Memory Retrieval
Pulling the right past information back into an agent’s working memory when it’s relevant.
Think of it like
Remembering a friend’s allergy right as you’re about to order for the table — the fact surfaces when you need it.
Example
Before answering, the agent retrieves the note it saved three sessions ago that this user prefers metric units.
How it actually works
An agent can store far more than fits in its context, so retrieval is the bridge — search the memory store and pull only what matters into the prompt. It usually reuses the same machinery as RAG: embeddings, similarity search, ranking. The quality ceiling is the retrieval step: if the right memory isn’t surfaced, it might as well not exist, and irrelevant memories just crowd out useful context.
For product teams
An agent’s memory is only as good as its ability to recall the right piece at the right moment.
For engineers
Query the memory store — usually by embedding similarity — and inject the top matches into the active context.
Related
- Episodic Memory — The durable events it retrieves from.
- Retrieval — The general technique it borrows from.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome