Memory Management
definition
Memory management governs when agents store new memories, how they retrieve relevant ones, and when they evict or summarize old information to stay within operational limits. Effective strategies include recency-weighted retrieval (favoring recent information), importance scoring (prioritizing high-signal memories), summarization (compressing old conversations into distilled knowledge), and explicit forgetting policies (removing outdated or contradictory information).
Memory management governs when agents store new memories, how they retrieve relevant ones, and when they evict or summarize old information to stay within operational limits. Effective strategies include recency-weighted retrieval (favoring recent information), importance scoring (prioritizing high-signal memories), summarization (compressing old conversations into distilled knowledge), and explicit forgetting policies (removing outdated or contradictory information). The fundamental tension is between recall (having enough context to make good decisions) and efficiency (not overwhelming the model with irrelevant history that degrades performance and increases cost). In production, most agent failures attributed to "forgetting" are actually retrieval failures — the information was stored but the agent couldn't surface it at the right moment. This concept connects to short-term memory and long-term memory for the storage layers being managed, context window budget for the constraint that drives memory management, and RAG patterns for the retrieval mechanisms that make stored memories accessible.