Semantic Procedures

This family is about semantic retrieval and diversity-oriented semantic retrieval.

The score column is a cosine-similarity relevance score where higher values are better. Use ORDER BY score DESC when you want to preserve relevance ordering after joining hits back to graph data.

The canonical user-facing semantic procedure is nexyron.semantic.search. It searches one unified semantic ANN index containing graph-node embeddings, graph-edge embeddings, abstractions, and abstraction artifacts in one call and returns one unified row shape:

Normal semantic search should use one procedure:

Assistant, Client, wiki, and report search surfaces consume these procedures through the service-owned graph context retriever. That retriever adds bounded graph expansion, source references, observed-data snippets, score components, provenance, and temporal hints while preserving the same canonical Cypher/cluster read path.

Semantic retrieval requires the unified derived ANN index to be hot. If indexed semantic vectors exist but ANN is cold, semantic procedures return an error instead of falling back to brute-force vector scanning. Check /admin/{db}/semantic/status and require ann_ready: true before treating semantic search as available.

Low-level index debugging references: