Meta Muse: The Personal AI Agent That Books, Buys, and Negotiates

Meta launched Muse, a personal AI agent that doesn't just answer questions — it acts. Give it a goal and it opens a browser, fills out forms, negotiates on your behalf, and pays with your card, pausing for approval on anything sensitive. This guide covers what Meta Muse does, what it costs, how its unusual security model works, and how it compares to the other agents racing to run your life.

What is Meta Muse?

Meta Muse is a personal AI agent that lives in a chat interface and executes multi-step tasks instead of stopping at advice. Meta launched Muse for people 18 and over who want help with day-to-day tasks like schedules and shopping; once you share a goal, it helps develop a plan, coordinates time and resources, and advances the work on its own — it can open a browser, fill out forms, and negotiate on their behalf.

The distinction Meta keeps drawing is between a chatbot and something that acts. Muse keeps working after the app is closed, makes suggestions nobody asked for, remembers preferences across sessions, and turns a saved recipe reel into a shopping list. In practice that means chasing a long-term goal over days rather than resolving a single prompt and forgetting it.

Muse was announced by Meta on Tuesday and is rolling out in the United States on iOS, Android and the web, with support for the company's AI glasses to follow. You talk to it the way you'd message a person — in the Muse app or directly inside WhatsApp.

What Meta Muse can actually do

The headline use cases are errands and negotiations that normally eat your attention:

• Shopping and deals — turn a saved recipe reel into a grocery list, or hunt for a better price on furniture.
• Bills and money — negotiate a bill down, or manage recurring costs.
• Travel and forms — book trips, fill out paperwork, and handle tedious multi-step flows.
• Ongoing goals — health plans, relationships, and career tasks it revisits over time.

The pattern underneath is consistent: Muse plans, executes in its own browser, and returns for sign-off. Muse is designed to require approval for more sensitive actions while allowing previously authorized, lower-risk tasks to proceed. A full audit trail shows what it has done and what it plans to do next.

Meta Muse pricing

There's a free tier plus two paid plans. Meta is offering a free tier of Muse, as well as two subscription options, at $20 per month and $100 per month. Which tier you need depends on usage.

Meta's own framing is that most people won't have to pay. "For the vast majority of users, they should be able to do what they need to within the free tier," Wang told Axios. Whether that holds depends on how heavily the agent is used — the $100 plan is clearly aimed at power users.

The security model: Muse Secure VM and Sentinel

The most technically interesting part of Muse isn't the chat — it's where the agent runs. Each user gets a dedicated, isolated environment. The Muse agent runs on a dedicated virtual machine in Meta's cloud, using a built-in browser that's visible to the user. You can watch it work rather than trust it blindly.

Meta's AI chief has emphasized that the agent is walled off from your real secrets. Alexandr Wang said the app runs within "its own isolated environment" inside the company's computing infrastructure, and "never sees your actual passwords or payment details."

Under the hood, the design leans on a separation of duties. According to a technical write-up of the launch, a separate approval agent — Sentinel — is the sole authority for network access and connector actions, and credentials are handled through surrogation: the agent only ever sees placeholder tokens, while real secrets are injected at the network boundary (MarkTechPost). That makes credential exfiltration via prompt injection structurally futile, since there is nothing real to steal. The browser sub-agent also sees an accessibility tree rather than the raw page and can't execute JavaScript, and the email connector filters out one-time passcodes and password-reset links by default.

For shopping, Meta adds card-level protection: a one-time card number is generated at checkout so your real card stays hidden from both the merchant and the agent. On the privacy question that always follows Meta around, the site states plainly that your conversations are not shared with Meta's ad systems.

What powers Muse: Muse Spark 1.3

Muse runs on Meta's latest frontier model. Muse runs on Muse Spark 1.3, released by Meta Superintelligence Labs; the model targets long-horizon agentic work — zero-shot CLI tool calling, multi-workflow threads, and self-correction across messy sources — and in internal comparisons used roughly 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2. If you want the model details rather than the consumer app, see our breakdown of Muse Spark 1.3.

Notably, the app and the model have different openness stories. Muse itself is a consumer service that developers cannot self-host, but its underlying model, Muse Spark 1.3, is available through the Meta Model API and Muse Code.

Why Meta is doing this

Muse is the first real product behind a much bigger pitch. The product, long in development, was touted as a key next step by CEO Mark Zuckerberg in his recent 6,500-word manifesto. The framing is "personal superintelligence" — an agent that works on your behalf across every part of your life.

The timing is delicate. Less than two weeks after Meta agreed to a massive multistate settlement in a lawsuit over social media's consumer harms, the company announced its biggest bet on consumer AI to date — one that requires significantly more trust than social media ever did. That trust gap is exactly why the security scaffolding is front and center in Meta's messaging.

Meta Muse vs. other AI agents

Muse enters a crowded field. Google, OpenAI, and Microsoft are all shipping agents that try to act, not just chat. The differentiators to watch:

• Execution vs. advice — many consumer AI tools still stop at drafting or suggesting. Muse is pitched as doing the final step itself, which is also where the risk lives.
• Isolation model — the per-user cloud VM plus a separate Sentinel approver is a stricter architecture than agents that run with broad access to your accounts.
• Where it runs — Muse is a hosted consumer service in Meta's cloud. That's the opposite trade-off from local, open-source agents, where your data and the browser stay on your own machine.

If that last point matters to you — running an AI workforce on your own hardware, with your data staying local — it's a fundamentally different design than a vendor-hosted agent, and worth weighing before you connect your email and card to any cloud agent.

Do this with your own AI workforce

Meta Muse shows where personal agents are heading: describe a goal, and a multi-agent system plans and executes it. Eigent brings that same "give it a goal, watch it work" model to a desktop app you run yourself — an open-source, multi-agent workforce that automates real workflows locally, so your data and credentials never leave your machine. Explore what open-source agents can already do, or download Eigent and put your own agents to work.