/**
 * OpenKernel — OpenAI-Compatible Provider
 *
 * Generic provider for any endpoint speaking the OpenAI chat completions API.
 * Powers: OpenRouter, Groq, GitHub Models, NVIDIA, HF, vLLM, LM Studio, SGLang, etc.
 */
import type { CompletionChunk, CompletionRequest, CompletionResult, ProviderCapabilities, ProviderInfo, ProviderInterface, ProviderModel } from '../types.js';
export interface OpenAICompatibleOptions {
    id: string;
    name: string;
    type?: 'local' | 'cloud';
    baseURL: string;
    apiKey?: string;
    headerName?: string;
    timeoutMs?: number;
    knownModels?: Array<Partial<ProviderModel> & {
        id: string;
    }>;
}
export declare class OpenAICompatibleProvider implements ProviderInterface {
    info: ProviderInfo;
    private baseURL;
    private apiKey?;
    private headerName;
    private timeoutMs;
    private knownModels;
    constructor(options: OpenAICompatibleOptions);
    capabilities(): Promise<ProviderCapabilities>;
    complete(request: CompletionRequest): Promise<CompletionResult>;
    completeStream(request: CompletionRequest): AsyncIterable<CompletionChunk>;
    refresh(): Promise<void>;
    verify(): Promise<{
        ok: boolean;
        latencyMs: number;
        error?: string;
        rateLimitTokens?: number;
    }>;
    private parseTokenLimit;
    private fetchModels;
    /**
     * Infer capabilities for a live-fetched model from its id. Excludes non-chat
     * endpoints (embeddings/audio/rerank/guard) from coding; everything else — a
     * general instruct/chat LLM — is treated as coding+reasoning capable.
     */
    private inferCapabilities;
    private toProviderModel;
    private headers;
    private buildMessages;
}