import type { CallOptions, ILlm, LlmError, LlmResponse, LlmStreamChunk, LlmTool, LlmUsage, Message, Result } from '@mcp-abap-adt/llm-agent';
/**
 * An ILlm decorator that reports token usage after every `chat` or `streamChat`
 * call via an injected `logUsage` callback. This is the authoritative path for
 * per-role usage logging in the Stepper runtime — it ensures every role's LLM
 * call is captured in the request-scoped byComponent breakdown regardless of
 * whether the role is the executor, planner, reviewer, or finalizer.
 *
 * Usage:
 *   const logged = new LoggingLlm(innerLlm, (u, d) =>
 *     requestLogger.logLlmCall({ component: 'planner', model, ...u, durationMs: d })
 *   );
 *
 * For `streamChat`, usage is accumulated from stream chunks and logged once the
 * stream ends.
 */
export declare class LoggingLlm implements ILlm {
    private readonly inner;
    /** Called once per LLM invocation with the accumulated usage and duration. */
    private readonly logUsage;
    constructor(inner: ILlm, 
    /** Called once per LLM invocation with the accumulated usage and duration. */
    logUsage: (usage: LlmUsage, durationMs: number) => void);
    get model(): string | undefined;
    chat(messages: Message[], tools?: LlmTool[], options?: CallOptions): Promise<Result<LlmResponse, LlmError>>;
    streamChat(messages: Message[], tools?: LlmTool[], options?: CallOptions): AsyncIterable<Result<LlmStreamChunk, LlmError>>;
    healthCheck?(options?: CallOptions): Promise<Result<boolean, LlmError>>;
    getModels?(options?: CallOptions): Promise<Result<import('@mcp-abap-adt/llm-agent').IModelInfo[], LlmError>>;
}
//# sourceMappingURL=logging-llm.d.ts.map