Updated September 26, 2026. This is a source-based explainer, not a hands-on review. Availability, rankings, and prices can change.
Muse AI has a compelling pitch: hand over an everyday task and let an AI work through it while you get on with your day. The app, officially named Muse from Meta, ranked No. 1 on Apple's US free iPhone apps chart when checked on September 26, 2026.
That makes it an interesting entry in the personal agent conversation. But before downloading, there are practical questions: what does it actually control, where can you use it, and how much work is free?
What is Muse AI?
Muse is Meta's personal AI agent: software designed to take actions across online services, maintain context, and pursue tasks over time. Meta announced Muse on September 8, 2026, powered by its Muse Spark model.
The distinction between those names matters. Muse is the consumer product; Muse Spark is the model powering it. Finding a model's API pricing does not tell you what the Muse app costs.
For example, the intended workflow goes beyond asking for travel suggestions. You can delegate the research and booking process, then review the steps that require your approval. Meta says longer tasks can continue after you close the app, with Muse returning when there is an update or a decision to make.
Think of it as a place to hand over work, with a conversation attached to that work.
Why is Muse attracting attention?
The chart result is concrete, but its scope matters: No. 1 among free iPhone apps in the United States on the date checked. It does not establish a worldwide download lead or prove the app reached first place on launch day. Apple's Canadian free-app chart also placed Muse first when checked.
Our reading of the appeal is straightforward. “Compare these options and help finish the job” is easier to connect to daily life than an abstract promise of better AI reasoning. A familiar messaging interface gives people a simple way to try delegation.
The important question after the initial curiosity is whether completed work saves more time than reviewing and correcting it takes. Download rankings cannot answer that.
How did Muse come about?
Meta's technical account of building Muse says the team had been developing and using it internally since early 2026. The work included training for tool use, extended instructions, and coordination between agents, alongside engineering protections for unattended operation.
The product story is equally revealing. In How We Designed Muse, Mona Sarantakos and Christine Awad describe a motivation rooted in balancing demanding work and family life. The design centers on an ongoing conversation, persistent memory, and proactive help.
Users can shape the agent's name, avatar, and communication style. Side chats organize separate topics, while interactive outputs called Artifacts can turn a response into something more useful, such as an itinerary or dashboard. The team also designed controls for adjusting how proactive Muse should be.
This combination explains the “personal” in personal agent: continuity and context are part of the product experience.
Is Muse a personal agent or a cloud agent?
Both descriptions can be useful, provided they mean different things. Personal agent describes a convenient way to access the service. Meta's Muse product page describes the underlying workspace as a persistent, isolated Linux computer in the cloud, with a browser shared by the user and agent.
That does not establish unrestricted control over every app on your phone. Muse works through its browser, supported connections, and the permissions you grant.
Meta calls those service connections Connectors. Its connector guide explains that connecting a calendar or email service gives Muse access relevant to the task. Some connectors support read-only access; others allow actions with approval controls.
A useful setup detail: Facebook, Instagram, and Threads can connect automatically when their accounts share the same Accounts Center. Review that grouping alongside the connections you add manually.
Muse is also broader than a mobile app. Meta's official download page offers iPhone, Android, and Mac access, and describes continuity with WhatsApp. The web entry point is muse.ai. Device options should not be confused with worldwide service availability.
Muse AI use cases: what can it do?
Meta's official app listing describes scheduling, subscription reviews, shopping, fitness planning, and larger tasks such as trip planning. The prompts below are suggested starting points, not results we obtained in testing.
Use case | Example task to delegate | What to review |
|---|---|---|
Calendar and email | “Find next week's scheduling conflicts and prepare replies.” | Times, recipients, proposed changes |
Subscriptions | “Identify recurring subscriptions I may no longer use.” | Charges, renewal dates, cancellation consequences |
Shopping | “Compare these headphones and monitor their prices.” | Exact model, seller, final purchase total |
Travel | “Build a weekend itinerary within my budget.” | Dates, booking conditions, transfer time |
Fitness routines | “Organize a weekly workout plan around my schedule.” | Whether the plan fits your actual circumstances |
Another useful category is producing something you can keep. Meta's getting-started guide documents reading, editing, and generating artifacts including PDFs, spreadsheets, and documents.
