Agent Memory
Agent memory is the mechanism by which an AI agent stores and retrieves information across turns or sessions, beyond the fixed context window of the underlying language model.
Full definition
Three common layers: (1) Working memory — the current context window itself; (2) Episodic memory — past conversations or events stored verbatim and retrieved by similarity (vector store of summaries); (3) Semantic / profile memory — distilled facts about a user or domain, stored as key-value or structured data. Frameworks: LangGraph memory, MemGPT (Packer et al., 2023), Letta, Zep. Memory write/read policy is the hard part — naive "store everything" memory rapidly degrades retrieval quality.
Why it matters
Memory turns one-shot chatbots into stateful assistants. It is also the leading source of privacy and security risk in agent products — what gets written to memory is often outside the user's mental model and can leak across sessions.
Example
A sales-coaching agent remembers each rep's pipeline accounts, previous objections handled, and skill-gap notes; after 8 weeks the agent's recommendations measurably outperform a memoryless baseline by 22% on close rate.
Related terms
- AI Agent / Agentic AIAn AI agent is an LLM-driven system that, given a goal, plans a sequence of steps, calls tools or APIs, observes the results, and iterates until the goal is reached — going beyond single-turn chat to multi-step autonomous action.
- Context WindowA context window is the maximum amount of text — measured in tokens — a language model can read and reason over in one inference call, equivalent to the model's working memory for that turn.
- Retrieval-Augmented Generation (RAG)Retrieval-Augmented Generation (RAG) is a technique where a language model retrieves relevant documents from an external knowledge base before generating a response, reducing hallucination by 30-60% and enabling citation of sources.
- Multi-Agent SystemA multi-agent system orchestrates multiple specialized AI agents — often LLM-powered, with distinct roles, tools, and prompts — that communicate, coordinate, and divide labor to solve a problem larger or more diverse than any single agent could handle reliably.
Source & further reading
Primary source: Packer et al. — "MemGPT: Towards LLMs as Operating Systems" (2023).
Citation policy: this entry is part of the AIDOLS AI Implementation Glossary and may be quoted for research, journalism, and education with attribution to aidolsgroup.com/nl/glossary/agent-memory/.