Agent Memory: Layers & Types
Agent memory two ways: four layers by lifetime and scope (context window, session, long-term user, shared org) and three types by content (episodic, semantic, procedural), with one map between them.
Last updated
After this section you can
- Place a memory requirement in the right layer: context window, session, long-term user or shared org
- Classify what is being stored as episodic, semantic or procedural, and say how each is written and retrieved
- Budget a turn’s working memory across history, retrieved memories and documents
16
Agent Memory: Layers & Types
The model remembers nothing between calls. Memory is what your system loads back into the window: four layers sorted by lifetime and scope, three kinds of content, and one map between them.
Key idea
An agent remembers only what your code puts into its context window. Sort memory two ways: where it lives (window, session, long-term user, shared org) and what kind of thing it is (episodic, semantic, procedural). The first decides storage; the second decides how you write, find and update it.
The model reads one thing, the context window. Every other layer is your code putting tokens back into it.