Graph retrieval

Expand diagram

Storage and source provenance

PostgreSQL is the source for chunk text and embeddings. Neo4j entity links retain chunk IDs so graph results can be resolved to citable passages.

Use the zoom controls, drag to pan, or pinch on a touch screen. Press Escape to close.

Graph search reduces a question to a bounded list of terms, matches entity names in Neo4j, and collects their source chunk IDs. The API then loads the corresponding passages from PostgreSQL.

Provenance requirementCopy Anchor Link

A graph match becomes usable evidence only after its chunk IDs resolve to source text. Entity relationships without source chunk IDs cannot produce a document-grounded citation. Graph coverage and provenance remain incomplete, so ordinary retrieval still relies mainly on document chunks.

Graph and vector retrieval address different patterns: graph lookup can help with named entities and relationships, while vector retrieval searches the full chunk corpus. For the stored data model, see Data model and storage.