Meta Launches Muse, a Personal AI Agent That Can Take Real-World Actions

Meta Muse personal AI agent is launching in the United States with the ability to plan tasks, take actions on a user’s behalf, remember personal context, and even complete online purchases.

Meta is launching Muse, a personal AI agent that users can access through a dedicated app or directly through WhatsApp.

Muse runs on Muse Spark, which Meta describes as its most capable model yet for real-world agentic tasks.

Muse can take actions on your behalf

Instead of only answering questions, Muse can plan and carry out tasks based on a user’s request.

The agent can send emails, book travel, open websites, and fill out forms. For more complex goals, Muse can create a personalized plan, coordinate resources, and continue progressing through individual steps.

Meta says Muse can also continue working after a user closes the app. It can return when circumstances change or when it needs approval before taking an important action.

Muse remembers personal context

Muse can remember information users share during conversations and use that context to provide more personalized suggestions later.

Users can also ask Muse to forget specific information it previously learned.

Muse can complete online purchases

Muse integrates with Link by Stripe to handle online checkout.

Link can generate a one-time-use payment card so merchants do not receive a user’s actual card details. Meta also plans to add Shop Pay support and 1Password integration.

Credentials remain securely stored, and Meta says Muse itself cannot access users’ passwords or payment information.

Each Muse agent runs inside a dedicated virtual machine

Every user’s Muse agent runs inside its own Muse Secure VM, a dedicated virtual machine hosted in the cloud.

The VM contains the agent alongside information and credentials for connected services. Meta says each environment remains isolated from other Muse agents.

This approach aims to prevent one user’s agent from accessing another user’s environment or stored information.

Sentinel controls Muse’s internet access

Meta also runs a separate Sentinel agent inside the same virtual machine.

Sentinel remains isolated from Muse at the system level and controls whether Muse can communicate with the internet.

According to Meta, nothing Muse does can reach the internet unless Sentinel approves the action. Sentinel can also request permission from the user when an action requires explicit approval.

Users receive an audit trail showing what Muse has done and what it plans to do next.

Users control which apps Muse can access

Users choose which services Muse can connect to and what permissions it receives.

They can change those permissions or disconnect services at any time.

Meta says Muse conversations and data stored inside its virtual machines are not shared with the company’s advertising systems.

Meta plans an encrypted Confidential VM

Meta plans to introduce Muse Confidential VM later in 2026.

The company says the system will encrypt the entire virtual machine, including user data and Muse conversations, using a key controlled only by the user.

Meta claims this architecture will prevent even Meta itself from accessing the contents stored inside the environment.

Muse is initially launching in the US

Muse is initially rolling out in the United States across iOS, Android, muse.ai, and WhatsApp.

Meta also plans to bring Muse to AI glasses in the future.

The service will remain free for most uses, while subscription plans will offer additional capabilities for users with heavier workloads.

In other Meta news, the company has agreed to an $18 billion settlement over teen safety on its platforms. Meanwhile, WhatsApp may soon flag scam messages before users open them.