import type { LlmAdapter, AdapterRequest, AdapterResponse, AdapterStreamChunk } from '@mast-ai/core';
import type { ReasoningEffort, UsageMetadata } from './OpenAIChatCompletionsAdapter.js';
/** Reasoning-summary verbosity that gates what the model emits as `thinking` events. */
export type ReasoningSummary = 'auto' | 'concise' | 'detailed';
/** Per-request defaults applied to every call made by an {@link OpenAIResponsesAdapter}. */
export interface OpenAIResponsesAdapterDefaults {
    /** Forwarded as `reasoning.effort` on every request. Reasoning models honour this. */
    reasoningEffort?: ReasoningEffort;
    /**
     * Forwarded as `reasoning.summary` on every request. Required to receive
     * reasoning summary deltas as `thinking` events. Defaults to `"auto"` so
     * reasoning models surface thinking content without extra configuration.
     * Set to `false` to disable.
     */
    reasoningSummary?: ReasoningSummary | false;
}
/**
 * {@link LlmAdapter} implementation backed by the OpenAI Responses API
 * (`/v1/responses`).
 *
 * Unlike {@link OpenAIChatCompletionsAdapter}, this adapter surfaces reasoning
 * content from reasoning models — `gpt-5`, `o-series` — as `thinking` events.
 * The adapter runs statelessly (`store: false`); the caller's `Conversation`
 * history is replayed on every request.
 */
export declare class OpenAIResponsesAdapter implements LlmAdapter {
    private client;
    private modelName;
    private onUsageUpdate?;
    private defaults;
    /**
     * @param apiKey - OpenAI API key.
     * @param modelName - Model identifier (defaults to `"gpt-4o-mini"`). Any
     *   Responses-API-compatible model works, including reasoning models.
     * @param onUsageUpdate - Optional callback invoked with token-usage data after each response.
     * @param defaults - Provider-specific defaults applied to every request.
     */
    constructor(apiKey: string, modelName?: string, onUsageUpdate?: (usage: UsageMetadata) => void, defaults?: OpenAIResponsesAdapterDefaults);
    private buildBaseParams;
    /** {@inheritDoc LlmAdapter.generate} */
    generate(request: AdapterRequest): Promise<AdapterResponse>;
    /** {@inheritDoc LlmAdapter.generateStream} */
    generateStream(request: AdapterRequest): AsyncIterable<AdapterStreamChunk>;
}