Muse (Meta)
Also called Muse, Meta Muse.
In plain English
Muse is Meta's personal AI agentAn AI agent is a system that uses an AI model to choose actions toward a goal, use the tools it has been given, and respond to the results. for carrying out tasks across apps and websites a person connects.
Example
You ask Muse to turn your notes and selected documents into a draftA draft is work that is not finalized yet. event plan, keep a list of open questions, and bring proposed messages back for you to approve.
Why it matters
Meta’s announcement, dated September 8, 2026 and updated September 30, describes Muse running on Muse Secure VMThe Muse Secure VM is Muse's dedicated cloud computer, with its own file system, terminal, and browser. with its own browser, powered by the Muse SparkMuse Spark is the family of AI models from Meta that powers the Muse assistant. model, and reachable through the Muse app or WhatsApp. It says Muse can keep working after the app is closed and asks for approvalAction approval is a person's permission for one proposed action to go ahead. A useful approval shows what will happen and what it will affect. before sensitive actions such as sending email or making a purchase.
Meta also describes a separate SentinelSentinel is a separate Meta agent that checks Muse's outbound internet actions before they happen. agent that approves outbound internet actions, an audit trail of what Muse did, a way to tell Muse to forget something, and an opt-out from training. A Muse Confidential VM is announced for later. Price and availability change, so those details stay on Meta’s pages.
Common confusion
Muse is the agent. Muse SparkMuse Spark is the family of AI models from Meta that powers the Muse assistant. is the model named in Meta’s announcement. They are not interchangeable. A safety system lowers risk but does not rule out mistakes or prompt injectionPrompt injection is an attempt to make an AI follow untrusted instructions instead of the task and rules it was given..
Sources
Updated October 1, 2026.


