Meta lets US users voice and video chat with animated Muse AI agent

Meta has rolled out real-time voice and video chat for its Muse AI agent to everyone in the US, letting users talk out loud with an animated avatar that answers and acts on their behalf. The company says the feature runs on two new AI models, Muse Realtime Avatar and Muse Realtime Voice, and is available now on mobile, web and Mac.

CEO Mark Zuckerberg announced the update on November 5, and Meta’s chief AI officer, Alexandr Wang, announced the rollout in a blog post and on X alongside a demo of his own Muse. The feature was first introduced at Meta’s Connect developer conference in September.

What Muse Realtime Avatar and Realtime Voice add to Muse

Until now, Muse was primarily a text-chat agent. TechCrunch reports the new capability under the name Muse Voice: a feature that lets users call and video chat with Meta’s consumer AI agent. Meta describes the release as voice plus “expressive avatars”.

Three things change for users:

• Spoken conversation in real time. Users can talk to their agent to give instructions or ask questions instead of typing.
• An avatar that animates as it talks. The avatar at the top of the conversation animates in time with the words it is saying. Meta says it uses “subtle expressions and gestures” that match the tone of the conversation, and the mascot’s facial expressions match what it is saying.
• Selectable voices. Meta says users can pick from hundreds of voices. The voice’s tone can be changed by asking the agent and letting it decide, or by browsing options in the avatar settings.

The “video” in video chat is one-directional. Muse does not actually see the user; it hears them and reacts based on the conversation. In practice, this is a voice assistant with a lip-synced, gesturing character on screen, not a model that processes the user’s camera feed.

Meta’s own announcement frames it this way:

“Starting today in the US, we’re rolling out voice and expressive avatars in the Muse app across mobile app, web and Mac app, giving you a new, natural way to interact with your Muse. Want New Yorker energy? Pirate flair? Pick from hundreds of voices and switch it up whenever your mood does.”

Meta has not published technical details for Muse Realtime Avatar or Muse Realtime Voice in the coverage available: there are no figures on latency, supported languages, model size or how the two models divide the work. TechCrunch describes them only as new AI models and technology that “help power the digital avatar that chats with the user”.

Where the Muse voice and avatar features are available

The rollout is US-only and, according to Meta, open to everyone in the country rather than a waitlist. The table below summarises what the sources state.

Item	What Meta and the reports say
Announced	November 5, 2026
Models	Muse Realtime Avatar and Muse Realtime Voice
Region	United States
Platforms now	Mobile, web and Mac
Voices	Hundreds of options
Camera input	None; Muse hears but does not see the user
Coming later	Meta’s AI glasses (“soon”) and the Muse Charm device

Meta says the voice mode and expressive avatars are coming “soon” to its AI glasses and will also be available on its forthcoming Muse Charm gadget, which is supposed to ship this year. TechCrunch describes Muse Charm as a Tamagotchi-like wearable planned before year-end, on whose small screen the avatar will appear so users can talk to the agent and have it reply or carry out tasks.

None of the sources mention an API, SDK or any developer access to Muse Realtime Avatar or Muse Realtime Voice. For now, the models are only reachable through Meta’s consumer Muse app.

How Muse’s voice mode performs in early hands-on use

The only independent impressions so far come from The Verge’s Jay Peters, who had the feature rolled out to him and tried it on the iOS app. His report is short and based on brief tests, so it is an early signal rather than an evaluation.

• Responsiveness: his Muse agent responds quickly over voice, though it occasionally pauses while looking up information or checking something, as it would with a text query.
• Animation: the avatar animates what it is saying well, but with his particular avatar the effect was “a little uncanny”.
• Voice limits: he set his avatar to look like Mark Zuckerberg and found seeing the digital facsimile talk “quite weird”; he also could not get the voice to sound exactly like Zuckerberg.

That last point suggests voice customisation works from a preset menu and tone adjustments rather than free-form voice cloning, though Meta has not described the limits explicitly.

