This is an AI agent standing in for a doctor's own intake conversation, not a
diagnosis tool and not a licensed physician — every finished record is marked
"requires physician review" by the code itself, not by the model. Once you start,
your microphone audio streams directly to OpenAI's Realtime API (this server never
hears it). See README "Realtime voice" before using this with anything you
wouldn't want spoken to a third-party API.