LoCoMo

Long-term conversational memory

We evaluate Cognee on LoCoMo, a benchmark of very long-term conversational memory across ten conversations of up to 32 dated sessions each. With standard Cognee memory components and one answer prompt for every question, Cognee answers 91% of the 1,762 evaluated questions correctly, graded by an LLM judge.

Overall0.91
Single-hop0.93
Multi-hop0.90
Temporal0.90
Adversarial0.91
Loading form...