Wang’s demo gives a sense of the intended tone. In the video he jokes about his love of posting memes on X, which he tells his Muse may have gotten him in trouble with his boss; his Muse, a fluffy white dog, suggests adding disclaimers to memes that could be misread as Meta endorsing anything risqué. This is a company demo, not independent evidence of quality.

Why Meta is giving its AI agent a face

Many AI competitors already offer voice chat; Meta’s difference is the animated avatar, nicknamed Jolly by default, which users talk to as they ask questions or hand off tasks. Users can customise it to look more like a person, an animal or another character.

According to TechCrunch, Meta believes anthropomorphising Muse (giving it human-like traits) could make people feel more connected to their AI agent, and that live conversation could deepen that connection. Mashable makes a similar observation about the wider industry: a cute character is a way to make AI seem friendly rather than intimidating, “for better or worse”.

Meta’s announcement also pitches practical uses where speaking aloud matters:

“Some things are better rehearsed out loud. Prepare for a job interview, run through a presentation, or talk through a pitch — hearing a response and answering in the moment makes it feel more like the real thing.”

The product has traction to build on, though the figures vary by source and date. Mashable says Muse passed five million app downloads after its September launch. The Verge’s story page lists its own October headlines reporting that Muse hit 10 million US downloads and that 10 million people had set up Muse agents. The two outlets are likely describing different points in time, but neither number comes with an official breakdown in these sources.

What it means for developers building voice agents

For builders, this launch is a signal rather than a tool. Meta is not offering Muse Realtime Avatar or Muse Realtime Voice to developers, according to everything the sources report. What it does show is where a large consumer platform thinks voice agents are heading:

• Avatars as part of the voice interface. Meta is pairing a speech model with a separate avatar model that animates in sync and adjusts expression to tone. Teams building voice products may face user expectations for a visual presence, not just audio.
• Agents that act during the call. Muse is described as handling tasks on users’ behalf, and The Verge observed pauses while it looked things up mid-conversation. Managing tool calls without awkward silence remains a design problem for any real-time agent.
• Voice as personality. Hundreds of selectable voices, plus the ability to ask the agent to pick a tone, frames voice choice as personalisation rather than a single brand voice.
• Multi-device reach. Meta plans to carry the same agent across phone, web, Mac, glasses and a dedicated wearable, which raises the bar for consistency across surfaces.

Open questions and risks around Muse voice

Several gaps remain in what Meta has disclosed:

• No technical disclosure. There is no model card, latency figure, language list or benchmark in the coverage. All performance claims are Meta’s, apart from one reporter’s brief test.
• Reliability in voice. Mashable notes that voice mode “can quickly go sideways,” pointing to OpenAI’s experience when creator Husk repeatedly went viral with videos of ChatGPT’s voice mode behaving poorly; it says it remains to be seen whether Muse fares better.
• The uncanny effect. Human-like avatars can unsettle users, as The Verge found when its avatar resembled a real person. The sources do not say what rules, if any, Meta applies to avatars modelled on real people.
• Emotional attachment. Meta’s stated aim of making users feel more connected to their agent is the same quality that drew Mashable’s “for better or worse” caveat.
• Regional reach. Only the US is named; there is no timeline for other countries.

What to watch next for Muse

The next milestones are hardware. Meta says the voice and avatar experience will come “soon” to its AI glasses and to Muse Charm, which is expected to ship this year. How the avatar works on Muse Charm’s small screen, and how voice-only interaction works on glasses, will test whether the two models hold up away from a phone or laptop.

Beyond that, the questions for builders are whether Meta publishes technical details for Muse Realtime Avatar and Muse Realtime Voice, whether it ever opens them to developers, and whether independent testers find the real-time experience as smooth as Meta’s demos suggest. Expansion beyond the US would also be a clear signal of how far Meta intends to push Muse as a voice-first agent.