Meta Muse Availability: US Only, For Now

Meta Muse availability is limited to the United States at launch. Meta's newsroom announcement on 8 September 2026 says Muse "is rolling out in the US on iOS, Android, and muse.ai, and coming soon to AI glasses," with no second country named and no waitlist mentioned for anywhere else. If you are reading this from London or Toronto or Sydney, the app store link will not help you today.

Key Takeaways:

• Muse launched 8 September 2026, US only, on iOS, Android and the web at muse.ai
• Meta says it is "free for most of what people need," with paid subscription tiers above that
• Each user gets a Muse Secure VM, a dedicated cloud machine holding the agent, its browser and their connected credentials
• The model behind it, Muse Spark, is already on OpenRouter at $1.25 per million input tokens, so the model is reachable from outside the US even though the app is not
• AI glasses support and an encrypted Muse Confidential VM are both promised for later, with no dates attached

What Meta actually shipped on 8 September

Muse is a personal agent that acts rather than answers. Meta's framing is that you hand it a goal and it builds the plan, then works the plan: booking travel, filling in forms, negotiating on your behalf, opening a browser and clicking through it. You talk to it in the Muse app or inside WhatsApp, which is the distribution decision that matters most, because WhatsApp is how a couple of billion people already type to somebody.

The architectural claim is the interesting part. Every user gets a Muse Secure VM: a dedicated virtual machine in Meta's cloud that holds the agent, its own browser, and the credentials for any service you connect. A second agent called Sentinel runs on the same machine, separated at the system level, and nothing Muse does reaches the internet without Sentinel approving it. Meta says Muse never sees your passwords or payment methods, that connected credentials go into secure storage the agent can use but not read, and that you get an audit trail of what it has done and what it plans to do next. Payments run through Link built by Stripe, which issues a single-use card so your real card number stays hidden, and Meta claims Muse is the first agent covered by Link's purchase protections: cover for damaged or lost items, price drops, no-fee returns. Shop Pay and 1Password support are listed as coming.

Whether the security model holds up is a question for people with more access than a newsroom post gives anyone. What is not in doubt is the shape of the thing: Meta decided a personal agent needs its own computer, and built one per user.

Where Meta Muse availability actually stands right now

One country. Three surfaces: iOS, Android, and the browser at muse.ai. The product page carries Google Play and App Store badges and a "Try Muse" button, which tells you the gate is an app store region gate rather than a server-side country check, and app store gates move on Meta's schedule, not yours.

Meta named no other country and gave no date for one. It also did not publish a waitlist for other markets, so there is nothing to sign up to and nothing to check back on except the newsroom. Two other things are promised without dates: AI glasses support, and Muse Confidential VM, where the whole virtual machine including your conversations gets encrypted under a key only you hold, which Meta says arrives "later this year."

A note on the VPN question, since it is the first thing people try. Changing your apparent location does not change which storefront your Apple or Google account belongs to, and Meta Muse availability is enforced at that layer. Neither we nor Meta recommend building your daily workflow on a workaround that breaks the first time an account gets reviewed.

The model is available even though the app is not

Nobody in the launch coverage picked this up. Muse runs on Muse Spark, which Meta calls its most capable model, and Muse Spark is already listed on OpenRouter with a public price: $1.25 per million input tokens and $4.25 per million output on version 1.3, against a context window of 1,048,576 tokens, as of 9 September 2026. There is a cheaper Glimmer 30B in the same family at $0.30 and $1.10.

So the intelligence is not geofenced; the consumer app is, and that gap is the most useful thing to understand about Meta Muse availability right now. If what you wanted from Muse was the model quality, you can have that today from anywhere OpenRouter serves. What you cannot have is the product wrapped around it: the VM, the Sentinel approval layer, the WhatsApp entry point, the Stripe-issued single-use card. That distinction is worth being clear about before anyone spends a weekend trying to recreate Muse out of API calls, because the API gives you the easy half.

The controls Meta says you get, in its own words

Since the whole pitch rests on trust, the specific promises are worth listing rather than summarising, because each one is a thing you can later hold Meta to.

You choose which apps Muse connects to and how far each connection goes; for email, Meta says you decide whether it can only read or can also send on your behalf. You can change that access or disconnect a service at any point. Muse checks with you before sensitive actions, naming sending an email and making a purchase as the examples. It shows a full audit trail of what it has done and what it intends to do. It remembers things you mentioned once, and you can tell it to forget specific items. Conversations and the data inside your VM are not shared with Meta's ad systems, and you can opt out of your interactions training Meta's models.

