import type Database from 'better-sqlite3';
import { type FtsResult, type LabelFilter } from '../storage/fts-store-search.js';
/** Turns text into L2-normalised vectors. Supplied by the host; monograph has no model dependency. */
export interface Embedder {
    /** Identifies the model; a change invalidates every stored vector. */
    model: string;
    dim: number;
    embedPassages(texts: string[]): Promise<Float32Array[]>;
    embedQuery(query: string): Promise<Float32Array>;
}
export declare function quantize(vec: Float32Array): Buffer;
export interface EmbeddingStatus {
    /** Symbols and files with a summary — the number that can be embedded. */
    total: number;
    embedded: number;
    model: string | null;
}
export declare function embeddingStatus(db: Database.Database): EmbeddingStatus;
export interface EmbedOptions {
    batchSize?: number;
    /** Stop after roughly this long; the next call resumes where this one stopped. */
    maxMs?: number;
    onProgress?: (done: number, pending: number) => void;
}
export interface EmbedResult {
    embedded: number;
    /** Still without a current vector after this call. */
    remaining: number;
}
/** Embeds every summary that has no vector yet, or whose text changed since its vector was made. */
export declare function embedPending(db: Database.Database, embedder: Embedder, options?: EmbedOptions): Promise<EmbedResult>;
export interface SemanticHit extends FtsResult {
    /** Cosine similarity to the query, roughly -1..1. */
    similarity: number;
}
/** Nearest summaries to `queryVec` by cosine similarity, optionally restricted to some node labels. */
export declare function semanticSearch(db: Database.Database, queryVec: Float32Array, limit: number, label?: LabelFilter): SemanticHit[];
//# sourceMappingURL=store.d.ts.map