ChatGPT’s saved memory is small — roughly 1,200 to 1,400 words — so heavy users fill it within days, and it keeps replying “memory updated” after it is full. OpenAI does not sell extra memory at any tier. JustOnce is a separate memory layer with a visible meter, full history, and nothing discarded silently.
Because the saved-memory store is small. Our July 2026 research put it at roughly 1,200 to 1,400 words in total — a few pages of notes for your entire relationship with the product. Anyone using it for a real project, a long piece of writing or a running journal fills it inside a week. Deleting entries buys days, not months, which is why the same people are back deleting again the following week.
Because the confirmation and the write are not the same thing. This is the failure people find hardest to forgive, and it is documented on OpenAI’s own community forum: “When the memory is full, the system does not provide any warning, yet ChatGPT still displays a message indicating that new information has been successfully remembered. However, these entries are not actually stored.” You believe it remembered. It did not. You find out weeks later, when it confidently gets something wrong.
Quoted from the OpenAI developer community thread on the missing memory-full warning, first posted February 2025 and still collecting replies in 2026.
People report that it does not. There are public accounts of users escalating from the US$20 plan to the US$200 plan specifically to get more memory, and hitting the same message again the next day. As of our July 2026 check, OpenAI does not sell extra memory at any tier — a thread on its own community forum is titled “I’ll pay for more memory!” and it is still open. The cap is not a billing setting you have not found. It is the size of the store.
Two things. First, there is no hidden ceiling that silently swallows a write: what you save is saved, and what you can add each month is a number you can see on your own usage page. Second, nothing is overwritten. When a fact changes, the new value supersedes the old one and the old one stays in the history — so “it forgot” and “it quietly replaced what I told it” both stop being possible failure modes.
Memories are bitemporal and verbatim-first: your exact words are kept alongside the distilled version, corrections supersede rather than erase, and every value a fact has held is retrievable.
No, and we are not going to pretend otherwise to sell against a cap. Paid plans meter how much you add each month — memories and document pages — and you can see the meter. What they never do is discard something you already saved, refuse a save while claiming success, or delete anything when you reach your allowance. Recall and search stay free on every plan, so nothing you have stored ever becomes unreadable because of a billing state.
A memory holds up to 1,000 words and a page is 2,000 characters of document text; longer items simply use more of the allowance. Nothing is rejected or deleted.
Ask ChatGPT to print out its saved memories, then hand that list to any assistant you have connected to JustOnce and tell it to import them. It is a five-minute job, and it is the first thing most people do. Every new account gets a one-time welcome bonus of extra memories precisely so the import does not eat your first month.
Bulk import runs 50 memories per call, each one classified and entity-extracted on the way in. The welcome bonus is 100 extra memories on Free, 1,000 on Plus and 2,500 on Pro, valid for 90 days.
Paste these into any assistant you've connected to JustOnce. They work as written.
Bulk-loads them 50 at a time, classifying each one and extracting the people, organisations and projects mentioned.
Here is the list of saved memories ChatGPT had about me. Import all of them into JustOnce, keeping my original wording: [paste the list]
Returns your categories with counts and the memories in them, so you can compare against what you pasted.
List everything you now have stored about me in JustOnce, grouped by category, so I can check the import.
Distils a long paste into separate durable memories rather than one blob, so each fact can be recalled, corrected or removed on its own.
Remember all of this before I lose it: [paste the context you have been re-explaining in every chat]
Pulls your verified, high-confidence facts as a compact block — the standing context a fresh chat would otherwise be missing.
Load my core profile from JustOnce before we start.
Supersedes the memory. The previous value stays retrievable, so past documents still make sense.
You have my old job title stored. Update it — I moved to a new role in June. Keep the old one in the history.
You cannot buy more from OpenAI — as of July 2026 no tier sells extra memory, and the US$200 plan has the same store as the US$20 one. The practical options are to keep pruning entries by hand, or to keep your memory somewhere outside the chat product and let ChatGPT read it through a connector. JustOnce is the second option.
New entries can silently fail to save while the interface still says “memory updated”, and users have reported months of accumulated context disappearing without warning. There is no notification and no restore. This is the specific behaviour documented in OpenAI’s community forum threads.
It sits alongside it. You connect JustOnce to ChatGPT as a connector and it becomes the durable store, while built-in memory keeps doing whatever it does. The difference is that JustOnce also serves Claude, Cursor and any other MCP client, so the same facts follow you when you switch.
Yes. Every memory is visible in the web app, with your original wording kept alongside the summarised version. You can edit it, scope it, or forget it — and forgetting is a reversible invalidation, not a shredder.
No. There is no eviction, no least-recently-used pruning and no silent overwrite. When a fact changes, the new value supersedes the old and the old one remains in the history. When you reach a monthly allowance you simply add less that month; everything already saved stays searchable.
Plus is US$8.99 a month and Pro is US$19.99. Free covers 20 memories and 10 document pages a month with two AI-assistant connections, and recall and search are free on every plan. It is a small addition to an AI subscription you already have, rather than a replacement for it.
It is one connector per tool. Claude Code takes a single command, Cursor and VS Code have one-click deeplinks, and ChatGPT and Claude Desktop take a pasted connector URL. Sign-in is OAuth, so there is no API key to manage.