Memory
The mechanism by which an agent remembers information across sessions — usually a vector store or structured key-value cache.
Memory is what makes an agent feel persistent. Without it, every session starts cold and the agent has no idea you asked the same question yesterday. With it, the agent builds up context — preferences, prior decisions, ongoing projects.
Two flavors dominate in 2026: episodic (recall specific past interactions) and semantic (learn general facts about the user and surface them when relevant). Production agents use both.
The harder problem is *forgetting*. Without explicit pruning, memory bloats and becomes a liability. The agents that handle this well treat memory as a working set, not an archive.
Where this shows up
Frequently asked
How is agent memory different from a chatbot keeping conversation history?+
Chatbot history is short-term, scoped to the session, and stored in the prompt. Agent memory is long-term, persists across sessions, and is queried only when relevant.