What Is Meta Muse? Meta's Personal AI Agent, Explained in Plain English

CodexFuse Team · 2026-10-27T16:46:49.361Z · 10 min read

Meta Muse is Meta's personal AI agent: instead of waiting for you to ask a question, it takes a goal you give it and works on it by itself — browsing websites, filling in forms, sending emails, and booking things — and comes back when it needs your approval. Meta launched it in the United States on September 8, 2026, and within six weeks the app had been downloaded 10 million times. If you have heard the name and wondered whether it is just another chatbot with a new label, it is not. A chatbot answers. Muse acts. This guide explains what Meta Muse is in plain English, how it works, what it can do, what it costs, and where it stops. It also compares Muse with its two closest rivals, so you can tell which one — if any — fits the way you already work. If you already use one of Anthropic's agents, the comparison will feel familiar: what is Claude Duet covers the other proactive agent in this race.

What Is Meta Muse?

Meta Muse is a personal AI agent built to understand your goals and work toward them around the clock, in Meta CEO Mark Zuckerberg's words at launch: "the personal agent that understands your goals and works 24/7 to get things done for you." You talk to it the way you message a person — in the standalone Muse app, on the web at muse.ai, or directly inside WhatsApp — and describe what needs doing. Muse turns the goal into a plan, opens a browser, fills out forms, and keeps working after you close the app, coming back when something changes or when it needs your approval. Meta's own examples range from sending an email or booking travel up to selling a car, lowering a bill, or maintaining a year-long training plan.

Here is how one task flows:

  1. You give Muse a goal. "Plan a three-day trip to New York next month. Find flights that fit my schedule, shortlist hotels near Manhattan, and keep the total under my budget."
  2. Muse makes a plan. It breaks the goal into steps: check your calendar, search flights, compare hotels, build an itinerary.
  3. Muse does the work in its own cloud computer. It browses, compares, and fills in forms while your phone is in your pocket.
  4. Muse checks back before anything sensitive. Sending an email, booking a flight, or spending money always waits for your explicit approval.

That last step is the one that matters most. Muse is designed to act, but it is also designed to ask first when the action is hard to undo.

How Does Meta Muse Actually Work?

Four pieces work together behind the chat window. None of them is visible while you use Muse, but each one explains a different part of what it can and cannot do.

Muse Spark plans the task

The reasoning engine behind Muse is Muse Spark 1.3, Meta's most capable model. Meta says the model has been specifically trained for long-running agentic workflows: instead of treating every prompt independently, it can keep track of information discovered earlier, operate across multiple workflows, use tools, identify gaps in a plan, and continue working toward a larger objective. It can also split a big job into parts and hand them to subagents that work at the same time. This is the same model family that powers the assistant you are reading about — Muse is both the product name and the name of the model family behind it.

Your own cloud computer

Every user gets an isolated Linux virtual machine — a dedicated computer in Meta's cloud that holds the agent, its workspace, its files, its browser, and its long-running tasks. Because the work happens there and not on your phone, Muse can keep going after you close the app. Your Muse therefore has something closer to a persistent computer than a temporary chatbot session: it can browse websites, work with files, run code, use command-line tools, manage several subagents, schedule recurring jobs, and keep state between sessions.

Sentinel, the guard

Giving an AI access to email, payment systems, and websites creates an obvious problem: what happens if the agent makes a mistake — or is manipulated by malicious content on a website? Meta's answer is another agent called Sentinel. Sentinel is separated from Muse at the system level and acts as the permission authority. Muse can propose an action, but Sentinel decides whether it should be allowed, blocked, or sent to you for approval, and all outbound network activity passes through these controls. The design is aimed at prompt injection — hostile text hidden on a webpage, email, or document that tries to trick an AI into following instructions it was never given. Meta combines model-level prompt-injection training, untrusted-content labels, detection classifiers, browser restrictions, system isolation, and human approvals to reduce the risk. It does not mean prompt injection has been solved, but it shows how differently security must be designed once an AI can take actions instead of simply generating text.

Connectors link your apps

Muse becomes more useful when you connect it to external services. These integrations are called connectors, and they cover email, calendar, banks, restaurants, music, health, and smart home apps, plus public APIs and a browser fallback for everything else. Meta also says Muse can write custom integrations itself for services that expose suitable APIs. On September 18, Meta opened Muse to developer-built connectors, so third-party services can register their own tools for it. Instead of opening five different applications yourself, you tell the agent the outcome you want while it determines which services need to be used.

What Can Meta Muse Actually Do?

The easiest way to understand Muse is through examples. The jobs below are the ones Meta and the early coverage keep returning to, because they show exactly who Muse is for.

