/**
 * Local embedding provider using @huggingface/transformers (transformers.js v4).
 *
 * Implements the EmbeddingProvider interface for brain memory vector search.
 * Uses all-MiniLM-L6-v2 (22MB, 384 dimensions) — matches the brain_embeddings
 * vec0 table schema. Model downloads on first call and is cached locally by
 * the transformers library.
 *
 * @epic T134
 * @task T136
 * @why Ship vector search out-of-the-box without external API keys
 * @what Local embedding provider using @huggingface/transformers all-MiniLM-L6-v2
 * @remarks Brain embeddings are a FIRST-CLASS CLEO feature — the transformers
 *   package is a regular dependency of `@cleocode/core`, not optional.
 *   Migrated from `@xenova/transformers` v2 to `@huggingface/transformers`
 *   v4 (upstream rename, same author) which drops the deprecated
 *   `prebuild-install` transitive via `sharp@0.34+`.
 */
import type { EmbeddingProvider } from './brain-embedding.js';
/**
 * Local embedding provider backed by @huggingface/transformers.
 *
 * Produces 384-dimension Float32Array vectors compatible with the
 * brain_embeddings vec0 table. The model is downloaded on first use
 * and cached locally by the transformers library.
 *
 * Use {@link initDefaultProvider} (in brain-embedding.ts) to register an
 * instance when brain.embedding.enabled=true and
 * brain.embedding.provider='local'.
 */
export declare class LocalEmbeddingProvider implements EmbeddingProvider {
    /** Number of dimensions produced — must match brain_embeddings vec0 table. */
    readonly dimensions = 384;
    /**
     * Whether the pipeline has been successfully initialized and is ready to produce embeddings.
     */
    isAvailable(): boolean;
    /**
     * Convert a single text string into a 384-dimension float vector.
     * Triggers model download on first call if not already cached.
     *
     * @param text - The text to embed.
     * @returns A Float32Array of length 384.
     */
    embed(text: string): Promise<Float32Array>;
    /**
     * Embed multiple texts in sequence, reusing the cached pipeline.
     *
     * @param texts - Array of text strings to embed.
     * @returns Array of Float32Array vectors, one per input text.
     */
    embedBatch(texts: string[]): Promise<Float32Array[]>;
}
/**
 * Get or create the shared LocalEmbeddingProvider singleton.
 *
 * @returns The shared LocalEmbeddingProvider instance.
 */
export declare function getLocalEmbeddingProvider(): LocalEmbeddingProvider;
//# sourceMappingURL=embedding-local.d.ts.map