Meta plans to bring Muse, its personal AI agent, to AI glasses in the coming months, letting wearers call it by name and ask it about an object, sign or list in view. The company announced the update at its Sept. 24 Connect event, alongside longer voice conversations that can continue while Muse works on a task in the background.
Muse is an agent service powered by Meta’s Muse Spark model. It can browse websites, fill forms and keep working after its app is closed. People can connect services such as email and choose how much access to give them. Meta says the system keeps a personal memory of details users share so it can use that context in later tasks. The current service is available in the United States through its apps and website, and through WhatsApp.
The glasses version is still a plan, not a released feature. Meta says a wearer will be able to say Muse’s name, ask about an object, sign or list in view, and have it take action using that visual context instead of requiring a description.
The underlying browser-task pattern is not new. TechRadar described Muse’s cloud-browser workflow as familiar from OpenAI and other assistants, while WIRED compared its browsing with the reviewer’s earlier test of ChatGPT Agent. Those were individual experiences rather than a matched test. Meta’s product distinction is its emphasis on persistent personal context, connections to services such as Instagram and Facebook, background tasks and a planned hands-free interface through its glasses.
Meta is also widening Muse’s connections. At Connect, it announced more shopping and payment partners, including Walmart, Best Buy, PayPal and Shop Pay, and named Notion, GitHub and Box among the services for work tasks. Meta said Expedia was coming soon and Instacart was already available. The announcement mixes current and planned integrations, so not every named connector is usable today. Meta also plans to give Muse its own email address.
Two recent hands-on reviews show what the agent can do before a person must step back in. TechRadar reviewer Eric Hal Schwartz asked Muse to find shorts; it asked for a size, returned options and a sale, then requested an Amazon sign-in before a purchase. After the reviewer granted email access, Muse found a utility-company message and drafted a reply for approval. WIRED reviewer Reece Rogers said Muse found nearby couches on Facebook Marketplace and offered to contact sellers. In a separate bakery test, it put a sandwich in the cart but stopped for payment setup and final approval; Rogers noticed it had selected no tip and chose to order in person. These are two individual tests, not typical-use data or a controlled benchmark.
Meta says each user’s agent runs in an isolated cloud computer with its own browser, and that a separate Sentinel system must approve actions that reach the internet. It says credentials are stored outside the agent’s view and asks users to approve sensitive actions such as sending email or making a purchase. Meta also acknowledges that agents can make mistakes and that prompt injection remains an open security problem. Those safeguards are the company’s description, not an independent audit; a more private “Confidential VM” is still planned for later this year.
Across both reviews, Muse moved an errand forward but left sign-ins and consequential actions to the person.
Image: ItzANormalFioko, Ray-Ban Meta second-generation smart glasses, CC BY-SA 3.0. The photo shows an existing glasses model, not the unreleased Muse interface; used without modification. License.
Sources
Meta Connect announcement: Muse, glasses, voice mode and connectors
Meta AI Research: Muse’s architecture, safeguards and known risks