premium AI subscriptions charge for comparable features.
Limitations to be aware of
First, availability: US and Canada only at launch, no UK timeline. Second, trust. An agent that acts on your behalf gets access to your email, shopping accounts, and payment methods. Approval-before-sensitive-actions helps, but still review every permission you grant. Third, the $20/month tier’s exact benefits aren’t fully detailed. And finally, real-world reliability (how Muse handles complex, multi-step tasks without errors) will only show with broader use. Remember, this guide is built from Meta’s documentation, not hands-on testing. Where Meta hasn’t said something, we say so rather than filling in the gaps.
Frequently asked questions
Is Muse AI free?
Free for most uses; paid plans start from $20 a month. Meta hasn’t published a full free-vs-paid comparison, so check the official pricing page for details.
Is Muse AI available in the UK?
Not at launch — US and Canada only. Meta AI glasses support and Mac desktop control are coming, but there’s no timeline for the UK yet.
How is Muse different from Meta AI?
Meta AI is a chatbot assistant that answers questions. Muse is a personal AI agent that completes tasks (shopping, booking, emailing, and more) using a dedicated Secure VM with its own browser. The difference is agency: Muse takes actions, not just provides information.
Does Muse work without my approval?
Yes — it asks for your approval before sensitive actions, specifically sending emails and making purchases. Routine steps within a task it handles on its own; the sensitive decisions stay with you.
What is the Muse Secure VM?
It’s the dedicated cloud computer assigned to each Muse user, with its own browser that Muse uses to interact with websites and services on your behalf. Because it’s persistent, Muse keeps working even after you close the app on your phone.
What is Muse for Small Business?
Announced September 29, 2026: a version of Muse tailored for small businesses. Full details aren’t published yet, but routine business tasks (booking, ordering, customer communication) are the focus.
If you are exploring how AI agents could work for your business, our team can help you understand the options.