JobWhat Muse does
Travel planningChecks your calendar, researches flights, compares hotels, prepares an itinerary, and brings the final choices back for approval
ShoppingResearches products across sites against your budget and requirements, then checks out with an approved payment method
EmailReads mail with permission, summarizes conversations, finds information, and drafts or sends messages after approval
Calendar and eventsFinds free times, books reservations, and adds the result to your calendar
GoalsTracks long-term objectives — a cheaper phone plan, a flight price, a fitness plan — and pings you when something changes
ResearchBrowses multiple sources, collects information, reconciles conflicting data, and produces a finished deliverable
Small businessConnects to Shopify, Stripe, QuickBooks, Klaviyo, Canva, Figma, Slack, Asana, and Notion through Muse for Small Business

Muse also remembers details you mention once and can act on them unprompted — turning a saved recipe reel into a grocery list, or recalling a friend's dietary restriction before sending invites — and you can tell it to forget specific things. It has a Feed of posts it writes based on your interests, an Ideas tab that proposes tasks it could take on, a Goals tab for long-term objectives, and a Library that stores the documents and charts it has produced.

How Much Does Meta Muse Cost?

Meta Muse is free to start, with two paid tiers for people who use it heavily. Meta's AI chief Alexandr Wang told Nikkei in October 2026 that the agent will stay free of advertising, and that subscriptions — not ads — will fund it as it expands internationally. That is the business model in one sentence: the basic agent is free, and power users pay.

PlanPriceWho it is for
Free$0Everyday use, with a weekly usage limit
Power$20 a monthHeavy users who want a much larger weekly usage allowance
Maximum$100 a monthPeople who run the agent constantly and want the largest allowance

An independent guide that checked the tiers against Meta's help page on October 1, 2026, reports the weekly limits as roughly 100 million tokens on the free tier, 500 million on Power, and 3 billion on Maximum — but Meta has not published those figures itself, so treat them as reported, not official. The prices are the firm part: free, $20, and $100 a month. Meta also takes commissions from sellers on purchases made through the agent, which is a second revenue stream beyond subscriptions.

How Do I Get Meta Muse?

Getting started takes a few minutes, but who can get it depends on where you live.

  1. Check the country list. Muse is available in the United States, Canada, and Japan. It launched in the US on September 8, 2026, in Canada on September 18, and in Japan on October 23, 2026. Neither the UK nor India has a date yet.
  2. Check the age rule. You must be 18 or older.
  3. Download the app or open the web. Get the Muse app on iOS or Android, open muse.ai in a browser, or talk to it inside WhatsApp. A Mac app was added on September 17.
  4. Connect the apps you want it to use. Email and calendar first, then anything else — banks, shopping, music, smart home — one connection at a time, with each permission granted separately.
  5. Give it one small job first. Ask it to summarize your unread email or find a restaurant booking. Watch how it asks for approval before it sends or books anything, then hand it bigger jobs.

The early numbers explain why Meta is pushing this hard. The app passed 5 million downloads in 22 days, reached 10 million downloads in its first six weeks, and topped the US App Store free chart on September 18, 2026, pushing ChatGPT down. Sensor Tower data reported by TechCrunch put Muse's US 7-day retention at 30%, against a 19% average for ChatGPT, Claude, and Grok — people who install it tend to keep using it.

Meta Muse vs Claude Duet vs ChatGPT

Muse did not arrive alone. Anthropic launched Claude Duet, its own proactive agent, on October 21, 2026, and OpenAI followed with Dots, its always-on agent. The three are the same idea from three companies: an agent that works in the background and messages you first. The differences are in who can get them, what they cost, and what they are built for.

Meta MuseClaude DuetChatGPT
LaunchedSeptember 8, 2026October 21, 2026Chatbot since 2022; Dots agent in 2026
What it isPersonal agent for anyoneProactive agent for Claude subscribersChatbot first, agent second
PriceFree; $20 or $100 a monthIncluded in Claude Pro ($20) and Max (from $100)Free tier; Plus $20 a month
AvailabilityUS, Canada, Japan; 18+US Pro and Max subscribers, waitlisted research previewWorldwide
How it worksOwn cloud VM, Sentinel guard, Link paymentsOwn cloud VM, morning brief, 6,000+ connectorsChat thread; Dots runs its own cloud computer
Approval ruleAsks before sending, booking, or buyingAsks before sending, paying, or signingAsks before agent actions
Best forEveryday personal tasks and shoppingWork tasks across Gmail, Slack, and ShopifyQuestions, writing, and coding

The deeper difference is the audience. What is Claude Cowork covers Anthropic's file-working agent, and Duet is built for people who already live in Claude for work. Muse is built for everyone else: it arrives inside WhatsApp, where billions of people already are, and its first jobs are personal — email, travel, shopping, goals. If you want to understand the category rather than one product, what is agentic AI explains the shift from chatbots that answer to agents that act. And if you still mostly need answers rather than actions, how to use ChatGPT covers the chatbot side of the same world.

Where Meta Muse Stops: Amazon, the Banks, and the Limits

Muse's limits are as informative as its features, because they show where the agent idea meets the real world — and the real world is pushing back.

