Memory is a short list of lasting facts about you that the assistant reads at the start of every chat: your name, your work, the thing you are building, how you like your notes written. It exists so you stop introducing yourself to your own notes. Memory is part of the Plus, Pro and Lifetime plans, along with custom instructions: on Free or Sync no memory goes out with a chat and none is saved, and the list waits for the plan.
How something gets remembered
You do not fill in a form. Mention in passing that you run a bakery while asking about customer notes, and the bakery is saved while the question gets answered. The assistant has tools to save, update and delete memories, and the chat labels it when one of them runs, so you can see the moment something was kept.
Reading the list
Open Settings and go to Personalization. Every memory is a plain sentence rather than a black box. Add one yourself, edit one that came out wrong, or delete one you do not want kept. Up to fifty memories are held, each up to two hundred characters.
Each memory has a grip at the start of its row, so you can put the list in the order you like. On desktop, drag it up or down; on the phone, touch and hold it, then drag. A new memory joins the end.
Asking it to forget
Say forget that in the chat and the memory is deleted rather than hidden. Naming what to forget works too.
Turning it off
The switch is in the same Personalization card, or ask the chat to turn memory off. While it is off, no memory text goes out with a chat on any device. Your existing memories stay in the list, so turning it back on picks up where it left off. An agent has a Use memory switch of its own, so you can keep memory out of one agent's chats and in the rest.
Across your devices
Memories travel with your settings, so with sync on there is one list on every device and the phone knows what the laptop learned. Without sync they stay on the device they were saved on.
Memory versus instructions
Memory is facts about you that the assistant collects. Custom instructions are rules you write about how it should answer. They work together: with both set, replies arrive already in your voice.