Meta Puts Its Muse Agent Behind an Early Access List
Meta opened requests for early access to new capabilities in Muse, its personal AI agent, on September 25, 2026. The program follows new Muse announcements at its Connect 2026 developer conference. TechCrunch
Access is opt-in. Users can join the list by asking Muse to add them, using the prompt "Can you let the Muse team know I want to be part of the Muse early access program?" Meta also posted a link on X that, when shared with Muse, logs interest in joining the program.
The features in early access center on more persistent and more agentic interaction. They include a digital avatar for Muse that users can video chat with, additional shopping partnerships and connectors, and an expanded Muse Mac app that will use the user's computer to complete tasks. Muse will also come to Meta's lineup of AI glasses, where users interact via a wake word and voice command.
Muse launched on September 8, 2026, in the US via a dedicated app and WhatsApp. Reuters Meta describes Muse as the world's first personal AI agent, a secure, private assistant that proactively helps people meet goals and suggests ideas. Meta In practice, it connects to email, calendar, Instagram and other apps, according to Meta's product documentation, as well as payments apps and health apps. It can send emails and book travel on behalf of users, and complete transactions such as selling a car. Reuters
Glasses are the more consequential surface. Meta said Muse would become the primary interface for its glasses. Through glasses, Muse allows users to shop, make appointments and complete tasks through voice. Reuters On AI glasses, Muse can act on what the user is looking at without requiring the user to describe it. Meta
Underneath is a model refresh. At Connect 2026, Meta showed Muse improvements including new features, connectors, and a state-of-the-art model. That work builds on Muse Spark, introduced by Meta as its most powerful model yet to power a smarter and faster Meta AI assistant. Meta A related system, Muse Image, added more than 30 AI-powered effects for Instagram Stories.
Looking at what this means for practitioners, the list is less interesting than the architecture it implies. Avatar video chat tests whether users want an embodied agent for sustained tasks rather than a chat box for discrete prompts. Computer use on macOS tests permissioning, sandboxing, and auditability when an agent operates local UI and files. Glasses with wake-word invocation and gaze-grounded context test always-available inference with tight latency budgets. None of those are new research problems. Combined in one consumer agent with connectors to mail, calendar, payments, and commerce, they become deployment and trust problems.
In this author's view, worth flagging is how Meta is sequencing distribution. The dedicated app and WhatsApp provide scale. The Mac app provides a place where multi-step work actually lives. Glasses provide a hands-free input where describing context is costly, so acting on what the user sees has practical value. If the early access cohort uses the avatar for longer sessions, delegates desktop tasks, and completes purchases or bookings by voice, Meta will have evidence for an agent that persists across devices rather than a feature inside any single app. The long arc here is familiar from the PC to mobile to cloud transitions. Interfaces that reduce friction get used for more consequential work, and the controls around identity, payments, and data access matter as much as model quality.
