Zep excels at memory recall but struggles with analytical queries, developer finds

A developer comparing AI memory tools found that Zep, a temporal knowledge graph built on the open-source Graphiti framework, performs well at storing and retrieving conversational facts with temporal context. However, when agents needed to run aggregation queries — such as counting topics or averaging deal sizes — Zep's architecture had no query planner, leaving arithmetic to the LLM and risking hallucinations. The developer contrasted this with nlqdb, a Postgres-backed tool that lets agents query stored data in plain English using standard SQL operations like GROUP BY and JOIN. The conclusion drawn is that the two tools are complementary rather than competing: Zep handles recall while nlqdb handles analytical workloads. The comparison was published on June 19, 2026, with the developer noting that both products' limitations were included in the analysis.