Memory

Memory, in the context of Agentic Reasoning, refers to AI Agents storing and loading information outside of their messages and prompts.

An LLM doesn't remember anything between calls; everything it knows about the current task has to fit in its context window. Short-term memory is just that context. Long-term memory is an external store, like files, a database or a vector store, that the agent can write to and retrieve from later, often using Retrieval Augmented Generation.

Papers like RAISE and Reflexion use memory so agents can remember past actions and outcomes, and do better next time.