import { type Embedder } from "./embeddings.js";
import { type SearchHit, type SearchOptions } from "./search.js";
/** Cosine similarity of two equal-ish-length vectors (0 when either is zero). */
export declare function cosineSimilarity(a: Float32Array | number[], b: Float32Array | number[]): number;
export interface SemanticOptions extends SearchOptions {
    /** Embedder for the query (defaults to OpenAI). Inject a fake in tests. */
    embedder?: Embedder;
}
/** Rank stored embeddings against a query vector (brute-force cosine), one hit per session. */
export declare function vectorSearchByEmbedding(queryVec: number[], opts?: SemanticOptions): SearchHit[];
/** Semantic search: embed the query, then cosine-rank stored embeddings. */
export declare function semanticSearch(query: string, opts?: SemanticOptions): Promise<SearchHit[]>;
/** Reciprocal-rank fusion of multiple ranked result lists. */
export declare function reciprocalRankFusion(lists: SearchHit[][], limit: number, k?: number): SearchHit[];
/** Hybrid search: blend full-text (FTS5) and semantic (vector) results via RRF. */
export declare function hybridSearch(query: string, opts?: SemanticOptions): Promise<SearchHit[]>;
//# sourceMappingURL=vector-search.d.ts.map