/**
 * OpenAI-compatible LLM client for Z.AI GLM models
 *
 * Provides a wrapper around the /v1/chat/completions endpoint
 * with support for both streaming and non-streaming responses.
 */
import type { ScampiConfig } from "../config/index.js";
/**
 * Role of a chat message participant
 */
export type ChatRole = "system" | "user" | "assistant";
/**
 * A single message in a chat conversation
 */
export interface ChatMessage {
    /** The role of the message author */
    role: ChatRole;
    /** The content of the message */
    content: string;
    /** Optional name for the participant */
    name?: string;
}
/**
 * Options for chat completion requests
 */
export interface ChatCompletionOptions {
    /** Sampling temperature (0-2). Higher = more random. Default: 1 */
    temperature?: number;
    /** Maximum tokens to generate. Default: model-specific */
    maxTokens?: number;
    /** Nucleus sampling probability. Default: 1 */
    topP?: number;
    /** Request timeout in milliseconds. Default: 30000 */
    timeout?: number;
    /** Whether to stream the response. Default: false */
    stream?: boolean;
    /** Stop sequences to end generation */
    stop?: string | string[];
}
/**
 * Request body for OpenAI-compatible chat completions API
 */
export interface ChatCompletionRequest {
    /** Model ID to use for completion */
    model: string;
    /** Array of messages in the conversation */
    messages: ChatMessage[];
    /** Sampling temperature */
    temperature?: number;
    /** Maximum tokens to generate */
    max_tokens?: number;
    /** Nucleus sampling probability */
    top_p?: number;
    /** Whether to stream the response */
    stream?: boolean;
    /** Stop sequences */
    stop?: string | string[];
}
/**
 * Token usage information from the API response
 */
export interface TokenUsage {
    /** Number of tokens in the prompt */
    prompt_tokens: number;
    /** Number of tokens in the completion */
    completion_tokens: number;
    /** Total tokens used */
    total_tokens: number;
}
/**
 * A single choice in the completion response
 */
export interface ChatCompletionChoice {
    /** Index of this choice */
    index: number;
    /** The generated message */
    message: ChatMessage;
    /** Reason the model stopped generating */
    finish_reason: "stop" | "length" | "content_filter" | null;
}
/**
 * Response from the OpenAI-compatible chat completions API
 */
export interface ChatCompletionResponse {
    /** Unique identifier for this completion */
    id: string;
    /** Object type (always "chat.completion") */
    object: "chat.completion";
    /** Unix timestamp of creation */
    created: number;
    /** Model used for this completion */
    model: string;
    /** Array of completion choices */
    choices: ChatCompletionChoice[];
    /** Token usage statistics */
    usage?: TokenUsage;
}
/**
 * A single chunk in a streaming response
 */
export interface ChatCompletionChunk {
    /** Unique identifier for this completion */
    id: string;
    /** Object type (always "chat.completion.chunk") */
    object: "chat.completion.chunk";
    /** Unix timestamp of creation */
    created: number;
    /** Model used for this completion */
    model: string;
    /** Array of delta choices */
    choices: ChatCompletionChunkChoice[];
}
/**
 * A single choice delta in a streaming chunk
 */
export interface ChatCompletionChunkChoice {
    /** Index of this choice */
    index: number;
    /** The delta content */
    delta: Partial<ChatMessage>;
    /** Reason the model stopped generating (only in final chunk) */
    finish_reason: "stop" | "length" | "content_filter" | null;
}
/**
 * Error response from the API
 */
export interface APIError {
    /** Error details */
    error: {
        /** Error message */
        message: string;
        /** Error type */
        type: string;
        /** Error code */
        code: string | null;
    };
}
/**
 * Custom error class for LLM client errors
 */
export declare class LLMClientError extends Error {
    /** HTTP status code if available */
    readonly status?: number;
    /** Error code from API */
    readonly code?: string;
    /** Whether the error is retryable */
    readonly retryable: boolean;
    constructor(message: string, options?: {
        status?: number;
        code?: string;
        retryable?: boolean;
    });
}
/**
 * OpenAI-compatible LLM client
 *
 * Provides methods for interacting with Z.AI's GLM models through
 * the OpenAI-compatible /v1/chat/completions endpoint.
 */
export declare class LLMClient {
    private readonly apiKey;
    private readonly baseUrl;
    private readonly defaultTimeout;
    /**
     * Create a new LLM client
     *
     * @param apiKey - API key for authentication
     * @param baseUrl - Base URL for the API (default: Z.AI endpoint)
     * @param defaultTimeout - Default timeout in milliseconds (default: 30000)
     */
    constructor(apiKey: string, baseUrl?: string, defaultTimeout?: number);
    /**
     * Make a chat completion request (non-streaming)
     *
     * @param messages - Array of chat messages
     * @param model - Model ID to use
     * @param options - Optional completion options
     * @returns Promise resolving to the completion response
     */
    complete(messages: ChatMessage[], model: string, options?: ChatCompletionOptions): Promise<ChatCompletionResponse>;
    /**
     * Make a streaming chat completion request
     *
     * @param messages - Array of chat messages
     * @param model - Model ID to use
     * @param options - Optional completion options
     * @returns AsyncGenerator yielding completion chunks
     */
    stream(messages: ChatMessage[], model: string, options?: ChatCompletionOptions): AsyncGenerator<ChatCompletionChunk, void, unknown>;
    /**
     * Collect all chunks from a streaming response into a complete response
     *
     * @param messages - Array of chat messages
     * @param model - Model ID to use
     * @param options - Optional completion options
     * @returns Promise resolving to the assembled response
     */
    streamToCompletion(messages: ChatMessage[], model: string, options?: ChatCompletionOptions): Promise<{
        content: string;
        finishReason: string | null;
    }>;
    /**
     * Handle error responses from the API
     *
     * @param response - The fetch Response object
     * @throws LLMClientError with appropriate details
     */
    private handleErrorResponse;
}
/**
 * Create an LLM client from Scampi configuration
 *
 * @param config - Scampi configuration object
 * @param baseUrl - Optional base URL override
 * @param timeout - Optional default timeout in milliseconds
 * @returns Configured LLMClient instance
 */
export declare function createClient(config: ScampiConfig, baseUrl?: string, timeout?: number): LLMClient;
//# sourceMappingURL=client.d.ts.map