import type { ChatResponse, LLMToolDef, ONIModelMessage } from "../../models/types.js";
import type { AgentLoopConfig, LoopMessage } from "../types.js";
/** Check if an error is retryable (rate limits, transient server errors). */
export declare function isRetryableError(err: unknown): boolean;
/** Get retry delay with exponential backoff. */
export declare function getRetryDelay(err: unknown, attempt: number): number;
export interface InferenceResult {
    response: ChatResponse | null;
    /** Yielded messages during retries */
    events: LoopMessage[];
    succeeded: boolean;
    lastError: unknown;
}
/**
 * Run LLM inference with retry/timeout/header-aware backoff.
 * Returns the response and any yield-able events (retry messages).
 */
export declare function runInference(messages: ONIModelMessage[], llmTools: LLMToolDef[], systemPrompt: string, config: AgentLoopConfig, sessionId: string, turn: number): Promise<InferenceResult>;
//# sourceMappingURL=inference.d.ts.map