Agent Memory

Agent memory is how an AI agent retains information beyond a single model call, so it can act coherently across many steps or sessions instead of forgetting everything each turn. It usually splits into short-term (working) memory — the running context of the current task, held in the context window — and long-term memory — facts, past interactions, and learned preferences stored outside the model, typically in a vector database or plain database and retrieved when relevant. People often further divide long-term memory into episodic (what happened before) and semantic (durable facts about the user or domain). For builders, memory is what turns a stateless chatbot into a product that remembers a customer's plan, past tickets, or brand voice. The practical challenge is curation: naively stuffing every past message into the prompt is expensive and degrades accuracy, so you summarize, score relevance, and retrieve only what each step actually needs.