Agent Memory
State an agent keeps beyond the current context window.
Overview
Agent memory is whatever an agent retains between steps or sessions - a scratchpad, a summary, a store it can search - because the context window is finite and the task is not. It solves the problem that an agent doing multi-step work exceeds what fits in front of the model. It exists as the reason long-running agents are mostly a retrieval problem with a model attached.
Technology Graph
How Agent Memory connects to the rest of the stack.
Uses and depends on
- Embedding concept Techniques Agent Memory uses Embedding High ✓ source
- Context Window concept Models Agent Memory based on Context Window High ✓ source
Used by
- Multi-Agent technique AI Agents Agent Memory used by Multi-Agent High ✓ source