Meta Launches Muse Agent To Plan Tasks Beyond Chatbot Responses
The U.S. rollout combines Meta’s Muse Spark model with a secure virtual machine that can keep working after users close the app.
Meta has begun rolling out Muse in the United States as an AI agent that can plan and complete tasks across connected services instead of only answering individual prompts.
Muse can turn long-term goals into step-by-step plans, including tasks such as sending email and booking travel. It can continue operating after the user closes the app and return when circumstances change or approval is needed.
The system runs inside Muse Secure VM, a dedicated cloud computer with its own browser. The environment holds the agent, user data and credentials for connected services. A separate Sentinel agent manages internet access and can stop an action until the user grants permission.
Users can select which applications Muse may access, change permissions, disconnect services and opt out of having their interactions used to train Meta’s artificial intelligence models. Conversations and virtual-machine data are not shared with Meta’s advertising systems.
Muse is available in the United States through iOS, Android and muse.ai. The service is free for most uses, while subscription plans cover additional usage. Support for Meta’s AI glasses is planned.
Muse is powered by Muse Spark, a multimodal reasoning model built for tool use and coordination among multiple agents. Meta introduced Muse Spark on April 8, 2026. Its “Contemplating” mode scored 58% on Humanity’s Last Exam and 38% on FrontierScience Research.
Muse Spark also powers broader functions in Meta AI, including connections to email and calendar applications, research, presentation creation and recurring tasks.
Muse currently supports checkout through Link by Stripe. Shop Pay and 1Password support are planned. Muse Confidential VM is also planned for later in 2026, with encryption keys held only by the user.
The product combines model-based planning with an isolated execution environment, browser controls, permission checks and an audit trail. Meta describes Muse as “a personal AI agent” that “doesn’t just answer questions, it actually does the work.”
