Agents extend what a model knows by loading stored memories and retrieved documents at run time. Every retrieval path is also an injection path, so you need to know where memories and documents come from, who can write to them, and how long they persist.
Read in this order
- OfficialMemory overview (LangChain Docs)Start here: the main memory types and how agents use each one.
- ArticleLLM Powered Autonomous Agents: Memory section (Lilian Weng)A clear survey of memory and retrieval inside an agent design.
- PaperCognitive Architectures for Language Agents (CoALA)A framework for describing any agent's memory, useful for mapping what it can write.