ChatGPT memory got an upgrade. It still is not a knowledge base.

On June 4, OpenAI started rolling out a more capable version of ChatGPT's memory. The background system behind it is called "dreaming." Instead of waiting for you to say "remember this," ChatGPT curates memories on its own from your past chats. It revises them as time passes, so a trip coming up in July is remembered as one you already took once July is over. There is a memory summary page where you can see what it knows, correct a detail, or tell it to drop something.

It is a real improvement. If you live inside ChatGPT, you will feel it. The personalization gets quietly better without you doing anything.

So it is worth asking the obvious question. If ChatGPT remembers you now, do you still need somewhere to keep your notes?

Yes. They do different jobs. Here is the honest version.

The key difference is this. ChatGPT memory is inferred. A knowledge base is authored. One is the assistant's impression of you. The other is the source material you chose to keep.

What Dreaming is good at

Give it credit first. Dreaming is a strong personalization layer.

• It learns your preferences without you managing anything. Tone, format, the projects you keep mentioning.
• It works passively. No "save" step, no folders to maintain.
• It keeps itself current. Old facts get revised instead of repeated back to you wrong.

If all you want is for ChatGPT to feel like it knows you, this is most of the way there. We are not going to pretend otherwise.

What it is not

Dreaming remembers things about you. That is not the same as holding the things you write down.

Three gaps, and none of them are bugs. They are just what built-in memory is.

It lives in one app

Your ChatGPT memory is in ChatGPT. It does not follow you to Claude, to Cursor, or to Copilot. Open a different assistant and you start from zero again.

A knowledge base does the opposite. You write notes once, and any AI you connect reads the same ones. Ask Claude to draft something and it reads your notes. Switch to ChatGPT to send it and ChatGPT reads the same notes. One store, every client.

Here is what that looks like. You ask Claude to turn a customer interview into product notes. Next week you ask ChatGPT to write the launch email from those same notes. The source stays the same. The assistant changes.

It decides what to keep, and you read a summary

Dreaming curates. It picks what is worth remembering and shows you a tidy summary of it. That is fine for "you prefer concise answers." It is the wrong shape for a runbook, a customer interview, a design decision, or a style guide.

Those are documents. You want the exact text, not a model's summary of it. You want to open the note, edit a line, organize it into a folder, and search it like the document it is. A knowledge base is notes, folders, and tags in Markdown that you author and own. Built-in memory is a model's running impression of you.

It is personalization, not knowledge

This is the core of it. Dreaming makes ChatGPT better at being ChatGPT for you. It does not store your team's architecture notes, your literature review, your interview transcripts, or the decisions you will need to quote next month.

That is the difference between a feature and a system. Built-in memory is a feature inside one product. The notes worth keeping belong in a system that outlives any one product's roadmap.

Use both

This is not replace-one-with-the-other. Let ChatGPT's memory do what it is good at. Let it remember that you like short answers and that you are working on a launch. Keep the notes worth writing down in a knowledge base your AI can actually read.

The line we keep coming back to: built-in memory is rented, your knowledge base is owned. Rent the convenience. Own the context.

Who should pick what

Built-in memory is enough if: you only use ChatGPT, you only need light personal facts remembered, and you do not want to think about where any of it lives.

You want a knowledge base if: you use Claude or ChatGPT (or both, or plan to), you want notes you can read, edit, search, and export like normal documents, you want shared memory across a team, and you care that the knowledge survives if a pricing page or a model changes.

Try it next to your ChatGPT memory

Try Hjarni next to ChatGPT memory. Let ChatGPT remember your preferences. Keep your actual notes in a place Claude and ChatGPT can both read.

Hjarni is a knowledge base with a built-in MCP server. You write notes. Claude and ChatGPT read them, follow your instructions, and remember what you told them across every conversation. Your ChatGPT memory keeps running alongside it. The two do not fight.

If you want the full feature-by-feature breakdown, see Hjarni vs ChatGPT Memory. If you are ready to wire it up, give Claude long-term memory or connect ChatGPT takes about five minutes. For the bigger argument, the AI-native second brain is the place to start.

ChatGPT can remember a few things about you now. That is a good upgrade. It is still not the place to keep what you write down.