We ran cognee on LoCoMo with the same standard components we used on BEAM: one remember() call per conversation, the hybrid retriever, one answer prompt, and a judge that rejects topical matches. Result: about 91% of 1,762 evaluated questions, adversarial included, with every prompt, exclusion and verdict published.
cognee on LoCoMo: 91% With Standard Memory Components and a Strict Judge
calendar_today
November 11, 2026
domain
cognee