studyArena/lib/ai/thinking-context.ts

/**
 * Async-context carrier for per-request ThinkingConfig.
 *
 * callLLM / streamLLM wrap each AI SDK call in thinkingContext.run()
 * so that the custom fetch wrapper in providers.ts can read the
 * current thinking preference and inject vendor-specific body params.
 *
 * IMPORTANT: This module uses node:async_hooks which is server-only.
 * providers.ts must NOT import this module directly (it's also used
 * on the client via settings.ts). Instead, providers.ts reads the
 * context via globalThis.__thinkingContext, which is set here at
 * module load time and guaranteed to be available before any fetch
 * wrapper runs.
 */
import { AsyncLocalStorage } from 'node:async_hooks';
import type { ThinkingConfig } from '@/lib/types/provider';

export const thinkingContext = new AsyncLocalStorage<ThinkingConfig | undefined>();

// Expose on globalThis so providers.ts can access the store without
// importing this module (which would pull node:async_hooks into the
// client bundle via the settings.ts → providers.ts import chain).
(globalThis as Record<string, unknown>).__thinkingContext = thinkingContext;