What Is Agent Memory? How AI Agents Remember Beyond a Single Conversation
A model itself has no persistent memory between sessions — agent memory is the system built around it that lets an agent recall relevant information over time.
The model doesn't remember on its own
An underlying language model has no built-in memory between separate conversations — everything it "knows" during a session comes from what's currently in its context window. Agent memory is the surrounding system that captures, stores, and retrieves relevant information across sessions so an agent can behave as if it remembers.
Short-term vs. long-term memory
Short-term memory is usually just the current conversation's context, discarded once the session ends. Long-term memory persists information — facts, preferences, past interactions — outside the context window, in a separate store, then retrieves and injects the relevant pieces back into context when they become useful again.
How retrieval usually works
Long-term agent memory is often implemented using the same building blocks as RAG: stored memories get embedded and searched via a vector database, pulling back only what's relevant to the current situation instead of dumping the agent's entire history into every prompt.
Frequently Asked Questions
What is agent memory?
The system built around a model that captures, stores, and retrieves relevant information across sessions, letting an AI agent behave as though it remembers past interactions despite the underlying model having no built-in memory of its own.
How is long-term agent memory usually implemented?
Often using the same approach as RAG — stored information gets embedded and searched via a vector database, pulling back only what's relevant to inject into the current context rather than replaying the entire history.