Short-Term Memory
Also called Working Memory (LLM)
What the model can "remember" right now — everything sitting in the current context window.
Think of it like
The notes spread on your desk during a meeting: instantly available, but swept away when you leave the room.
Example
A chatbot recalls what you said three messages ago because it is all still in the prompt — not because it learned anything.
How it actually works
Models have no memory between calls; "short-term memory" is just the tokens you feed in this turn. It is fast and lossless while it lasts, but bounded by the context window and gone the moment the conversation is trimmed or a new session starts. Anything you want kept longer has to be written somewhere and retrieved later.
For product teams
It sets how much a session can "hold in its head" before you must persist and retrieve context yourself.
For engineers
Non-persistent state = the current token sequence; nothing survives outside the request without external storage.
Related
- Context window — Lives entirely inside the context window.
- Conversation History — The running transcript that fills it.
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