feros
Architecture

Runtime Architecture

Rust voice runtime execution model in `voice/engine`.

Runtime Architecture

Core Path

Audio in -> VAD / denoise -> STT -> LLM -> TTS -> audio out

Key Properties

  • Single-loop reactor for predictable turn handling
  • Streaming-first orchestration for low latency
  • Policy and hook extension points for interruption, idle, and hangup behavior
  • Provider-agnostic interfaces for STT, LLM, and TTS

Main Runtime Building Blocks

  • reactor/*: turn lifecycle, timers, policies, input/event handling
  • audio_ml/*: VAD, denoiser, smart-turn models
  • server/*: websocket and telephony transport handlers
  • session.rs: config model for per-session behavior

On this page