Notes on coding agents and the pages they leave behind, on moving between notes apps, and on the plain mechanics of keeping what you think. Written and published from a Sumibako vault.
ChatGPT says memory full. Which memory it means, and what to do
Memory full in ChatGPT now means two different things. How to tell which one you have, what OpenAI changed since 2025, and what to do about each.
Prashant Shaw
Founder, Sumibako
8 min read
An elephant never forgets. ChatGPT, it turns out, occasionally runs out of room to remember, and tells you so in two words: memory full. Since late September those two words can mean two quite different things, depending on where they appear, and the fix for one does nothing for the other. Somewhere right now a person is deleting their favourite memories to rescue a chat those memories had nothing to do with.
So: how to tell which one you have, what OpenAI changed in the last year (a lot), and what to do about each. Everything links to OpenAI's own pages where they exist. It was checked on 18 October 2026, and updated on 22 October after a bug put the second kind in short chats. I build a notes app, which comes up once near the end.
NOTE:Updated 22 October 2026. From about 15 October, and in a rush on 20 and 21 October, people on Free and paid plans saw "Chat memory full" in short and even brand-new chats, some with the send button greyed out. One forum thread collected more than a hundred replies in two days. On 21 October, OpenAI Support said it was "investigating reports" of the message "blocking short or newly started chats", had "applied a mitigation" and was "monitoring recovery". Within the hour, people in the same thread wrote that "my chats that were bricked are back to working normally". By the next day most people in the thread reported it gone, though a few still saw it. If it happened to you in a short chat, it was almost certainly that bug rather than your conversation. The section near the end says what to do if it is still happening.
🐘 Two messages, one pair of words
Where you see it
What is full
What fixes it
A notice about saved memory, and the list in Settings
Your saved memories: the facts ChatGPT keeps about you, across every chat
Deleting or merging memories, or moving to improved memory
Inside one conversation: "Chat memory full"
That one conversation
A new chat. Your saved memories are not involved
The second one is new. Since about 30 September, people have been posting it in OpenAI's forum: "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." OpenAI's support account explained it on 2 October: the message "indicates that the conversation reached its available memory/context limit". One person in that forum put it best: "I'm not talking about the saved Memory feature being full. I'm talking about conversations becoming unusable out of nowhere."
Saved memories used to be a hard wall. OpenAI's help page, as archived in September 2025, called them a "notepad" and said: "Once your memory storage is full, ChatGPT won't save new memories until you clear space." It never said how big the notepad was, and it still has not. The word counts quoted in other guides are readers' estimates, and I am not going to add a fresh one to the pile.
15 October 2025. Automatic management, for Plus and Pro on the web. ChatGPT keeps "the most relevant details prioritized" and moves less important ones to the background, which "helps avoid 'memory full' state". You can turn it off, and pick what stays on top yourself.
4 June 2026. Improved memory, for Plus and Pro in the US first. Memories "are now updated automatically", the old list stays available for anyone who prefers it, and Plus and Pro get "twice as much memory capacity". Free and Go were to follow "over the next few weeks".
12 June 2026. More controls for the memory summary, with the note that memory updates are "still rolling out across ChatGPT plans and countries."
The pricing page lists memory as "Limited" on Free and "Expanded" on Plus and Pro, again without a number. And the current memory page no longer mentions memory being full at all. That does not mean it cannot happen. It means the list that fills up is now the old way of doing things, and some accounts are still on it.
🔍 Which one do you have?
Open Settings, then Personalization, then Memory.
If you see a memory summary, a description of you that you can correct by typing into it, you are on improved memory, which OpenAI describes as a summary rather than a list. Its page says nothing about that summary filling up.
If you see Saved memories, a list of separate entries, you are on the legacy list, which is the one that fills. Where the option exists, OpenAI's page says you can select "Try improved memory" to switch.
If you chose the list on purpose because you like knowing exactly what a machine remembers about you, that is a perfectly reasonable thing to want, and the next section is for you.
