# LongMemEval

VarOps built a memory system for AI agents working across an organization that distills each observation into a claim carrying who said what, when it held, and which source span proves it, instead of keeping raw transcript chunks. On LongMemEval, a public 500-question benchmark over long chat histories, this deliberately lossy distillation still put the right evidence in the top ten results for 99.8% of questions; ingesting about 35 million tokens cost $8.24 on two CPUs with no GPU. The system also makes time and absence explicit and resolves permissions before any model runs.

VarOps built a memory system for AI agents working across an organization that distills each observation into a claim carrying who said what, when it held, and which source span proves it, instead of keeping raw transcript chunks. On LongMemEval, a public 500-question benchmark over long chat histories, this deliberately lossy distillation still put the right evidence in the top ten results for 99.8% of questions; ingesting about 35 million tokens cost $8.24 on two CPUs with no GPU. The system also makes time and absence explicit and resolves permissions before any model runs.

The work introduces CogMem, a training-free cognitive memory architecture built on the PEC2F (Person-Event-Concept-Claim-Fact) graph schema: dedicated Claim nodes preserve the source and target of subjective statements, dialogue turns are incrementally converted into provenance-aware graph records and consolidated into higher-level facts, and a rule-based controller driven by LLM intent parsing composes four deterministic graph operators—anchoring, traversal, intersection, and evidence grounding—to reconstruct query-relevant context; on LoCoMo and LongMemEval it performs strongly on multi-hop, temporal, and knowledge-update tasks, with ablations and a semantic-collapse probe supporting complementary contributions from epistemic separation, consolidation, and agentic retrieval.

The work introduces CogMem, a training-free cognitive memory architecture built on the PEC2F (Person-Event-Concept-Claim-Fact) graph schema: dedicated Claim nodes preserve the source and target of subjective statements, dialogue turns are incrementally converted into provenance-aware graph records and consolidated into higher-level facts, and a rule-based controller driven by LLM intent parsing composes four deterministic graph operators—anchoring, traversal, intersection, and evidence grounding—to reconstruct query-relevant context; on LoCoMo and LongMemEval it performs strongly on multi-hop, temporal, and knowledge-update tasks, with ablations and a semantic-collapse probe supporting complementary contributions from epistemic separation, consolidation, and agentic retrieval.