import { RealtimeAgent, RealtimeSession, RealtimeAgentConfiguration } from '@openai/agents-realtime';
import { RealtimeProvider, RealtimeProviderConfig, RealtimeProviderCallbacks, RealtimeMessage, AudioFormat } from './RealtimeProvider';
interface WebSocketRealtimeProviderBaseOptions {
    voiceMap?: Record<string, string>;
}
/**
 * Shared base for OpenAI-compatible WebSocket realtime providers.
 *
 * This covers the common pieces used by both `vowel-prime` and Grok:
 * - OpenAI Agents SDK tool definitions
 * - manual audio streaming over WebSocket transports
 * - queued text/image messages until the session is fully ready
 * - tool response resolution
 * - session.updateAgent() based instruction updates
 */
export declare abstract class WebSocketRealtimeProviderBase extends RealtimeProvider {
    protected agent: RealtimeAgent | null;
    protected session: RealtimeSession | null;
    protected isConnected: boolean;
    protected originalAgentConfig: RealtimeAgentConfiguration | null;
    protected pendingToolExecutions: Map<string, (result: any) => void>;
    protected messageQueue: Array<{
        type: 'text' | 'image';
        data: string;
    }>;
    protected isFullyReady: boolean;
    protected sessionCreatedResolver: (() => void) | null;
    protected voiceMap: Record<string, string>;
    private pendingToolContinuationTimer;
    private hasLoggedAudio;
    constructor(config: RealtimeProviderConfig, callbacks: RealtimeProviderCallbacks, options?: WebSocketRealtimeProviderBaseOptions);
    getAudioFormat(): AudioFormat;
    getInputAudioFormat(): AudioFormat;
    getOutputAudioFormat(): AudioFormat;
    handlesAudioInternally(): boolean;
    protected getLogLabel(): string;
    protected getDefaultTurnDetectionMode(): 'client_vad' | 'server_vad' | 'disabled';
    protected getResolvedTurnDetectionMode(): 'client_vad' | 'server_vad' | 'disabled';
    protected getConfiguredAudioFormat(direction: 'input' | 'output'): Partial<AudioFormat>;
    protected mapVoiceName(voice: string): string;
    protected createSDKTools(): any[];
    disconnect(): Promise<void>;
    sendAudio(audioData: string, _format?: AudioFormat): void;
    commitAudio(concatenatedAudioBuffer: ArrayBuffer): void;
    sendText(text: string): void;
    sendImage(imageUrl: string): void;
    interrupt(): void;
    sendToolResponse(toolCallId: string, toolName: string, result: any): void;
    private clearPendingToolContinuationTimer;
    protected normalizeMessage(_rawMessage: any): RealtimeMessage | null;
    sendSessionUpdate(updates: {
        instructions?: string;
    }): void;
    protected updateAgentAsync(newInstructions: string): Promise<void>;
    /**
     * Set up common SDK event listeners for all WebSocket providers.
     * This consolidates event handling that was previously duplicated across
     * OpenAI, VowelPrime, and Grok providers.
     *
     * Providers should call this method during their connect() after session is created.
     */
    protected setupSDKSessionEventListeners(): void;
}
export {};
//# sourceMappingURL=WebSocketRealtimeProviderBase.d.ts.map