Score agent memory retrieval with three parallel passes — semantic, keyword, and entity matching — then fuse

Mem0·medium signal

2026 memory systems are moving past pure vector similarity: the retrieval stack runs semantic-similarity, keyword-match, and entity-match scoring passes in parallel and fuses the results. Pure embeddings miss exact entity names and identifiers that keyword and entity passes catch, which is where single-vector recall silently fails. For anyone building agent memory, adding the entity/keyword passes alongside vector search is a cheap recall win over a similarity-only retriever.