Memory Systems Design

Memory Systems Design is an essential skill for building agents with persistent state across sessions. Simple agents lose all information when sessions end, but this skill teaches layered memory architectures for maintaining continuity and reasoning over accumulated knowledge. Memory operates on a spectrum from immediate context to permanent storage, and this skill helps you balance zero-latency access against information persistence.

The skill covers multiple memory layers and their characteristics. Working Memory operates within the context window with immediate access but limited capacity. Short-term Memory provides session-persistent searchable storage. Long-term Memory enables cross-session persistence for learning patterns. Entity Memory tracks consistent information about specific entities throughout conversations. Temporal Knowledge Graphs add validity periods to facts, enabling time-aware reasoning.

Progressive memory architectures are explored, from simple vector stores to knowledge graphs to temporal knowledge graphs, each adding structure for improved reasoning and retrieval accuracy. Practical implementation patterns include file-system-based memory (simple, no infrastructure needed), vector RAG with metadata (enables semantic search), knowledge graphs (models relationships between entities), and temporal knowledge graphs (time-aware queries).

Activate this skill when developing agents that must persist across multiple user sessions, maintain consistent entity information throughout conversations, reason over accumulated knowledge, learn from past interactions, or track how information changes over time. The core principle: memory exists on a spectrum from immediate context to permanent storage, requiring architectural choices that balance latency, capacity, persistence, and retrieval complexity based on specific system requirements. Essential for anyone building agents that need to remember and learn.