ten-framework

Why we included this project

Building a voice agent usually means wiring a speech-to-text model, a language model, and a text-to-speech service together yourself. TEN instead provides a runtime that handles that coordination for you, and its extension-based design lets you add memory, voice activity detection, turn detection, or streaming over WebSocket and RTC without reworking the whole pipeline. The bundled examples cover a lot of ground, from a low-latency voice assistant to speaker diarization, lip-synced avatars, and phone calls over SIP. If you need a production base for a multimodal agent that actually talks and listens, rather than another demo, TEN is a practical place to start.