Information an agent can persist and retrieve beyond the immediate context.
Agent memory may store facts, events, preferences, decisions, or procedural knowledge across sessions. Writing, retrieval, consolidation, correction, permissions, and forgetting are all design problems; more stored material is not automatically better memory.
Suppose you explain a project constraint today and open a new conversation tomorrow. The agent can use that constraint again only if the system saved it and brings it into the new conversation. Ask where the fact is stored, how it is retrieved, and how you can correct it.
Saving a conversation does not guarantee that the next answer uses it. Memory requires both persistence and a way to select relevant information. A remembered fact can also become outdated, so its source and date matter.
Background: LangChain: short-term and long-term memory.