For a first task, combine a clear outcome with a boundary: “Use my calendar to draft a plan for tomorrow. Show it to me before changing any events.” This makes success easier to check than a broad request to organize your life.
Where is Muse AI available?
The US launch is confirmed, and Canadian iPhone distribution is also evidenced by Apple's current storefront chart. A complete current country-by-country service list was not established from the official sources reviewed.
Market | What the official evidence supports |
|---|---|
United States | Named in Meta's launch rollout; US iPhone listing and chart are live |
Canada | Muse appears in Apple's Canadian free iPhone chart; this establishes storefront distribution, not identical feature access for every account |
UK and EU markets | Broad service availability was not confirmed in the official materials reviewed |
Mainland China, Hong Kong, Taiwan, India, Japan, Singapore, Australia, and other markets | This review did not establish official service availability; check the local store and Muse's account onboarding |
The last two rows are unconfirmed availability, not an official blacklist. Meta's Muse Help Centre explicitly says the app and website are not available everywhere, and features can vary by area even where the service is accessible.
For readers who cannot download Muse, the practical distinction is between three checks: local app distribution, account eligibility, and access to a specific feature. Reaching the website alone does not settle all three. Nor does a translated help page establish a launch in that language's home market.
Muse AI pricing: do you need a subscription?
No paid subscription is required for the limited free tier. Meta's official subscription guide lists the following plans, checked September 26, 2026:
Plan | Published monthly price | Usage allowance |
|---|---|---|
Free | $0 | Limited; the cited page does not quantify the allowance |
Power | $20 | 500 million Muse tokens per week |
Maximum | $100 | 3 billion Muse tokens per week |
If free usage runs out, you can wait for it to refresh or upgrade. Paid subscriptions renew monthly until canceled. Meta notes that benefits and availability can vary by account and region, so check the offer displayed before subscribing rather than treating the table as a universal local-currency quote.
The published unit is Muse tokens. It should not be translated into a guaranteed number of completed bookings, messages, or research tasks. Our suggestion is to start free and judge a subscription against a workflow you already find useful.
What should you know about privacy and control?
Muse's cloud workspace makes access control central to the experience. Meta's technical explanation describes a separate Sentinel system that mediates actions outside the agent's environment, plus protected credential storage. These are architectural safeguards, not proof that every permitted action will be correct.
Two settings deserve particular attention:
Model training: Meta's data controls documentation says the option to use interactions to improve its models starts enabled. You can turn it off under Settings → Data controls → Help improve our AI models.
Stored memory: The same guide says deleting a message can leave information learned from it in memory. Memory management, forgetting, and resetting are separate controls.
Disconnecting a service also stops future exchange without necessarily removing information already retained, according to the connector guide.
Meta announced a stronger Confidential VM protection for later in 2026. The launch announcement should not be read as confirmation that this future protection is already enabled for your account.
Muse AI benchmarks: what does No. 1 actually prove?
An App Store ranking measures chart position, not task accuracy. This review did not establish an official app-level benchmark that predicts how often Muse will complete your particular workflow successfully.
For a personal agent, a useful evaluation is concrete: did it finish the task, respect the budget, preserve the right context, and ask at the right moment? Count correction time as well as execution time. Meta itself warns in its getting-started guide that Muse can be inaccurate or take unexpected actions.
Should you try Muse AI?
Muse AI offers an accessible way to explore personal delegation: begin with one bounded task, inspect the output, and see whether repeated use reduces your workload. Its appeal as a personal agent is the convenience of assigning work from your phone while a persistent service handles the steps behind it.
For readers outside confirmed markets, availability remains the first practical hurdle. For those with access, the decision is less about the chart position and more about useful results, manageable permissions, and whether the free allowance covers the work you want done.