That last pair is the one to watch, because opt-out and not-shared are both weaker guarantees than they sound, and neither has been audited by anyone outside Meta. TechCrunch's write-up framed the launch around exactly this question of whether consumers will extend that trust to this company in particular. The Confidential VM promised for later this year is the version where the question stops mattering, since a key only you hold makes the promise structural rather than contractual. It is not shipped.

What Meta Muse costs when you can get it

Meta's wording is that Muse is "free for most of what people need, with subscription plans for people who want to do more." No tier names, no prices, no credit allowances, no statement of what falls on the paid side. That is all the pricing information that exists in the announcement, and anything more specific you read today was inferred by somebody.

The free tier is a deliberate move against the $20 floor that most agent products sit on, and against the $200 ceiling that heavy users of coding assistants pay. Meta can afford to run agent workloads at a loss for reach in a way that a company selling agent-hours cannot, which is worth remembering when the comparison tables start appearing. It also means the free tier's real boundary will only become visible once enough people hit it and complain, probably a few weeks from now, and until then every number you see attached to Muse pricing is somebody's guess.

Why a personal agent needs its own computer

Strip away the branding and Meta has validated a design decision the rest of the category has been arguing about for two years. A personal agent cannot live on your phone, because your phone sleeps, moves between networks, and holds your entire life on a device that gets lost in taxis. It cannot live on your laptop for the same reasons plus one more: the laptop is where you download things, which makes it the worst place to keep an agent's credentials. That argument has been made repeatedly in the abstract, and it has been made expensively in practice by Claude subscribers whose tokens were drained after infostealer malware lifted their session cookies.

So the agent gets its own machine. That is what Zentor already is: a hosted cloud AI computer where your agents run on infrastructure that isn't the device in your pocket, and it is available now rather than after a national rollout. The overlap with what Meta described is not a coincidence; it is two teams reaching the same conclusion from the same constraint.

The second half of that constraint is time. Meta makes a point that Muse "keeps working after people close the app" and comes back when it needs approval, which only works if something stays awake. Zentor solves that with the same move: the machine stays up, so scheduled and long-running work lands whether or not your laptop lid is open. Our write-up on always-on AI agents covers why that turns out to be the hard requirement rather than the obvious one, and personal AI agents explained sets out the category Muse just joined.

Where the two genuinely differ is scope, and it would be dishonest to blur it. Muse is aimed at a consumer running errands, and it is one integrated product from one vendor. Zentor sits next to whatever you already use rather than replacing it, which is the right shape when the assistant you depend on is a work tool you are not going to give up. If you are outside the US, the practical difference today is simply that one of these you can open.

FAQ

Is Meta Muse available in the UK?

No. Meta's announcement names the United States only, on iOS, Android and muse.ai. No other country was named, no date was given for expansion, and there is no waitlist to join for other markets.

Is Meta Muse free?

Partly. Meta says Muse is "free for most of what people need, with subscription plans for people who want to do more." Neither the prices nor the limits of the free tier have been published.

What model does Meta Muse use?

Muse Spark, which Meta describes as its most capable model to date. The Muse Spark family is separately available through OpenRouter, where version 1.3 is priced at $1.25 per million input tokens with a context window of just over one million tokens as of 9 September 2026.

Can I use a VPN to get Meta Muse?

App store region gating follows the country on your Apple or Google account rather than your IP address, so a VPN alone will not do it. Treat any workaround as temporary.

What is Muse Secure VM?

A dedicated virtual machine Meta runs per user, holding the agent, its browser, and the credentials for services you connect. A separate Sentinel agent on the same machine has to approve anything Muse sends to the internet.

Meta Muse availability is a distribution question, not a technology one

Nothing about Muse is technically confined to one country; the model is on a public API right now, priced in dollars, callable from anywhere. What is confined is a consumer app that Meta wants to roll out at its own pace, through app stores that enforce nationality. So the honest answer for everyone outside the United States is that there is no trick and no date, only the newsroom to watch. In the meantime, the idea Muse is built on, that an agent deserves a computer of its own that stays awake, is not something you have to wait for.