Echo (Universal Voice)

Architecture

Echo uses a WebSocket connection to stream audio data.

Client: Sends {"type": "audio_chunk", "data": "base64..."}.

Server (VAD): Voice Activity Detection determines end of speech.

Server (LLM): Transcribes -> Thinks -> Streams text response.

Server (TTS): Converts text to audio chunks -> Sends back to client.

To prevent abuse:

Max Duration: 60 seconds per turn.

Rate Limit: 50 turns per hour per user.

Cost Control: Admins can set a hard spend_limit (USD) per user in the database.