Memory Store
Also called Memory Store
A place an AI writes things down so it can remember them in a later conversation.
Think of it like
A notebook the assistant keeps on the desk — not everything it ever said, just the facts worth carrying forward.
Example
A coding agent saves "this project uses pnpm, not npm" to its memory store, and pulls it back up automatically next session so you don’t repeat yourself.
How it actually works
Models are stateless between calls — once the context window ends, the conversation is gone. A memory store is external storage (often a vector database plus some structured records) that the agent reads from and writes to across sessions. The hard parts aren’t storage but curation: deciding what’s worth remembering, when to update it, and how to retrieve the right memory without dredging up stale ones.
For product teams
The difference between an assistant that feels like it knows you and one that resets to a stranger every session.
For engineers
Persist salient facts to external storage keyed for retrieval; on each turn, fetch relevant memories into context and write back new ones under a curation policy.
Related
- Vector database — Where memories are usually stored for lookup.
- Context window — Why external memory is needed at all.
- RAG — Retrieval into the prompt works like RAG.
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