Meta ships its personal agent: Muse has its own computer and keeps working after you close the app

Meta has released Muse, a personal AI agent. It runs on Muse Spark 1.3, the model announced on 2 September; the agent itself arrived today. The product is framed not as a chat box but as an assistant that finishes tasks.

It has its own computer

The headline description: Muse has its own computer, with a file system and a terminal, so it can write its own code. It searches the web, navigates sites, fills in forms and completes transactions. What comes out the other end includes documents, PDFs, web pages, dashboards and interactive trackers.

It keeps working after you close it

The second claim matters more: Muse runs multiple tasks in the background while the app is closed and sends a notification when something worth your attention changes. The waiting is handed to the agent. Meta AI chief Alexandr Wang's line about personal agents that "work 24/7 on your behalf" described exactly this.

How much of it can you see

With an agent acting on its own, oversight is the real question. Meta names three things here: activity logs, a permissions interface, and access to the agent's memory file. So you can read back what it did, see what you allowed, and inspect what it remembers. For an agent that fills forms and completes transactions, those three are not decoration; they are the minimum.

The model underneath

Muse Spark 1.3 went live in Muse Code and Meta's API on 2 September, with paid developer access opening on 3 September. Pricing is unchanged from 1.2: $1.25 per million input tokens and $4.25 per million output tokens. The agent itself is offered with free and paid subscription tiers.

One line from the announcement is worth quoting: the agent's system instruction reads "Your purpose is to make your user's life better." That is not a measurable target, and the announcement offers no benchmark table or success rate.

Sources: Muse introduction, Muse Spark 1.3 announcement.