RealtimeInput
Input streamed to a live model outside the turn-by-turn content exchange.
Realtime input is not a turn: media streams in continuously and the model decides when the user has finished, unless activity detection is disabled and the caller marks the boundaries with ActivityStart and ActivityEnd; realtime text has no arm here and is sent as content.
Audio and Video are separate arms because the Kotlin GenAI SDK has no combined media parameter, so the caller chooses which it is sending.
Inheritors
Types
Marks the end of user activity, under the same condition as ActivityStart.
Marks the start of user activity.
A chunk of audio, conventionally 16-bit mono PCM at 16kHz (audio/pcm;rate=16000).
Marks the end of the audio stream, forcing the model to flush what it has.
A frame of video, conventionally a JPEG image.