What Is Meta Muse and How Does It Work?

Muse, Meta's personal AI agent, is a new kind of assistant announced on September 8, 2026. Meta describes it as "a secure, private personal AI agent that proactively helps people meet their goals and suggests ideas," with a one-line pitch that captures the whole product: "It doesn't just answer questions, it actually does the work."

That distinction matters. Chat assistants — including Meta's existing Meta AI — respond in text and wait for your next message. Muse is built to act: send email, book travel, fill out forms, negotiate on your behalf, and keep working after you close the app. This guide explains what it does, how it works under the hood, and where its limits are. For a broader overview, see our main Muse AI guide.

What Muse can do

Meta's announcement names specific capabilities:

• Send emails on your behalf
• Book travel
• Open and use a browser to work on websites
• Fill out forms
• Negotiate for you — Meta cites outcomes like "selling a car for more" and "lowering a bill"
• Turn an Instagram recipe reel into a grocery list
• Remember details like friends' dietary restrictions
• Work asynchronously — "Muse keeps working after people close the app" and "comes back when something changes or when it needs approval"

Set expectations honestly: Reuters reported that Meta's internal tests surfaced failure modes (one flight-monitoring test stopped refreshing), and Meta's Vishal Shah told the outlet that Muse will "sometimes... make mistakes and sometimes... fail to finish things."

How Muse works: the Secure VM

Muse is not just a model answering in a chat window — it runs on its own computer in the cloud.

• A dedicated virtual machine. Muse runs on a dedicated, secure VM that houses the agent, your data, and the credentials for services you connect.
• Its own browser. Where no app integration exists, Muse can open and operate websites the way a person would. TechCrunch reports you opt in to services one at a time — email, calendars, payments, shopping, dining, and more.
• It runs while you're away. Because the work happens in the cloud VM rather than on your phone, Muse can keep churning after you close the app, then come back with results.

Sentinel: the guardrail agent

An agent that acts needs brakes. Sentinel is a second agent running on the same machine, isolated from Muse at the system level. Per Meta, "Nothing Muse does reaches the internet unless the Sentinel approves it" — and when a request is unclear, Sentinel asks you. Meta's safety write-up covers this design in depth, and we break down the privacy details in our Muse privacy guide.

Approvals and the audit trail

Two controls you'll use constantly:

• Approval checks. "Muse checks with the person before sensitive actions like sending an email or making a purchase."
• Audit trail. "Muse shows people a complete audit trail of everything it has done and plans to do." If Muse went off track, the trail is where you'll see it.

Memory — with a delete button

"Muse remembers what matters to a person, and they can always tell it to 'forget' specific things it's learned." In practice: it can hold onto dietary restrictions or preferences, and you stay in control of what it keeps.

What powers Muse — and what Muse is not

Muse runs on the Muse Spark model family (Reuters reports the current version as Muse Spark 1.3), introduced in April 2026 as the first model family from Meta Superintelligence Labs. Several Meta products now share the "Muse" name, and they are easy to confuse:

Name	What it actually is
Muse	The personal AI agent — this guide's subject
Muse Spark	The model family Muse runs on
Muse Image	Meta's image generator (July 2026)
Muse Code	Meta's developer coding agent
Meta AI	The free chat assistant on WhatsApp, Instagram, Messenger, and meta.ai — chat only, no agentic VM

Muse the agent is a new, separate product — not a rebrand of Meta AI.

Where you can get it, and what it costs

Muse is rolling out in the US on iOS, Android, and muse.ai, and you can also use it directly in WhatsApp; support on Meta AI glasses is coming soon. See our availability guide for the current state of the rollout.

On price, Meta's exact wording is that Muse is "free for most of what people need, with subscription plans for people who want to do more." Meta has not published plan pages, but TechCrunch and CNBC report two subscriptions: Power at $20/month and Maximum at $100/month. Our pricing breakdown covers the details, and TechCrunch adds that a payment card is required at signup, with a usage meter warning before free limits are hit.

Privacy: what Meta promises, and what's still open

• Muse "doesn't share a person's conversations or the data in their VM with Meta's ad systems."
• You choose which apps Muse connects to and how much access each gets — email can be read-only or allowed to send — and you can disconnect a service whenever you want.
• Meta says you can opt out of your interactions training its AI models. Press reporting (CNBC) describes this as opt-out, meaning training use may be on unless you change it; Meta has not confirmed the default. Check the privacy settings in the Muse app.
• A "Confidential VM" — encrypted with a key only you hold, so not even Meta could read your data — is listed as coming soon. Until it ships, Wired's reporting flags that Meta's no-access guarantee is policy-based rather than cryptographic.

Keep reading

• How to download and set up the Muse app
• How to Use Muse AI: A Beginner's Guide
• Muse vs ChatGPT: how the two agents compare