import { type CompletionOptions, type LLMProvider, type ProviderCapabilities } from '@cortex/core';
export interface OllamaProviderOptions {
    host?: string;
    model?: string;
    embeddingModel?: string;
    numCtx?: number;
    numGpu?: number;
    timeoutMs?: number;
    keepAlive?: string;
}
export declare class OllamaProvider implements LLMProvider {
    readonly name = "ollama";
    readonly type: "local";
    private host;
    private model;
    private embeddingModel;
    private numCtx;
    private numGpu;
    private timeoutMs;
    private keepAlive;
    private streamInactivityTimeoutMs;
    readonly capabilities: ProviderCapabilities;
    constructor(options?: OllamaProviderOptions);
    getModel(): string;
    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>;
    ensureModel(): Promise<void>;
    listModels(): Promise<Array<{
        name: string;
        sizeBytes: number;
        modifiedAt: string;
    }>>;
    getHost(): string;
    getNumCtx(): number;
    getNumGpu(): number;
    getEmbeddingModel(): string;
    private mapError;
}
//# sourceMappingURL=ollama.d.ts.map