​Most production agent architectures combine relational state, graph relationships and vector retrieval across separate systems. That flexibility comes with trade-offs: different consistency models, extra round trips, and more application logic to keep everything aligned.



​In this session, Matthew will show where that complexity starts to affect retrieval quality and operational reliability, especially when filtering, ranking and relationship context live in different places. He’ll then walk through an alternative approach: keeping graph, documents, vectors and state in one engine, so teams can simplify the stack and improve accuracy.