Skip to content

Guides

Task-oriented pages for the subsystems around the core engine.

  • Turn-taking: Silero VAD + Smart Turn v3. When the bot answers, and when the user can cut in. The setting that most decides whether a bot feels good.
  • Services: swapping STT/LLM/TTS providers, tool calling, and speech-to-speech.
  • Audio: codecs, sample rates, noise reduction, background mixing, recording.
  • Telephony: phone calls over Twilio, Telnyx, Plivo and Exotel, plus DTMF and the idle watchdog.
  • RTVI: the event protocol that gives a client live transcriptions and speaking state.
  • Observability: observers, metrics and tracing, and which numbers actually matter for a voice agent.
  • Expressive conversation (alpha): the LLM hears how the user spoke and writes expression into its reply as tags, and the TTS performs them.
  • Development runner: run a bot behind one server that speaks WebRTC, LiveKit, telephony and WebSocket, with a session per call.