Avatar API
One runtime holds the whole character. VRM avatars load through one interface, and lip sync, expression, idle motion and gaze are handled for you. Swap the art without touching client code.
Enter dashboard
AI
Build one real-time agent. Give it a voice, a face, and the context to stay in the conversation. Now in invite-only beta.
Narya brings the agent and its avatar into one workspace without collapsing them into one thing. Configure instructions, tools, voice, and live presence; review each revision before your team takes it live.
One runtime holds the whole character. VRM avatars load through one interface, and lip sync, expression, idle motion and gaze are handled for you. Swap the art without touching client code.
Speech built for conversation, not narration. Audio starts streaming while the model is still writing, so the character answers inside the pause. Interruption and turn taking are handled by the API.
Ship a character without writing code. Pick the character, set the voice, write the persona, publish. You get three lines to paste into a website, a desktop app or a phone app.
Every session lands as structured data. You see how long people stayed, the turn where they walked away, and whether latency got in the way. Scoring runs on real conversations, not synthetic tests.
The agent owns the conversation. The avatar owns the live presence. Build and review them together while each layer keeps a clear job.
Set the instructions, model, voice, and tools that shape every turn. Publish a revision only after your team has reviewed exactly what changed.
Instructions helpful, concise Model gemini-3.5-flash Voice default companion Tools approved only Status ready to publish
Choose the character surface, voice, and stage used in a live session. Speech and expression stay synchronized while the agent listens, answers, and yields the turn.
Agent published revision Avatar selected character Voice conversation preset Transport WebRTC Status ready to join
Choose the character surface that fits the agent and the moment. The avatar carries expression, motion, and presence while the agent owns what is said.
One dashboard holds the agent revisions, integrations, organization access, and live-session setup needed for the invite-only beta.
Live Hostdraft—DRAFTCreate an agent, review its revision history, and publish it from the same focused surface your team uses to manage integrations and organization access.
LUMI, DEFAULT, ON A CALL
Pair the published agent with the avatar, stage, and voice used for a live session. The setup stays explicit: the agent owns behavior and the avatar owns presence.
Narya is currently invite-only. Request access and we will send a one-time invitation link when a beta place is available.
The invite opens the normal sign-in flow and creates your trial organization. Invitations and redemptions stay auditable.
Builders configure agents, integrations, voices, and live-session settings. Owners and admins manage organization access, credit, usage, and API keys.
They are two layers of one workspace. The agent owns conversational behavior; the avatar owns the live presence. A session can use them together.