/**
 * Local SQLite-backed vector store.
 * Replaces ChromaDB (JS client only supports HTTP servers, not local embedded).
 * Stores embeddings as JSON blobs, performs cosine similarity in JS.
 *
 * API surface mirrors Python ChromaDB PersistentClient to keep the rest
 * of the codebase consistent with the plan.
 */
import type { IEmbeddingFunction } from './embeddings.js';
export interface VectorRecord {
    id: string;
    document: string;
    metadata: Record<string, string | number>;
    embedding: number[];
}
export interface SearchResult {
    id: string;
    document: string;
    metadata: Record<string, string | number>;
    distance: number;
}
export declare class VectorCollection {
    private db;
    private embedFn;
    private name;
    private constructor();
    /** Create or open a collection backed by a SQLite file at dbPath. */
    static open(dbPath: string, name: string, embedFn: IEmbeddingFunction): VectorCollection;
    private init;
    /** Check if an id already exists. */
    has(id: string): boolean;
    /** Add a document. Does nothing if id already exists (explicit dedup gate). */
    add(params: {
        id: string;
        document: string;
        metadata: Record<string, string | number>;
    }): Promise<void>;
    count(): number;
    /** Get records by ids or all records if no ids specified. */
    get(params?: {
        ids?: string[];
        where?: Record<string, unknown>;
        limit?: number;
    }): {
        ids: string[];
        documents: string[];
        metadatas: Record<string, string | number>[];
    };
    /**
     * Semantic search — cosine similarity between query embedding and all stored vectors.
     * Returns top nResults sorted by ascending distance (0 = identical, 1 = opposite).
     */
    query(params: {
        queryTexts: string[];
        nResults: number;
        where?: Record<string, unknown>;
        include?: string[];
    }): Promise<{
        ids: string[][];
        documents: string[][];
        metadatas: Record<string, string | number>[][];
        distances: number[][];
    }>;
}
//# sourceMappingURL=vector-store.d.ts.map