Zep gives my agent perfect recall. It still can't answer "average per group" about its own memory.

Short summary

Zep (Graphiti) excels at temporal knowledge-graph recall for AI agents but cannot perform aggregations like GROUP BY or JOIN on stored memory. The author's product nlqdb fills this gap by provisioning a real Postgres that agents query in English, enabling analytical questions over memory. The two tools compose: Zep for recall, nlqdb for analysis — pick based on the question type your agent needs to answer.

• •Zep is strong for temporal recall and entity resolution but lacks a query planner for aggregations
• •nlqdb provides a Postgres-backed analytical store agents query in English for GROUP BY/JOIN/HAVING operations
• •The two tools compose: Zep as recall layer, nlqdb as analytical store

Generated with AI, which can make mistakes.