Agent Memory
What an agent remembers beyond the current conversation — facts, past steps, preferences it can pull back up later.
Think of it like
A notebook the assistant keeps between meetings, so you don’t have to reintroduce yourself every time.
Example
An agent remembers from last week that you prefer metric units and morning meetings, and applies that without being told again.
How it actually works
Because the context window is finite and resets, real continuity needs external memory: notes stored in a database, often as embeddings, retrieved when relevant and slotted back into context. The design questions are the hard part — what’s worth saving, when to recall it, when to forget. Done badly, it either forgets everything or drags in stale, irrelevant detail.
For product teams
The difference between an assistant that knows you and one that meets you fresh each time.
For engineers
External storage (often a vector store) that persists facts and history across sessions, retrieved into context on demand.
Related
- Inference — The frozen limit that memory works around.
- Embeddings — How stored memories are usually found again.
- RAG — The retrieve-then-answer pattern 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