ChatGPT's memory is full. Here's how to fix it.

You hit a limit. Either a long conversation stopped with "Chat memory full", or ChatGPT told you its saved memory is full and new facts about you are no longer being saved.

Those are two different limits with two different fixes. Each takes two minutes. A better setup stops either one from costing you context again.

"Chat memory full" in a conversation

Since around late September 2026, users report ChatGPT stopping long conversations with this notice:

Chat memory full. Start a new chat with an automatic summary of this conversation, or upgrade to continue here with expanded memory in every chat.

This is not your saved memory. It means this one conversation has used up its room, so deleting memories in Settings will not unlock it. We could not find a published limit for each plan.

What to do:

1. Start the new chat with the summary. ChatGPT carries a summary of the conversation over. Read it before you carry on: a summary keeps the gist and drops the details, so restate anything specific it missed.
2. Save that summary as a note before you carry on. Otherwise the next long chat ends the same way, with a summary of a summary. Install Hjarni from the ChatGPT Apps Directory (free, one click), select it from the + menu in the new chat, and ask: "Save this summary as a note called Project X context, and add the details it missed: …"
3. Pick up from the note, not a summary. The next time a chat fills up, open a new one and ask: "Read my Project X context note and continue where we left off." Ask ChatGPT to update the note as decisions land, so it stays current instead of going stale.
4. Or upgrade, as the notice offers, if you need to keep going in the same thread. The thread still fills up eventually; the note does not.

The note lives in your knowledge base, not in ChatGPT, so it survives any limit, and Claude can read the same note if you switch. The durable fix below covers the rest.

Saved memory full: clear space

1. Open Settings in ChatGPT.
2. Go to Personalization, then Memory. (OpenAI moves these labels around; if yours differ, look for the memory section of settings.)
3. Review what ChatGPT has saved about you. Newer versions show a memory summary you can read and correct; there is also a list of saved memories you can go through entry by entry.
4. Delete the entries you no longer need. Duplicates, stale projects, trips you already took. Or clear everything and start clean.

Two things people get wrong here:

• Deleting a chat does not delete its memories. Memories live in your memory settings, not in your chat history. If you want a fact gone, remove it there.
• Deleting memories does not break old chats. Past conversations stay intact. ChatGPT just stops knowing those facts going forward.

Before you clear anything, skim the list for facts you actually care about and save them somewhere you control. More on that in a minute.

Why saved memory fills up

ChatGPT's memory saves short facts about you as you chat: your name, your projects, your preferences. Since the Dreaming upgrade it also curates these in the background from past conversations, which means memories accumulate without you asking. That curation does some housekeeping of its own, combining and dropping memories as they go stale, so the total is not purely one-way. It still accumulates faster than it prunes when you use ChatGPT for real work.

The storage is finite. Paid plans get more space than free, but every plan has a ceiling, and there is no setting to raise it. Clearing space is a reprieve, not a fix: the feature is doing its job, collecting crumbs from every conversation, with no sense of which ones matter to you.

Why the important stuff shouldn't live there anyway

Here's the uncomfortable part: the facts you most want ChatGPT to know are the worst fit for its memory.

Your positioning, your customers, your pricing decisions, your project history. That's not a handful of fact crumbs. It's documents. Memory was never designed to hold documents, which is why it fills up the moment you try to make it carry your working context.

And even when a fact fits, you can't organize it into folders, version it, or take it with you to Claude or any other assistant. It's a buffer inside one app. Buffers overflow. That's what the notice is telling you. ChatGPT's memory is genuinely good at personalization. It's just not a place to keep knowledge.

The durable fix: memory for preferences, a knowledge base for knowledge

The setup that stops the limit from mattering:

• Let memory hold the light stuff. Your name, your tone preferences, the fact that you hate bullet-point answers. Small, personal, fine to lose.
• Put your knowledge in a knowledge base ChatGPT can search. Connect it over MCP and ChatGPT looks notes up when they're relevant instead of storing them in a fixed allowance. There's no ceiling, because nothing is being crammed into ChatGPT. Your notes stay in your knowledge base; ChatGPT reads them on demand.

Notes you can read, edit, organize into folders, and export. When a fact changes, you change the note, and every future conversation gets the current version. The same notes work in Claude and any other MCP client, so your knowledge isn't trapped in one app's settings screen. See the full Hjarni vs ChatGPT memory comparison for how the two differ in practice.

Set it up in five minutes

1. Create a free Hjarni account and save the facts you rescued from your memory settings, or the context you keep re-explaining in new chats, as a few Markdown notes.
2. Install Hjarni from the ChatGPT Apps Directory in one click, then sign in. The setup guide covers the developer-mode path too.
3. Test it: "Search my notes for our positioning and use it to rewrite this intro."

From there, build it out into a proper second brain: the five notes you keep re-explaining, a decisions log, your style guide.

ChatGPT's memory can fill up as often as it likes. Your knowledge doesn't live there anymore.