ConnectionScript\_Human

false makes the persona uninterruptible: the bot cannot barge in on it, its speech plays out even while the bot talks (Gemini activityHandling: NO_INTERRUPTION, OpenAI interrupt_response: false). Use it for an answering machine whose greeting must overlap the bot's intro. Default true: normal barge-in.

Audio-to-audio model, provider-routed by prefix: openai/<model> (e.g. openai/gpt-realtime-2.1-mini) or gemini/<model> (e.g. gemini/gemini-3.1-flash-live-preview). An unprefixed value routes to OpenAI. Defaults to gemini/gemini-3.1-flash-live-preview.

Stage direction the persona receives for its first turn when speakFirst is set.

Make the emulated party talk as soon as the line opens (a callee saying "hello?", an answering machine playing its greeting) instead of waiting for the bot to speak. The bot's audio is withheld from the persona until that first utterance has played, so the bot's intro cannot cut it off. Without this the persona only ever answers the bot.

Persona tools: finish hangs up, dtmf sends touch-tone digits, tone plays an audible beep into the call (e.g. the voicemail beep after "leave your message after the tone") that the app's tone_detector / #getTone() picks up. wait lets the persona stay silent for a time it chooses and then continue (hold, a pause before hanging up): without it a persona that goes quiet gets no further turn until the app speaks. finish, dtmf and wait are on unless set to false; tone is off unless enabled.