RealtimeVoiceConfig

data class RealtimeVoiceConfig(val instructions: String? = null, val voice: String? = null, val modalities: Set<RealtimeModality> = setOf(RealtimeModality.AUDIO, RealtimeModality.TEXT), val audioFormat: RealtimeAudioFormat = RealtimeAudioFormat.PCM16, val turnDetection: VadConfig = VadConfig.Server(), val tools: List<LlmTool> = emptyList(), val toolChoice: ToolChoice = ToolChoice.Auto, val temperature: Double? = null, val maxResponseOutputTokens: Int? = null, val inputAudioTranscriptionModel: String? = "whisper-1", val providerExtras: Map<String, String> = emptyMap())

Vendor-neutral configuration for a realtime voice session.

Maps onto a provider's session-config message (for OpenAI Realtime, the initial session.update event). Only the cross-port subset is surfaced; provider-specific knobs go through providerExtras.

Constructors

Link copied to clipboard
constructor(instructions: String? = null, voice: String? = null, modalities: Set<RealtimeModality> = setOf(RealtimeModality.AUDIO, RealtimeModality.TEXT), audioFormat: RealtimeAudioFormat = RealtimeAudioFormat.PCM16, turnDetection: VadConfig = VadConfig.Server(), tools: List<LlmTool> = emptyList(), toolChoice: ToolChoice = ToolChoice.Auto, temperature: Double? = null, maxResponseOutputTokens: Int? = null, inputAudioTranscriptionModel: String? = "whisper-1", providerExtras: Map<String, String> = emptyMap())

Properties

Link copied to clipboard

Sample format used for ingress and egress audio.

Link copied to clipboard

System prompt analogue for the session.

Link copied to clipboard

Cap on tokens per assistant response.

Link copied to clipboard

Output modalities the model should produce. Defaults to [audio, text].

Link copied to clipboard

Escape hatch for provider-specific options; merged verbatim into the gateway's session-update payload.

Link copied to clipboard

Sampling temperature, when supported.

Link copied to clipboard

Provider-neutral tool-choice hint.

Link copied to clipboard

Tools the session may call.

Link copied to clipboard

Voice-activity-detection strategy.

Link copied to clipboard

Provider voice name (e.g. alloy, verse, shimmer).