What Muse Is and What It Can Do

Muse is Meta's personal AI agent: software that carries out tasks for you, not just answers questions. It launched on September 8, 2026 at muse.ai. Meta Superintelligence Labs built it, and it runs on a model called Muse Spark. When you open muse.ai, it signs you in through Meta. That's how you can tell the site is Meta's.

Several things have similar names, and mixing them up causes real problems when IT or counsel review your request:

• The Meta AI app is a separate Meta product with its own terms. Approval for one doesn't cover the other.
• Muse Spark is the model under Muse, not an app you sign up for.
• Project MUSE (an academic journal service) and the Muse website builder have nothing to do with Meta's agent.

Example: you email your IT lead, "Can I use Muse for meeting prep?" A precise version is: "Meta's Muse agent at muse.ai, which has its own Supplemental Terms at muse.ai/terms, separate from the Meta AI app." Being that precise saves a round of back-and-forth.

A chatbot answers when you ask and stops there. An agent takes a goal and works toward it. Meta says Muse turns long-term goals into plans, keeps working in the background after you close the app, and comes back when something changes or when it needs your approval. It asks through an approval card, an in-app prompt where you allow or deny a specific action.

Take this course's running example. A national chain offers a three-year contract worth nearly a fifth more volume, capacity may be short, and the board chair wants your view by Friday.

• With a chatbot, you paste in a question, get an answer, and have to come back to ask again.
• With Muse, you can set a standing job, such as "each weekday morning, flag new public news about the chain and list my meetings about it." It runs on schedule without being asked again.

Meta's examples go further: sending email, booking travel, filling in forms, buying things, negotiating. Muse can do these, but in this course it prepares and drafts, and you approve, send and decide.

Meta lists five capabilities. Each one matches part of your recurring preparation work:

1. Goals in the background. Muse holds an aim and checks in on a schedule. Fits: turning an annual goal into quarterly milestones with a weekly check-in.
2. Memory you can edit. It remembers what you tell it, and you can correct or delete entries. Fits: your role, your working hours and rules like "draft, don't send."
3. Its own cloud computer, the Muse Secure VM. Muse has a browser, a file system and a terminal, and it can write code for a task. Fits: researching public sources on the chain and its market.
4. Artifacts. These are outputs it creates, such as documents, PDFs, web pages and trackers. Fits: a one-page decision brief or a draft board update.
5. Connectors. These are links to other services, such as Gmail and Google Calendar. Fits: a weekday brief built from your inbox and calendar, read-only.

Not everything it can do belongs in your role. Board, deal and employee material stays out, and later lessons cover why.

Learn these locations now, because every later lesson sends you to one of them:

• Chats: one long main chat for your ongoing work, plus side chats that keep a project separate. You might keep the chain contract in its own side chat.
• Goals: where long-running goals and their plans are tracked and edited.
• Ideas: suggestions Muse offers without being asked.
• Library: files, including System Files, which give a closer look at what Muse has done.
• The Assistant icon, which opens:
  • Identity, which holds Memory (what it has saved about you) plus its personality and name settings.
  • Activity log, a dated record of the actions Muse took and the permissions it used.
  • Upcoming, a list of scheduled tasks and reminders.

A typical week: you set up a brief in Chats. It appears in Upcoming. It runs and posts each morning. On Friday you open the Activity log to confirm it only read your email and never replied.

Meta has announced more than it has shipped. At its Connect event on September 23, 2026, it announced a voice mode, Muse Realtime Avatar and a Muse email address you could copy on threads, with no ship date given. Muse on AI glasses is promised "in the coming months." Press have reported phone calling. Meta tested a version in which human contractors placed calls, for Meta employees only, then rolled it back and admitted the missing disclosure was a mistake.

None of these is confirmed as available as of October 28, 2026, so this course doesn't teach them. If you see a guide that shows them, treat it as unconfirmed.

One more limit matters for leaders: Muse has no team workspace, shared agent or team plan. Every person needs their own account and subscription. If your chief of staff also wants Muse, they set up their own, and their Muse has no access to yours.

Meta updates Muse often. The menu names in this course come from Meta's Help Center as of October 28, 2026, and your screen may already differ. Labels move, get renamed, and can vary between iPhone, Android, Mac and the web.

When a name doesn't match, don't guess at a setting that controls access. Use this routine:

1. Note the name the course uses, such as Settings > Permissions.
2. Look for the same setting under a nearby or renamed label.
3. If it isn't obvious, open Meta's Help Center page for that feature and follow its current wording.
4. Don't rely on Muse's own account of how it works. Meta has said the agent's explanation of itself was wrong in at least one reported case. Rely on the Help Center and the Activity log.

This lesson's main point carries through the whole course: Muse prepares and drafts, and you approve, send and decide. Next lesson covers who can get Muse, what the plans cost, and the incidents since launch.