Voice

A voice interface turns the app's agent into a live, phone-quality conversation. The user speaks, the agent speaks back in sub-second, interruptible turns, and calls the app's methods mid-sentence, as the signed-in user. The platform runs the realtime pipeline (one speech-to-speech model, or a cascaded transcribe-think-speak stack), the audio transport, and the telephony, so none of it is yours to stand up or operate.

This section is the backend surface that conversation reaches. voice.call places an outbound call, dialing a number and connecting the person to this same agent, with its persona, engine, and tools. Inside the methods the agent runs as tools, the session context says which call is on the line: its id, whether it came in over the web or a phone, and the numbers if it is a phone call. That lets a tool correlate its result back to the call, and answer differently depending on whether the user is on a screen or a phone. The browser client and the phone-number and call-log commands live in the Interface and Admin SDKs.