Agent Memory

What Is Agent Memory?

A stateless LLM forgets everything between calls. Agent memory gives an agent continuity: short-term memory holds the current task scratchpad, while long-term memory persists facts, preferences, and past outcomes in a vector store or database the agent retrieves from later. Good memory design is what separates an agent that repeats its mistakes from one that improves. Poor design leaks stale or irrelevant context and degrades responses.