Provides a coherent memory layer for LLM agents that models beliefs as nodes in a constraint network to detect and surface semantic contradictions. It uses local embeddings and a hybrid geometric-logical energy model to identify conflicting information that requires review.
View & run on Rokha →