import { AIRequest, AIResponse, StructuredOutput } from './types';
export declare const DEFAULT_LOCAL_MODEL = "onnx-community/Qwen2.5-0.5B-Instruct";
export declare const DEFAULT_LOCAL_MODEL_WARNING_TOKENS = 2048;
export interface MemorySummaryConfig {
    enabled: boolean;
    maxTokens: number;
    targetTokens: number;
    summaryModel: string;
}
export interface MemoryTrimResult {
    text: string;
    summarized: boolean;
    tokensBefore: number;
    tokensAfter: number;
    summaryModel?: string;
    usage?: AIResponse['usage'];
    error?: string;
}
export interface VideoGenerationRequest {
    model: string;
    prompt: string;
    images?: string[];
    ratio?: string;
    duration?: number;
    resolution?: string;
    negativePrompt?: string;
    audio?: boolean;
    task?: string;
    pollInterval?: number;
}
export declare class AIRuntime {
    private _client;
    private _openAIClient;
    private localPipelines;
    private localTokenizers;
    private conversationHistory;
    private embeddingCache;
    private memorySummaryConfigs;
    private memoryTrimLocks;
    constructor();
    private isGPTModel;
    private isGPTImageModel;
    private isLocalModel;
    private resolveLocalModelName;
    private getLocalCacheDirectory;
    private isLocalModelCached;
    private getLocalModelDirectory;
    private getLocalPipeline;
    private getLocalTokenizer;
    private countEncodedTokens;
    private getLocalWarningThreshold;
    private extractLocalText;
    private normalizeLocalTools;
    private normalizeLocalToolCall;
    private parseLocalToolCallText;
    private extractLocalToolCall;
    private callLocalModel;
    private mapThinkingEffort;
    private get openAIClient();
    private postOpenAIResponses;
    private streamOpenAIResponses;
    private extractOpenAIToolCall;
    private extractOpenAIText;
    private normalizeOpenAITools;
    private normalizeGeminiTools;
    private resolveOpenAIImageParts;
    private resolveOpenAIImageSize;
    private callGPTImageModel;
    private callGPTModel;
    private normalizeModelName;
    private get client();
    countTokens(model: string, contents: string): Promise<number>;
    callVideo(request: VideoGenerationRequest): Promise<string>;
    private getCacheFile;
    private readCache;
    private writeCache;
    private computeCacheHash;
    private resolveImageParts;
    synthesizeSpeech(text: string, voice?: string, model?: string): Promise<string>;
    transcribeSpeech(audioData: string, mimeType: string, language?: string, model?: string): Promise<string>;
    callModel(request: AIRequest): Promise<AIResponse>;
    parseStructuredOutput(response: string, schema: Record<string, any>): Promise<StructuredOutput>;
    initializeMemory(memoryId: string, initialValue: string): void;
    appendToMemory(memoryId: string, content: string): void;
    getMemory(memoryId: string): string;
    updateMemory(memoryId: string, content: string): void;
    private positiveInteger;
    private defaultMemorySummaryConfig;
    configureMemorySummary(memoryId: string, options: Partial<MemorySummaryConfig>): MemorySummaryConfig;
    getMemorySummaryConfig(memoryId: string): MemorySummaryConfig;
    estimateTokens(text: string): number;
    embedText(text: string): Promise<number[]>;
    searchMemory(memoryId: string, query: string, topK?: number): Promise<Array<{
        text: string;
        score: number;
    }>>;
    autoTrimMemory(memoryId: string): Promise<MemoryTrimResult>;
    trimMemory(memoryId: string, maxTokens?: number): Promise<string>;
    trimMemoryDetailed(memoryId: string, config: MemorySummaryConfig): Promise<MemoryTrimResult>;
    private performMemoryTrim;
}
export declare const aiRuntime: AIRuntime;
//# sourceMappingURL=ai-runtime.d.ts.map