Skip to main content
A memory is a durable fact about you or how you want to be worked with. It applies to every conversation, not just the one it was learned in.

​
How they are made

The assistant saves them itself, quietly, when you state a lasting preference. Say that you always want Python examples in a particular style, or that your company name is spelled a certain way, and it is remembered without a confirmation step and without being mentioned. You cannot add one by hand. That is deliberate: a memory is meant to come from something you actually said in the course of working.

​
Managing them

Open Memories in the sidebar. You can read every memory, edit its wording, and forget any of them. The footer shows how many of the fifty you have used. Forgetting is immediate and permanent.

​
Limits

Near duplicates are rejected, so the assistant cannot slowly fill the list with fifty phrasings of the same preference.

​
How they are applied

Memories are put in front of the model at the start of every conversation, with an instruction to act on them without being asked and without mentioning them. Anything you say in the current conversation overrides a memory that contradicts it. A memory is a default, not a rule.