/**
 * pgvector backend for iranti's vector search layer.
 *
 * Stores and queries embedding vectors in the `knowledge_base.embedding`
 * column using the `pgvector` Postgres extension. Falls back gracefully
 * when the extension is not installed or the `embedding` column is missing
 * (checked once via `hasVectorSupport()` and cached for the process lifetime).
 *
 * Vector writes use raw SQL (`$executeRaw`) so they can run inside a Prisma
 * transaction alongside knowledge entry mutations. The `VectorMutationDbClient`
 * type accepts either the full `PrismaClient` or a Prisma transaction client.
 *
 * Search uses a cosine-distance `<->` operator order-by with an IVFFLAT or
 * exact scan depending on what pgvector supports. Results are capped and
 * filtered by `isProtected = false`.
 */
import { VectorBackend, VectorConsistencyFilter, VectorMutationDbClient, VectorSearchResult, VectorUpsertParams } from '../vectorBackend';
export declare class PgvectorBackend implements VectorBackend {
    upsert(params: VectorUpsertParams, db?: VectorMutationDbClient): Promise<void>;
    delete(id: string, db?: VectorMutationDbClient): Promise<void>;
    search(vector: number[], topK: number, filter?: Record<string, unknown>): Promise<VectorSearchResult[]>;
    listIndexedIds(filter?: VectorConsistencyFilter): Promise<string[]>;
    ping(): Promise<boolean>;
}
//# sourceMappingURL=pgvectorBackend.d.ts.map