Short-Term vs Long-Term Agent Memory

Short-term memory helps an agent stay coherent inside a thread. Long-term memory lets information survive beyond that thread.

Short-term memory

Examples include conversation history, current plan, tool results, and intermediate state. LangGraph describes this as thread-level state that can be persisted with a checkpointer.

Long-term memory

Long-term stores hold user- or application-level information across sessions. Retrieval should be selective rather than injecting an entire store into every prompt.

Failure mode: stale memory

Durable facts can become outdated. Store timestamps and provenance, and revalidate facts that materially affect decisions.