/**
 * pi-loom: Embedding Client
 *
 * Two-tier embedding strategy:
 *   1. API (primary): EMBED_API_KEY → OpenAI-compatible API (text-embedding-3-small, 1536d)
 *   2. Local (fallback): character n-gram TF vector when no API key is set (256d, zero deps)
 *
 * Configurable via env:
 *   EMBED_API_KEY / EMBED_API_BASE / EMBED_MODEL
 * Falls back to DEEPSEEK_API_KEY → OPENAI_API_KEY.
 *
 * For production-quality semantic search without an API key:
 *   npm install @xenova/transformers
 * Set EMBED_LOCAL_MODEL=all-MiniLM-L6-v2 (or any HF sentence-transformers model).
 */
/** Dimension of the embedding model. API default: 1536, local fallback: 256. */
export declare const EMBED_DIM: number;
/**
 * Generate embeddings for one or more texts via API.
 */
export declare function embedTextsAPI(texts: string[]): Promise<number[][]>;
/**
 * Generate embeddings for one or more texts.
 *
 * Priority:
 *   1. API (EMBED_API_KEY set) → OpenAI-compatible, 1536d
 *   2. Local Transformers.js (if installed) → all-MiniLM-L6-v2, 384d
 *   3. N-gram TF fallback → character trigrams, 256d
 *
 * Returns empty array if no embedding method is available.
 */
export declare function embedTexts(texts: string[]): Promise<number[][]>;
/** Shorthand for single text. */
export declare function embedText(text: string): Promise<number[]>;