Architecture

The full path of a single turn.

Six stages, wired by the router. Barge-in loops the last stage back to the first.

Wake loop
openWakeWord · barge-in
Router
VoxRouter
STT
faster-whisper
Agent chain
A2A + LLM · MCP · memory
TTS chain
edge · minimax · 11labs · openai
Audio
cached or streamed
Abstract illustration of one VoxRouter voice turn traveling from microphone through the router, speech recognition, agent, and speech synthesis before returning as audio.
Turn anatomy Every stage is observable, replaceable, and able to fall through without losing the turn.
barge-in loops back to the wake loop  ·  every stage reports its provider in the response so you know exactly what served the turn.
As text
README.md
microphone/wake loop (openWakeWord, barge-in)
-> VoxRouter
-> faster-whisper /v1/audio/transcriptions
-> chat backend chain: A2A peer and/or OpenAI-compatible LLM
(+ MCP tools, per-agent conversation memory)
-> TTS chain: Edge (free) | MiniMax | ElevenLabs | OpenAI-compatible
-> audio response (cached file or streamed as it synthesizes)

Ready to route your own voice stack?