import type { AssistantMessage } from "../types.js";
export interface ContextOverflowOptions {
    /** Maximum output tokens requested for the exhausted response. */
    maxOutputTokens?: number;
    /** Estimated request-input tokens when the provider omitted usage. */
    estimatedInputTokens?: number;
}
/** Whether the agent exhausted its bounded retries at the configured output limit. */
export declare function isOutputLimitExhaustion(message: AssistantMessage): boolean;
/**
 * Check if an assistant message represents a context overflow error.
 *
 * This handles three cases:
 * 1. Error-based overflow: Most providers return stopReason "error" with a
 *    specific error message pattern.
 * 2. Context-constrained truncation: dreb exhausted its length retries and the
 *    provider signal or input-plus-configured-output budget indicates that the
 *    response may not have fit in the remaining context window.
 * 3. Silent overflow: Some providers accept requests beyond the configured window
 *    and return successfully (the server's hard limit may be higher). For these,
 *    we compare input + cacheRead + cacheWrite against the configured window;
 *    output tokens are not part of the input size.
 *
 * ## Reliability by Provider
 *
 * **Reliable detection (returns error with detectable message):**
 * - Anthropic: "prompt is too long: X tokens > Y maximum"
 * - OpenAI (Completions & Responses): "exceeds the context window"
 * - Google Gemini: "input token count exceeds the maximum"
 * - xAI (Grok): "maximum prompt length is X but request contains Y"
 * - Groq: "reduce the length of the messages"
 * - Cerebras: 400/413 status code (no body)
 * - Mistral: "Prompt contains X tokens ... too large for model with Y maximum context length"
 * - OpenRouter (all backends): "maximum context length is X tokens"
 * - llama.cpp: "exceeds the available context size"
 * - LM Studio: "greater than the context length"
 * - Kimi For Coding: "exceeded model token limit: X (requested: Y)"
 *
 * **Unreliable detection:**
 * - z.ai: Sometimes accepts overflow silently, sometimes returns rate limit errors.
 * - GitHub Copilot: May accept input beyond a conservative configured window.
 *   Pass contextWindow to detect these cases using input + cacheRead + cacheWrite.
 * - Ollama: Silently truncates input without error. Cannot be detected via this function.
 *   The response will have usage.input < expected, but we don't know the expected value.
 *
 * ## Custom Providers
 *
 * If you've added custom models via settings.json, this function may not detect
 * overflow errors from those providers. To add support:
 *
 * 1. Send a request that exceeds the model's context window
 * 2. Check the errorMessage in the response
 * 3. Create a regex pattern that matches the error
 * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or
 *    check the errorMessage yourself before calling this function
 *
 * @param message - The assistant message to check
 * @param contextWindow - Optional configured window for usage-based overflow detection
 * @param options - Output limit and input estimate used only after bounded length retries are exhausted
 * @returns true if the message indicates a context overflow
 */
export declare function isContextOverflow(message: AssistantMessage, contextWindow?: number, options?: ContextOverflowOptions): boolean;
/**
 * Get the overflow patterns for testing purposes.
 */
export declare function getOverflowPatterns(): RegExp[];
//# sourceMappingURL=overflow.d.ts.map