TL;DR

  • Muse works through its app and WhatsApp, according to its launch site.
  • The product is built around persistent conversation, side chats, goals, and background work.
  • Its action model includes browser use, app connections, approval cards, and an activity log.

Muse is trying to be more than a chat app

Muse is Meta’s new personal AI agent, built to take on goals and everyday tasks. The launch materials make a strong claim: Muse should feel like a long-running conversation with an assistant that can also do work across the web and your connected apps.

That leaves a fair question: what can Muse actually do?

The official materials give a more detailed answer than the launch post. They describe an agent that can plan, browse, create, track work in the background, and ask for approval before it takes sensitive actions. This is what each piece means in practice.

Talk to Muse in the app or on WhatsApp

Muse says people can message their agent in the Muse app or directly through WhatsApp. The product team designed the main chat as one ongoing conversation, rather than a series of isolated prompts. Its memory persists across conversations, while side chats let people separate more complex projects.

That structure matters for the tasks Muse is aiming at. Planning a trip, organizing school preparation, tracking spending, or building a health routine can all stretch across days and involve many smaller decisions. The product is meant to carry that context forward.

Muse also lets people personalize the agent with an avatar, name, and style. That is a product choice, not a capability claim, but it reflects the intended relationship: an assistant you return to rather than a blank chat window each time.

Set a goal and let it keep working

The Goals view is where Muse tracks ongoing tasks and its plan to reach them. The company says Muse can break a goal into actionable steps, work in the background while the app is closed, and respond to relevant events or a schedule.

Muse says it will only notify people when there is something meaningfully new or when it needs input. That is a sensible design target. Persistent agents can become noisy quickly if every background step turns into a notification.

The product also has an Ideas area. Muse says it can suggest ideas based on your goals, patterns, and prior conversations. The useful test is whether those suggestions become more relevant with real use, rather than adding another stream of generic prompts to ignore.

Browse, connect apps, and take actions

Muse says it can use a full web browser to search, navigate websites, fill out forms, and complete transactions. It also says people can connect daily-use apps such as email, calendar, and Instagram, while choosing what the agent can see and do.

The team says Muse has its own computer environment and can write code or build a tool if a task requires it. That helps explain the range of work shown in the launch materials. A personal task can move from an email to a calendar, then to a website, a form, and a purchase without fitting neatly inside one app.

The product is also designed to produce more than a written reply. Muse can create documents, PDFs, web pages, trackers, study guides, and dashboards. These outputs are called Artifacts and can live outside the chat.

Approvals, permissions, and the activity log

Action-taking agents create an obvious question: when does the person take back control?

Muse says it asks for approval before actions such as sending an email or making a purchase. The product includes structured approval cards, a list of permissions you have approved, and an activity log showing what Muse is working on. The company says it defaults to stopping for actions that are difficult to reverse, while letting ordinary web browsing proceed.

Users can adjust the caution level in settings. That does not remove the need to pay attention to what an agent is doing, especially around money or communication. It gives users visible controls instead of asking them to trust an invisible process.

The first use cases are everyday coordination problems

The company is showing Muse for planning, finances, health, shopping, and relationships. In its design story, a product leader describes using Muse to coordinate back-to-school work across school emails, calendars, websites, a shopping cart, and a restaurant booking.

That is still an internal launch example. Independent user patterns will take time to emerge. For now, it is the best window into what Muse is designed to handle: tasks that are too scattered for a checklist but too routine to deserve constant manual attention.

Read Muse’s launch overview for the product context, or Muse vs. Claude vs. Manus for where it sits among other action-oriented AI products.

Sources