noun · AI concept
Long-term memory
also called persistent memory
A notebook kept outside the backpack. The AI can read from it next time, even though it forgot the last chat.
Information stored outside the model between sessions and loaded back into the context when it is relevant.
Facts are saved outside the model, in files or a database, and the harness loads the relevant ones into the next request. The model stays stateless, so memory is only as good as what gets saved and what gets retrieved.
Real-life example
You tell an assistant you prefer short answers and use a certain framework. Next week it still follows that. The model did not remember you. The app saved a note, found it when you started a new chat, and slipped it into the request.
Where the analogy breaks
-
The model does not hold the memory. The app stores it and chooses what to load, so a note that is not retrieved might as well not exist.
-
Saved notes can be wrong or out of date. The assistant will trust them, so old or mistaken entries quietly steer later answers.
-
It raises privacy questions. Anything saved is data kept about you, so check what is stored, who can see it and how to delete it.