META LAUNCHES MUSE, AN AI AGENT THAT SENDS EMAILS AND SELLS YOUR CAR

On September 8, 2026, Meta launched Muse, a personal AI agent that performs tasks on a user's behalf: sending emails, booking travel, buying tickets, filling out forms, negotiating bills and building long term plans. It runs on Meta's Muse Spark model inside a dedicated virtual machine with its own browser, monitored by a separate Sentinel agent, and Meta says it cannot see passwords or payment methods and does not share conversations with its ads systems. Muse is available to US adults through an app, the web at muse.ai and WhatsApp, with a free tier, a Power plan at $20 a month and a Maximum plan at $100 a month.
Key Points
- Muse launched September 8, 2026, for US adults through a standalone app on iOS and Android, the web at muse.ai, and WhatsApp, with Meta's AI glasses to follow.
- The agent runs on Meta's Muse Spark model inside a dedicated virtual machine with its own browser, with a separate Sentinel agent watching it, and Meta says it never sees passwords or payment methods.
- Pricing: a free tier with weekly limits, a Power plan at $20 a month, and a Maximum plan at $100 a month. A payment card is required to start even on the free tier.
- The pitch comes from Zuckerberg's 6,500 word August essay promising everyone a personal agent working around the clock, and from Meta's $2 billion purchase of the agent startup Manus.
Meta launched Muse today, and it is not another chatbot. Muse is a personal agent that acts for you: it sends the email, books the flight, lists the car, calls the cable company to lower the bill, and fills out the permission slip for the field trip. It went live Tuesday, September 8, for adults in the United States through a standalone app, the web at muse.ai, and WhatsApp, with Meta's AI glasses promised next. Meta says it can "open a browser, fill out forms, and negotiate on their behalf." That last verb is the whole product.
Emails, Travel, a Car Sale, a Lower Bill
The examples Meta put in front of reporters are deliberately domestic. Buy movie tickets. Book a tennis lesson. Turn a recipe into a grocery list. Send the invitations. Build a yearlong exercise plan. Set up a new business. Sell a car. Get a bill lowered. The framing is that a chatbot retrieves and a personal agent does, and Muse is the second thing.
It connects to the services where that work actually happens: email, calendars, payments, health and fitness apps, smart home systems, dining, shopping, music and events. Connections are made one at a time, on purpose, so the opt in is visible every time. Checkout runs through Link by Stripe, with Shopify's Shop Pay and 1Password integrations on the way, and anything without a built in connector can be reached through a public API or by letting Muse drive a browser. You can name the agent and give it an avatar, the same personified shape the whole category has been drifting toward.
A Computer of Its Own, With a Guard Inside
The engineering claim Meta leads with is containment. Muse runs on Meta's Muse Spark model inside what the company calls Muse Secure VM, a dedicated virtual machine with its own browser that holds both the agent and the user's data. A second agent, Sentinel, runs on the same machine but is kept apart from Muse at the system level, watching what it does. Meta says Muse has no visibility into passwords or payment methods, and that it does not share conversations or data with Meta's ads systems.
Read that list as a set of promises Meta expects to be tested on. Every one of them answers a specific question a skeptical user would ask first, which tells you the company knows exactly which questions those are.
Free, Then Twenty, Then a Hundred
Muse is free to use with weekly limits. Above that sit two subscriptions: Power at $20 a month and Maximum at $100 a month, the same two price points the rest of the industry settled on for assistants and agents. One detail matters more than the tiers: Meta requires a payment card to start, even on the free plan. An agent that can buy things needs a way to pay for them, and Meta wants that wired up before the first task.
Meta's Record Is the Part It Cannot Patch
The obvious question is whether people will hand Meta their inbox, their calendar and their card. TechCrunch put the company's history in one paragraph, and it is worth restating: a 2011 FTC settlement over deceiving users about privacy, a record $5 billion FTC penalty in 2019 for eight separate privacy violations, a 2023 FTC charge of violating those orders, a 2019 disclosure that user passwords had been stored in readable form, and Cambridge Analytica. The secure virtual machine and the Sentinel agent are engineered answers to a trust problem that is not an engineering problem.
Meta is not alone in asking. Google's Gemini Spark, Claude Cowork and the startup Instinct are chasing the same job, and Instinct's early testers were shocked by licensing terms that asked for the right to "access, use, host, cache, store, reproduce, transmit, display, publish, distribute, and modify" their material. The whole category is racing on access, and access is the thing users are most reluctant to give. Meta already learned that lesson in miniature this summer, when Instagram's Muse image feature shipped with your photos switched on by default. Same name, same instinct.
Zuckerberg Wrote the Pitch Last Month
None of this is improvised. Mark Zuckerberg published a 6,500 word essay in August that critics called fantastical, in which he wrote: "Everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about. Your agent will work 24/7 on your behalf to improve your relationships, health, career, finances, home management, hobbies, and more." Muse is that sentence turned into an app. Behind it sits the Muse Spark model family Meta's lab has been shipping this year, and the $2 billion Meta paid for Manus, the agent startup, to buy its way into the category.
The glasses are the tell for where this goes. Meta's glasses launched in June in 26 styles and already translate speech in under three seconds. An agent that can negotiate for you, living on your face, is the product Meta has been describing since it renamed the company.
Prediction: the free tier gets millions of signups because the card requirement is the only friction, the Power tier becomes the default for anyone who lets it touch their calendar, and the first story about Muse buying the wrong thing lands before October.
Frequently Asked Questions
What is Meta Muse?
Muse is Meta's personal AI agent, launched September 8, 2026. Unlike a chatbot, it takes actions for you: sending emails, booking travel, buying tickets, filling out forms, negotiating bills and building long term plans.
How do you get Meta Muse?
Through the Muse app on iPhone and Android, on the web at muse.ai, or by messaging it in WhatsApp. It is available in the United States only, for people 18 and over, with Meta's AI glasses promised later.
How much does Meta Muse cost?
There is a free tier with weekly limits, a Power plan at $20 a month, and a Maximum plan at $100 a month. Meta requires a payment card on file to start.
What model does Muse run on?
Meta says Muse runs on its Muse Spark model, the family its AI lab has been shipping this year, inside a dedicated secure virtual machine Meta calls Muse Secure VM.
Can Meta Muse see your passwords?
Meta says no. Muse runs in its own virtual machine with its own browser, a separate Sentinel agent is kept apart from it at the system level, it has no visibility into passwords or payment methods, and checkout runs through Link by Stripe.
Does Meta use Muse conversations for ads?
Meta says Muse does not share people's conversations or data with its ads systems. Users connect apps and services one at a time on an opt in basis.
Topics: shopify, manus, google, techcrunch, meta, muse-spark, personal-agent, stripe, whatsapp, mark zuckerberg, ai-agents, mark-zuckerberg, 1password, muse