Local developer demo

Hear a voice agent change tone in real time.

Run the sidecar on your machine. Talk into your mic and escalate naturally. When tone shifts, ProsodyAI steers the agent — the next utterance softens.

Run it

pip install -r demo/requirements.txt
# OPENAI_API_KEY + PROSODY_API_KEY in .env at repo root
python demo/sidecar.py

Needs an OpenAI API key. ProsodyAI picks up PROSODY_API_KEY from the repo root .env.

What it does

mic audio (16 kHz PCM)
    │
    ├──► voice agent (OpenAI Realtime, LiveKit, Vapi, …)
    │
    └──► ProsodyAI WebSocket  →  ~2s chunks
                                      │
                                      ├── tone + speaker turns
                                      ├── transcript (parallel)
                                      └── agent steering cues

Each chunk returns delivery signals, transcript, speaker labels, and steering cues your agent can apply to the next turn.

Recording tip

Wear headphones when recording so speaker output does not bleed into the mic and pollute the tone read.

Full integration contract: Docs.