Context Graph Boosts Multi-Agent Memory Accuracy 89% vs 50% Vector RAG
A benchmark of three memory architectures shows flat transcripts hit 61% accuracy, vector-only RAG drops to 50%, while a context‑graph layer reaches 88.9% with only 27 tokens per query. The graph stores entities and relations, solving the structural blind spot that makes multi‑agent recall brittle.