Show HN: AI memory with biological decay (52% recall) | Hacker News

A Hacker News post highlighted a prototype AI memory system for retrieval-augmented agents that uses biological-style forgetting curves and spaced repetition.

The system adds a graph layer on top of vector search to better retrieve related concepts and keep context more relevant over time.

Reported results include 52% Recall@5 on the LoCoMo dataset with much less token waste than storing everything.

The approach suggests a leaner way to manage long-running agent memory by pruning stale context instead of endlessly growing prompts.