Module vector_actions 

Functions

handle_hybrid_search

handle_refresh_graph_cache

handle_vector_batch_upsert

Bulk-ingest embeddings: { items: [{entityName, embedding, model?}, ...] }. Each item is upserted independently; per-item failures are reported rather than aborting the batch — the shape RAG ingestion pipelines expect.

handle_vector_delete_embedding

handle_vector_get_embedding

Fetch the stored embedding for an entity: { entityName }.

handle_vector_mmr_search

Maximal Marginal Relevance search: diversified semantic retrieval. { embedding, topK?, fetchK?, lambda?, entityType? }. lambda in [0,1] trades relevance (1.0) against diversity (0.0). Reduces near-duplicate hits — a common RAG context-selection step. The reported score is the MMR score.

handle_vector_recommend

Example-based recommendation: build a query from positive (and optional negative) example entities and search. { positive: [names], negative?: [names], topK?, entityType? }. The example entities are excluded from results.

handle_vector_reindex

Rebuild/retrain the ANN index (IVF k-means; HNSW is a no-op). {}.

handle_vector_search_by_entity

“More like this”: find entities nearest to a given entity’s own embedding. { entityName, topK?, entityType?, excludeSelf? }.

handle_vector_search_entities

handle_vector_store_stats

handle_vector_upsert_embedding