Meta’s Muse keeps working after you close the app
Meta’s new personal agent can pursue tasks in the background. Its next destination is AI glasses, with permissions deciding how far it can go.
The task continues after the conversation
Meta wants Muse to carry a task beyond the chat window. Announced on September 8, the personal agent can browse websites, complete forms and keep working while its user is elsewhere. It returns when there is progress to report or a decision that needs approval.[1]
The rollout began in the US on phones and the web, with access through WhatsApp too. Meta describes uses ranging from arranging travel to negotiating a bill. Those jobs depend on the services a person connects and the access they grant.[1]
That makes the handover the important part of the experience. A useful agent has to remember the goal, pick up where it left off and recognize when the next step belongs to its user.
A computer of its own, with limits
Muse runs in a dedicated cloud computer. Meta’s security account describes a separate system called Sentinel that checks requests to contact websites or connected services. It can allow an action, reject it or pause for permission. Credentials are supplied separately, so the model does not need to read the real password or access token.[2]
When approval is needed, the request appears in a separate interface. A purchase, for example, requires confirmation of its details. These controls matter because an agent browsing the web can encounter malicious instructions hidden in the material it reads. Meta acknowledges that Muse can still make mistakes.[2]
There is also a distinction between today’s privacy protections and a promised upgrade. Meta says it can currently access data when necessary to operate, support or secure the service. Its planned Confidential VM is intended to prevent that access cryptographically; it was still being tested when announced.[2]
Next, Muse moves into glasses
At Connect, Meta showed where it wants this to go. Its September 24 recap says Muse will reach AI glasses in the coming months, letting people ask about something in front of them without first describing it.[3]
The update also outlined voice conversations that continue while Muse works, more shopping and workplace connections, and an email address for the agent itself. Glasses support remains a coming feature, rather than part of the initial rollout.[3]
For users, the appeal is less time spent carrying information between apps and following up on unfinished tasks. The test will be whether Muse can do that reliably while keeping its next action understandable and easy to stop.
Sources & context
Based on the sources below, with LLMCurrent’s analysis. Original reporting dated .
- Meta · Muse launch (opens in a new tab)
- Meta · Muse security architecture (opens in a new tab)
- Meta · Connect 2026 recap (opens in a new tab)
About this article
Prepared with AI assistance from the cited material. It does not claim firsthand reporting or independent testing. Source claims are attributed; interpretation is identified in the text. Contact us about a correction.