🧹 How to clear ChatGPT's saved memories
The quickest wins, in order:
1️⃣ Duplicates. A topic you talk about often leaves several near-identical entries. Keep the best one.
2️⃣ Finished things. The trip you planned, the job you left, the kitchen you renovated.
3️⃣ Things that are no longer true. These do more harm than the rest, because ChatGPT will keep using them with total conviction.
4️⃣ Long entries. A memory that is a paragraph is a document wearing a name badge. It belongs somewhere else, which is the next section.
On Plus and Pro on the web, also check that automatic management is on. The release notes say you can see which memories are top of mind and "prioritize or deprioritize any specific memory yourself in settings".
Two things that surprise people, both from OpenAI's memory page:
Deleting a chat does not delete what it taught ChatGPT: "Deleting a chat alone does not necessarily delete a separate saved memory created from that chat."
Deleting a memory does not delete it everywhere: it "does not remove mentions of that information from past conversations", and "OpenAI may retain logs of deleted saved memories for up to 30 days".
You can also hand the tidying to ChatGPT:
Treat the answer as a draft. The list in Settings is the truth, and checking there is quicker than arguing.
📦 What does not belong in memory
OpenAI's old word for saved memories, notepad, is the right one. Memory is for small, stable facts: your name, your job, that you like short answers. Anything long, exact or tied to one piece of work belongs somewhere built to hold it:
What it is
Put it in
Why
How every answer should sound
Custom instructions
Applied to every chat. 1,500 characters on Free and Go, 5,000 on paid plans
One project's background, files and rules
A project's instructions and files
Only that project's chats see it
Things you mentioned in passing
Reference chat history
OpenAI's memory FAQ: "There is no separate storage limit for what ChatGPT can reference through chat history"
There is one more reason to keep memory lean. The pricing page says each chat's context window is shared with "memories (if enabled)". Every memory rides along in every conversation, taking up room the conversation could have used.
💬 If it said "Chat memory full"
Nothing above will help, because your saved memories were never the problem.
First, rule out the October bug. If the chat is short, reload the page and try it again: OpenAI said on 21 October that it had applied a fix, and people reported blocked chats coming back. If a short chat is still blocked, add your report to the forum thread, which is where OpenAI has been answering.
If the chat really is long, it ran out of room. The word "upgrade" in the message is not new, by the way: it has been there since the message first appeared in late September. The pricing page does give paid plans a bigger window, 54K tokens for everyday answers on Go and Plus against 27K on Free, which moves the wall further away rather than removing it. I put every limit OpenAI publishes on one page.
Either way, press "Continue in a new chat" if it is offered, which starts a new conversation with an automatic summary of the old one. Then ask the new chat to tell you in two sentences where things stand, to check the summary it was handed. Next time, ask for a handoff note before the wall arrives. I wrote up the whole routine, template included.
📍 Keep the long stuff in a document
Both messages teach the same lesson: ChatGPT's memory is a convenience, not storage. Anything you would be upset to lose belongs in a document you own and can read without ChatGPT: a file, a project, a note. The same goes for Claude's memory, which also summarises rather than stores.
That last one is where I come in, so weigh this paragraph accordingly. Sumibako is a notes app that AI apps can read and write pages in. On ChatGPT, OpenAI's page on custom servers allows that on Business, Enterprise and Edu, and reading only on Pro; on other plans you paste. Claude can read and write it on every plan, Free included, through a custom connector. Either way the page is yours: you can open it, fix it, and hand it to whichever assistant you use next year. It has no AI of its own and will not remember anything by itself, which in this particular post may count as a feature.
An elephant never forgets, but it also never tries to remember everything. Keep the notepad short and the rest in a notebook.
Eight eMudhra keys were integers under 256, one of them 4. Why browsers blocked them in hours, why your Docker image did not, and what to check by 27 October.
10 min read
List every saved memory you have about me, word for word, grouped by topic.
Then suggest which to delete because they are duplicates, finished or no
longer true, and which to merge into one shorter memory. Change nothing
until I tell you which suggestions to apply.
