export type { Lens, HydeTargetCorpus } from '../hyde/lens.inferrer.js';
/** A single lens embedding ready for search. */
export interface LensEmbedding {
    /** Free-text lens label (e.g. "crypto infrastructure VC"). */
    lens: string;
    /** Which corpus to search. */
    corpus: 'profiles' | 'intents' | 'premises';
    /** 2000-dim embedding vector. */
    embedding: number[];
}
/** Options for searchWithHydeEmbeddings (network scope, limits, min score). */
export interface HydeSearchOptions {
    /** Network IDs to scope the search (members / assigned intents only). */
    indexScope: string[];
    /** Exclude this user ID from results (e.g. source intent owner). */
    excludeUserId?: string;
    /** Max results per lens before merge (default 10). */
    limitPerStrategy?: number;
    /** Max results after merge/rank (default 20). */
    limit?: number;
    /** Minimum cosine similarity for intent searches (default 0.40). */
    minScore?: number;
    /**
     * Discovery corpus gating, composed by the caller (defaults preserve legacy behavior).
     * Omitted fields default to: intents true, profile true, profileCorpus 'premise'.
     */
    corpusGating?: {
        /** Search the intents corpus. */
        intents?: boolean;
        /** Search the active profile corpus. */
        profile?: boolean;
        /** Which corpus backs 'profiles' lens hints and profile searches. */
        profileCorpus?: 'premise' | 'user_context';
    };
}
/** A single candidate from HyDE search (intent, premise, or user_context), with score and which lens matched. */
export interface HydeCandidate {
    type: 'intent' | 'premise' | 'user_context';
    id: string;
    userId: string;
    score: number;
    /** Free-text lens label that produced this match. */
    matchedVia: string;
    networkId: string;
    /** Candidate document text (populated for user_context matches; used as candidatePayload). */
    text?: string;
    /** Set after merge when user matched via multiple lenses. */
    matchedLenses?: string[];
}
export interface EmbeddingGenerateOptions {
    signal?: AbortSignal;
}
export interface EmbeddingGenerator {
    generate(text: string | string[], dimensions?: number, options?: EmbeddingGenerateOptions): Promise<number[] | number[][]>;
}
export interface VectorSearchResult<T> {
    item: T;
    score: number;
}
export type VectorStoreOption<T> = {
    limit?: number;
    filter?: Record<string, any>;
    candidates?: (T & {
        embedding?: number[] | null;
    })[];
    minScore?: number;
};
export interface VectorStore {
    /**
     * Search for similar items in the vector store.
     *
     * @param queryVector - The embedding vector to search for
     * @param collection - The logical name of the collection (e.g., 'profiles', 'intents')
     * @param options - generic options including limit, filter, and candidates
     */
    search<T>(queryVector: number[], collection: string, options?: VectorStoreOption<T>): Promise<VectorSearchResult<T>[]>;
}
/**
 * Embedder: generate embeddings and run vector / HyDE search.
 * Implementations: OpenAI/OpenRouter for generate, pgvector for search.
 */
export interface Embedder extends EmbeddingGenerator, VectorStore {
    /**
     * Multi-lens HyDE search: run one vector search per lens embedding,
     * then merge, deduplicate by userId, and rank (boost for multiple lens matches).
     *
     * @param lensEmbeddings - Array of lens embeddings to search with
     * @param options - indexScope, excludeUserId, limits, minScore
     * @returns Deduplicated, ranked candidates (intent or premise) with scores
     */
    searchWithHydeEmbeddings(lensEmbeddings: LensEmbedding[], options: HydeSearchOptions): Promise<HydeCandidate[]>;
}