You don’t need a separate Graph RAG stack.
Taxonomy mapping → rule edges → graph retrieve. Chunks stay separate.
Co-occurrence Graph RAG
Same PDFs → same graph for every tenant. Connections are a corpus guess.
graph source: co-occurrence
PrismRAG
volatility + drawdown → risk
rule edge weight=1.0
mode: graph_rag · both chunks
prismrag — smoke · demo_taxonomy_connection.py
Step 1 / 6
Try a query (precomputed from live smoke)