Simli is a speech-to-video API that gives a real-time AI agent a face. You send audio and receive a lip-synced video avatar over WebRTC, and the platform slots in as the final stage of a voice pipeline after speech-to-text, the LLM, and text-to-speech. Typical uses include sales assistants, mock interviews, language training, coaching, customer-success training, and historical characters.
Three integration paths are offered: LiveKit, which the platform recommends, the Pipecat framework, and Simli's own SDKs for JavaScript and Python when you need finer control or a custom WebRTC client. Sessions can run peer-to-peer over WebRTC, through LiveKit, or as Simli Auto sessions that bundle a preconfigured or custom pipeline. Faces are created from your own footage or picked from default faces by ID, with newer Trinity face models and a Gaussian-based visual model for higher-resolution, expressive avatars.
Features
- Real-time avatar rendering: streams lip-synced video for incoming audio over WebRTC
- Face management: create custom avatars, list, delete, and check generation status, or use default faces by faceID; Trinity and Legacy model generation
- Session types: peer-to-peer WebRTC, LiveKit-backed WebRTC, and Simli Auto sessions with preconfigured or custom pipelines
- Compose session tokens: temporary tokens and ICE server retrieval for browser clients
- Media generation: static video from audio, HLS streaming segments, and MP4 export
- Session reporting: session history and active-session counts via the API
- SDKs: JavaScript and Python SDKs plus OpenAPI and AsyncAPI specifications
- Integrations: LiveKit agents, Pipecat, and starter repositories for OpenAI and ElevenLabs
- Emotive faces: a Gaussian-model face generation option for high-resolution, expressive avatars
