Meta's Muse Agent Now Talks Back in Real Time, With a Face
Mark Zuckerberg announced that Meta now lets users hold real-time voice conversations with Muse, the company's personal AI agent. The Verge reported that the voice mode and expressive avatars are rolling out now in the US for the Muse app. During a voice conversation, the agent's avatar animates in time with the words it speaks.
Users can change the agent's voice tone by asking for a different tone or through options in the avatar settings. The avatar layer is built on Muse Realtime Avatar, which Meta Superintelligence Labs describes as state-of-the-art embodiment technology related to Muse Realtime Voice. Meta says the technology turns Muse Realtime Voice into expressive, interactive avatars.
The voice stack has been building for several months. In June, Meta introduced Meta AI Voice Conversations, powered by Muse Spark, which let users interrupt the assistant, switch topics, or swap languages mid-conversation. In October, Zuckerberg's Meta Connect announcements included voice chat for Muse. Meta's AI voice mode also became available on Messenger, Facebook, WhatsApp and Instagram DMs, with the assistant responding out loud. Meta describes the voice mode as faster and conversational, supporting 13 languages. The Meta AI app has a settings control to toggle "Ready to talk" on, which makes voice the default.
Muse is the more capable layer on top of that conversational base. Reuters reported that Muse can book trips, make purchases, schedule appointments and make phone calls. In September, Reuters reported that Muse could access other apps to send emails and make payments, and that it had launched in the US through a dedicated app and WhatsApp. Meta introduced Muse Spark earlier this year as its most powerful model yet.
Hardware is the next distribution step. Meta said Muse voice mode and expressive avatars are coming soon to its AI glasses, and that they will also be available on the forthcoming Muse Charm, a handheld AI device that Meta expects to ship this year. Zuckerberg's October post described Muse Charm as launching by December.
In my view, the more consequential change is the combination of speech and action rather than the avatar. A voice agent that only answers questions is a convenience. One that can book, buy, schedule, call and move money across apps shifts the failure modes from wrong answers to wrong actions, and voice removes the visual confirmation step that a screen would normally provide. How Muse handles authorization, interruption during a transaction and error recovery will likely matter more to adoption than the lip-sync-style animation.
The broader context here is that the interface is moving from typed prompts to ongoing spoken exchanges, with a persistent avatar as the face of the agent. For developers and enterprise architects, the practical questions are latency budgets for full-duplex speech (where the system listens and talks at the same time, as on a natural phone call), how the agent's tool access is scoped, and whether the same voice and avatar layer will be exposed beyond Meta's own apps and devices.
For consumers, the appeal is straightforward. Talking to an assistant in the language you already use, interrupting it, and changing its tone without digging through menus lowers the friction that has kept many people from using these tools. That gain is real. So is the responsibility that comes with an agent that can spend money and contact people on a user's behalf.
