Feature · Relationships

Graph RAG

Some answers live in the relationship between documents: a decision in one note, a constraint in another, and the source that explains why the plan changed.

What it solves

Standard retrieval can find passages that mention a topic. Graph RAG helps when the question depends on entities and relationships distributed across several documents.

01 · RetrieveStart with the strongest local passages.
02 · ExtractIdentify entities and relationships.
03 · ExpandFollow connected entities.
04 · CiteReturn source markers intact.

How Mneme handles it

Graph RAG combines standard retrieval with entity-relationship expansion and alpha fusion. Its schema-validated JSON cache is tied to the index fingerprint.

What you can inspect

  • The source passages used to identify relationships.
  • The index fingerprint that bounds the graph cache.
  • The final cited files, pages, and chunks.
LLM-backed step

Entity extraction is an LLM-backed feature. Retrieved snippets are sent to the endpoint you configure for that step; local indexing and retrieval remain local.

Try it

python -m src.graph_rag --files /path/to/docs --collection my_docs --alpha 0.7

Search Mneme documentation