Context Windows Forget What Matters — I Built a Usage-Reinforced Decay Engine for AI Agent Memory

Standard AI agent memory systems often discard important, foundational information if it hasn't been accessed recently. This content proposes a usage-reinforced decay engine inspired by the Ebbinghaus forgetting curve to solve this problem. In this model, each recall of an item reinforces its stability, non-linearly extending its survival time in the agent's memory. A benchmark comparison showed this method significantly outperformed a simple recency-based window, especially for retaining core facts in long-running sessions. The system is deterministic and its failure case is also defined: it offers no advantage for facts that are introduced but never recalled.