Real-time AI avatars,
dropped into your own pipeline.

Talking-head video generation with sub-second latency — bring your own STT, LLM, and TTS, we handle the face. Ship it inside your existing Pipecat or LiveKit stack.

Real-time, not pre-rendered

Sub-second lip-synced video generated live from streaming audio — not a queued render job.

Bring your own stack

Plug into whatever STT, LLM, and TTS you already use — Deepgram, OpenAI, Cartesia, or anything else. We only handle the avatar video.

Deploy in your own pipeline

One package (koi-avatar) drops straight into a Pipecat or Pipecat + LiveKit pipeline you already run — no separate hosting to manage.

39 languages

Every avatar speaks and listens in any of 39 verified-supported languages, picked per session.