Memory on ChatPlus: stop explaining who you are in every chat

You know the drill. You open a new chat and type it all out again: "I run a small coffee shop, there are three of us, keep it short and skip the intro." You typed the same thing yesterday. And the day before. Without that preamble, the answer comes out generic: advice for a chain of two hundred stores and three paragraphs of warm-up before the point.

The ChatPlus assistant now remembers you across chats. Who you are, what you are working on, how you like your answers — you no longer have to repeat it. Memory is already on for everyone who is signed in, on every plan.

Below: what it is made of, what the assistant remembers, what it never saves, and where to switch it all off.

Three layers of memory

ChatPlus memory has three parts, and each has its own job.

Instructions are written by you. Think of them as a standing note to the assistant: who you are and how to talk to you.

Saved memory is a list of individual facts the assistant noted down during your conversations: your name, your job, a long-running project.

Chat history is a short summary of what you have been up to over the past few weeks. The assistant writes it itself, once a day.

All three live in one place: Settings → Personalization. That is where you edit them, delete them or turn them off.

Instructions: a note read in every chat

Personalization has two fields. The first, "What should the assistant know about you?", is about you: your job, what you do, which languages you speak. The second, "How would you like the assistant to respond?", is about format: short or detailed, lists or prose, code in TypeScript or Python.

Each field takes up to 1,500 characters, roughly half a page, which is plenty. Everything saves automatically; there is no button to hunt for.

Be specific. "I run a coffee shop and handle the social media and ordering myself" tells the assistant far more than "I'm an entrepreneur". And "no intros, straight to the point" works better than "answer well".

If you use projects, they have instructions too. When the two disagree, the project's instructions win in that project's chats. Handy if, say, your work project needs a formal tone while everywhere else you keep things casual.

Saved memory: what the assistant wrote down

The second layer is a list of entries. The assistant adds one when something lasting comes up in conversation: your name, where you work, a project you will be on for months, the answer format you prefer.

When it saves something, a "Memory updated" note appears above the reply, showing the entry itself and a "Manage" link to Settings. Nothing piles up behind your back.

The most reliable way is to ask directly: "remember that I'm vegetarian" or "remember: I want reports as a table". The assistant won't skip a direct request, even if it wouldn't have picked that fact on its own. It works the other way too: "forget that I work at a bank", and the entry is gone, with a "Removed from memory" note to confirm it.

What the assistant does not save:

• one-off task details, like today's shopping list or the email you asked it to polish;
• the contents of files you upload;
• sensitive things — health, political or religious beliefs, finances, passwords — unless you explicitly ask it to remember them.

There is room for up to 60 entries, and the "full" gauge above the list shows how much space is used. When it runs out, the assistant will suggest clearing out old entries in Settings.

One detail that may surprise you: deleting a chat does not delete what the assistant remembered from it. The entry lives separately from the conversation. Each entry has a delete button in Settings, and "Clear memory" wipes everything at once.

Chat history: a summary of the past month

The third layer works without any effort from you. Once a day the assistant reads through your messages from the last 30 days and writes a short summary: what you do, what you are working on, what you are into right now. It uses that summary in new conversations.

The effect is what you get with a person who knows you: you spent a week discussing a delivery launch, and in a new chat about flyers the assistant already knows which delivery you mean.

The summary follows strict rules.

• Only your own messages go into it. The assistant's replies don't, so it can't invent "facts" about you from its own words.
• Delete a chat, and it drops out of the summary straight away; the summary is rebuilt from what remains.
• Health, beliefs, finances and other people's details are left out.

You can read the whole summary in Settings, under "What the assistant learned from your chats". Don't like it? Reset it: the old one is forgotten and a new one is written from your next chats. Don't want a summary at all? Turn off "Reference chat history". Saved memory and instructions keep working.

Where memory works and where it doesn't

In regular chats, memory is fully on: the assistant reads your instructions, entries and summary, and can add new entries.

Agents see your memory too, but read-only. Ask an agent to find coffee bean suppliers and it will keep in mind where your shop is. Only the assistant in a regular chat saves new entries.

Image and video generation don't use memory at all: your bio has no business ending up in a prompt for a birthday card.

Turning it off, and what happens then

The main switch is "Memory", in the same Personalization section. Turn it off and the assistant stops saving anything new and stops using what it already has, entries and chat summary alike. Your instructions keep working, since you wrote them yourself.

Nothing is deleted when you switch off. Turn memory back on and everything returns. To erase it for good, use "Clear memory".

You can also mix and match: memory on, chat history off. Then the assistant only knows what you asked it to remember and draws no conclusions from your conversations.

Getting the most out of it

A few tips from our own use.

Start with the instructions. Five minutes on two fields gets you further than a week of the assistant figuring you out. Say what you do, and how not to answer: "no emoji", "don't repeat my question back to me".

Say "remember" when it matters. The assistant picks up lasting facts on its own, but a direct request always lands.

Check Settings now and then. A couple of months in, there may be stale entries: an old job, a finished project. Each one goes with a single click.

Start a new chat for each task. People used to keep long chats going for the context, so they wouldn't have to explain everything again. Now the assistant knows the essentials in any new chat, and short chats answer faster and more accurately.

Try it now: open a chat and type "remember that…". A "Memory updated" note will appear above the reply. Every switch is explained in the help center.

In short

ChatPlus memory has three parts. You write the instructions yourself. Saved entries — up to 60 — are added by the assistant as you talk, or when you say "remember". A summary of your last 30 days of chats is refreshed once a day. Everything can be reviewed and deleted in Settings → Personalization, sensitive details aren't saved unless you ask, and image and video generation never use memory. It works on every plan; you just need to be signed in.