import { GthConfig } from '#src/config.js';
import { GthAgentInterface, GthCommand, StatusLevel } from '#src/core/types.js';
import type { Message } from '#src/modules/types.js';
import { RunnableConfig } from '@langchain/core/runnables';
import { IterableReadableStream } from '@langchain/core/utils/stream';
import { BaseCheckpointSaver } from '@langchain/langgraph';
import type { Connection } from '@langchain/mcp-adapters';
import { MultiServerMCPClient } from '@langchain/mcp-adapters';
export type StatusUpdateCallback = (level: StatusLevel, message: string) => void;
export declare class GthLangChainAgent implements GthAgentInterface {
    private statusUpdate;
    private mcpClient;
    private agent;
    private config;
    constructor(statusUpdate: StatusUpdateCallback);
    init(command: GthCommand | undefined, configIn: GthConfig, checkpointSaver?: BaseCheckpointSaver | undefined): Promise<void>;
    /**
     * Invoke LLM with a message and runnable config.
     * For streaming use {@link #stream} method, streaming is preferred if model API supports it.
     * Please note that this when tools are involved, this method will anyway do multiple LLM
     * calls within LangChain dependency.
     */
    invoke(messages: Message[], runConfig: RunnableConfig): Promise<string>;
    /**
     * Induce LLM to stream AI messages with a user message and runnable config.
     * When stream is not appropriate use {@link invoke}.
     */
    stream(messages: Message[], runConfig: RunnableConfig): Promise<IterableReadableStream<string>>;
    getMCPClient(): MultiServerMCPClient | null;
    cleanup(): Promise<void>;
    getEffectiveConfig(config: GthConfig, command: GthCommand | undefined): GthConfig;
    /**
     * Extract and flatten tools from toolkits
     */
    private extractAndFlattenTools;
    protected getDefaultMcpServers(): Record<string, Connection>;
    protected getMcpClient(config: GthConfig): Promise<MultiServerMCPClient | null>;
}