import { type CompletionOptions, type LLMProvider, type ProviderCapabilities } from '@cortex/core';
export interface OpenAICompatibleProviderOptions {
    baseUrl: string;
    apiKey?: string;
    primaryModel?: string;
    fastModel?: string;
    embeddingModel?: string;
    timeoutMs?: number;
    maxRetries?: number;
}
export declare class OpenAICompatibleProvider implements LLMProvider {
    readonly name = "openai-compatible";
    readonly type: "cloud";
    private client;
    private primaryModel;
    private fastModel;
    private embeddingModel;
    private isGemini;
    readonly capabilities: ProviderCapabilities;
    constructor(options: OpenAICompatibleProviderOptions);
    getModel(preference?: 'primary' | 'fast'): string;
    /** Gemini uses max_completion_tokens; others use max_tokens */
    private tokenLimitParams;
    complete(prompt: string, options?: CompletionOptions): Promise<string>;
    completeWithSystem(systemPrompt: string | undefined, userPrompt: string, options?: CompletionOptions, modelPreference?: 'primary' | 'fast'): Promise<{
        content: string;
        inputTokens: number;
        outputTokens: number;
        model: string;
    }>;
    completeStructured<T>(prompt: string, _schema: Record<string, unknown>, options?: CompletionOptions): Promise<T>;
    stream(prompt: string, options?: CompletionOptions): AsyncGenerator<string>;
    streamWithSystem(systemPrompt: string | undefined, userPrompt: string, options?: CompletionOptions, modelPreference?: 'primary' | 'fast'): AsyncGenerator<string, {
        inputTokens: number;
        outputTokens: number;
        model: string;
    }>;
    embed(texts: string[]): Promise<Float32Array[]>;
    isAvailable(): Promise<boolean>;
    private mapError;
}
//# sourceMappingURL=openai-compatible.d.ts.map