Amazon blocked it

On the night of September 20, 2026 — twelve days after launch — Amazon began blocking Muse from completing purchases on Amazon.com. Amazon said Muse did not identify itself as an automated agent, arrived without Meta asking, and appeared to capture and store customer credentials, including access to account pages and order history. Users who ask Muse to buy on Amazon now see a pop-up saying continued access by an unauthorized AI agent violates Amazon's Conditions of Use. Meta disputes the credential part: its launch post says credentials go into secure storage so Muse can use them without seeing them. The split is the story: Shopify announced a Muse partnership on September 21, and Shop Pay became available for eligible purchases on September 24, while Walmart, Best Buy, Wayfair, Gap, Sephora, Ulta, and Dick's Sporting Goods signed on as Muse partners. One camp partners with agents; the other blocks them. Where you shop decides which camp your agent can serve.

The banks are blocking it too

On October 13, JPMorgan Chase CEO Jamie Dimon said on the bank's third-quarter earnings call that the bank will not let outside AI agents operate inside customer accounts: "We're not going to allow just any outside agent to come in inside the company, because it's just too risky for clients and for JPMorgan. You don't want rogue agents, and it could be thousands of them." A JPMorgan spokeswoman confirmed the bank blocks external agents from acting on behalf of retail customers, while balances can still be seen read-only through open-banking connectors such as Plaid. Citigroup and Bank of America have put similar restrictions in place, and in Canada, RBC's CEO said on October 20 that the bank is considering blocking Muse from customer information, while BMO's CEO warned a day later that handing agents your account numbers, passwords, or PINs is a bad idea. The practical rule for you: an agent can watch your money, but the big banks will not let it move your money — and Dimon added that customers who authorize outside agents anyway should understand the liability shifts to them.

Meta can still see inside the VM

Muse's Secure VM isolates your data from other users, but at launch it does not technically prevent Meta from accessing the VM when required to support, secure, or operate the service. Meta says a future Muse Confidential VM will encrypt the environment with a key only you hold. Until that arrives, understand the difference between isolated from other users and cryptographically inaccessible to Meta.

Agents still make mistakes

A language model can generate an incorrect answer. An agent can generate an incorrect answer and then act on it. Meta itself acknowledges this, and its security researchers state plainly that Muse "can and will still make mistakes" — which is why the product uses multiple layers of system-level protection, including Sentinel, approval gates, and an audit trail you can read. Meta has also opened a bug bounty program with rewards reaching hundreds of thousands of dollars for serious Muse vulnerabilities. AI glasses integration is coming within months, announced at Meta Connect on September 23, 2026, which will extend the same questions to a device you wear.

Should You Use Meta Muse?

Muse is worth trying if you are in the US, Canada, or Japan, you are 18 or older, and you have the kind of recurring jobs agents are good at: inbox triage, travel research, price watching, appointment booking, and shopping comparisons. Start with the free tier, connect one app at a time, and give it small jobs first so you learn how its approval prompts work before you hand it anything involving money. Check every approval it asks for — the approval step is the whole safety model, and it only works if you actually read it.

Wait if you are outside the three launch countries, if your bank is one of those blocking agents and you wanted it to manage money, or if you do most of your shopping on Amazon, where it is blocked. And keep one rule in mind whatever you decide: an agent is only as trustworthy as the approvals you give it. Muse asks before it sends, books, or buys. The answer should always be yours.

Sources

Frequently Asked Questions

Is Meta Muse free?

Yes. The basic version of Meta Muse is free, with a weekly usage limit. If you use it heavily, Power costs $20 a month and Maximum costs $100 a month, both raising the weekly allowance. Meta says the agent will stay free of advertising, with subscriptions funding it instead.

How do I get Meta Muse?

Download the Muse app on iOS or Android, open muse.ai in a browser, or talk to it inside WhatsApp. You must be 18 or older, and it is available in the United States, Canada, and Japan. There is no date yet for the UK, India, or other countries.

Can Meta Muse buy things on Amazon?

No. Amazon blocked Muse from shopping on Amazon.com on the night of September 20, 2026, saying the agent did not identify itself and accessed customer accounts without authorization. Muse can still shop with partner retailers — Shopify stores through Shop Pay, plus Walmart, Best Buy, Wayfair, and others — and it can compare Amazon prices and leave the checkout to you.

Is Meta Muse safe with my bank account?

Partly. Muse can see balances read-only through open-banking connectors such as Plaid, and it asks for your approval before spending. But JPMorgan, Citigroup, and Bank of America restrict outside agents from logging in to customer accounts or moving money, and JPMorgan says customers who authorize outside agents anyway carry the liability. Never hand an agent your bank password or PIN.

What is the difference between Meta Muse and ChatGPT?

ChatGPT is mainly a chatbot: you ask a question, it answers. Muse is a personal agent: you give it a goal, it plans the steps, works in its own cloud computer, and acts — sending emails, booking travel, and making purchases — asking for approval before sensitive actions. ChatGPT's own agent, Dots, works the same proactive way, and Anthropic's Claude Duet is the third rival in this race.