Save it once to memory or a profile. Permanent facts about you are exactly what memory is for: save them once and they show up in every conversation by themselves.

What’s the right move?

Keep the password out of memory and log in yourself. A password in memory is a password on someone else’s server. When a login is needed, an agent like ChatGPT agent hands control back to you and you type it yourself.

Where is it best to ask?

In a temporary chat. A temporary chat doesn’t create or update memories, so the surprise won’t pop up in other conversations.

How do you find out what the assistant remembers about you?

Ask it and open the memory settings. Asking gives a quick snapshot, and in settings you can see and delete every saved memory.

Memory: an agent that knows you

Lesson 4 of 22 in the course “AI Agents for Life”. Free, with animations and a quiz.

Two kinds of memory

Remember the brief from the last lesson? Half of it is data about you. So you don’t have to type it every time, assistants have memory. ChatGPT has two kinds. Saved memories are short facts: “I’m vegetarian,” “I study design.” You ask it to remember them, or the assistant saves them when it thinks they matter. The second kind is referencing past chats: the assistant pulls details from old conversations by itself. Both can be switched on and off in settings. Claude and Gemini have memory too, but it works and is named in their own way.

Why an agent needs memory

Without memory, every conversation starts from scratch. Liza asks what to cook and gets steak, even though she hasn’t eaten meat in three years. With memory, the assistant knows she’s vegetarian, works shifts and has just come off one, and gives three 20-minute recipes straight away. Same model, just with context about the person. For agents this matters even more: in every brief, the “data” part is already half filled in. Less typing, less guessing, fewer mistakes.

An “about me” profile

Waiting for the assistant to learn everything on its own takes ages. It’s faster to write a short profile once. It has four blocks: what you do and your schedule, your goals for the next six months, your limits (budget, what you don’t eat, where you live) and how you like your answers. 100–150 words is enough. You can paste the profile into your assistant’s personalization settings or put it in a project if you work in projects. Liza wrote hers in 10 minutes, and since then her assistant has stopped suggesting meat recipes and morning plans on her shift days.

What never goes into memory

Memory is stored on the company’s servers, not on your phone. If your account gets hacked, a stranger sees everything in there. On top of that, an agent can accidentally drop a fact about you where it doesn’t belong, like an email to someone else. So the rule is simple: passwords and codes, card and document numbers, and other people’s secrets don’t go into memory. Health details are your call. OpenAI says it plainly: sensitive information can end up in memory if you share it. That means the filter is you.

Check what it remembers

Do a check-up once a month. First, just ask: “What do you remember about me?” Then open the memory settings (in ChatGPT it’s Settings → Personalization → Memory) and delete what shouldn’t be there. One important detail: deleting a chat doesn’t remove a memory saved from it, you have to delete that separately. For one-off topics, like a surprise gift or an awkward question, there’s temporary chat. It doesn’t create or update memories, so the topic won’t pop up later in another conversation.

Try it: the interviewer agent

Too lazy to write a profile from scratch? Let an agent interview you. Paste the brief below into a regular chat: the agent asks 10 questions one at a time, then builds your profile from the answers. Look at the card. The goal is checkable: 150 words max. There’s a stop rule: at the end it asks what to cut. And passwords and documents sit under “never.” Read the finished profile, cross out anything you don’t want to share, and only then save it to settings or a project.

Check yourself

What will help most?
Correct answer: Save it once to memory or a profile. Permanent facts about you are exactly what memory is for: save them once and they show up in every conversation by themselves.
What’s the right move?
Correct answer: Keep the password out of memory and log in yourself. A password in memory is a password on someone else’s server. When a login is needed, an agent like ChatGPT agent hands control back to you and you type it yourself.
Where is it best to ask?
Correct answer: In a temporary chat. A temporary chat doesn’t create or update memories, so the surprise won’t pop up in other conversations.
How do you find out what the assistant remembers about you?
Correct answer: Ask it and open the memory settings. Asking gives a quick snapshot, and in settings you can see and delete every saved memory.

← The brief: how to give an agent a task Skills: your own agent in 10 minutes →