Defined in: realtime/types.ts

Configuration for a realtime session. Passed to the provider to configure model behavior, voice, tools, and VAD settings.

optional model: string;

Model to use for the session.

optional voice: string;

Voice to use for audio output.

optional instructions: string;

System instructions for the assistant.

optional tools: Array<RealtimeToolConfig>;

Tools available in the session.

optional vadMode: 'server' | 'semantic' | 'manual';

Voice activity detection mode.

optional vadConfig: VADConfig;

Detailed VAD configuration (threshold, padding, silence duration).

optional outputModalities: Array<'audio' | 'text'>;

Output modalities for responses (e.g., ['audio', 'text']).

optional temperature: number;

Temperature for generation (provider-specific range, e.g., 0.6-1.2 for OpenAI).

optional maxOutputTokens: number | 'inf';

Maximum number of tokens in a response.

optional semanticEagerness: 'low' | 'medium' | 'high';

Eagerness level for semantic VAD.

optional providerOptions: Record<string, any>;

Provider-specific options.