ChatGPT Memory Full: Why and How to Fix It
The "ChatGPT memory is full" message belongs to the saved-memories list, which has a fixed storage cap. As of June 2026, Plus and Pro accounts default to the new "dreaming" system that synthesizes memory automatically from past chats and doubled paid capacity, so the full message mainly affects accounts using the opt-in legacy saved-memories list. Fix it by deleting or editing entries in Settings, Personalization, Manage memories, or switch to the dreaming default.
What "ChatGPT memory is full" actually means
The message means the saved-memories list has reached its storage cap. Saved memories are facts you explicitly asked ChatGPT to remember, such as your name, your job, your writing style, and ongoing goals. When that list reaches capacity, ChatGPT shows a notice like "Memory full" and stops adding new entries until you make room.
This cap applies to the saved-memories list specifically, not your chat history and not the newer synthesized memory. A full list does not erase old chats and does not stop ChatGPT from working. It only blocks new explicit saved entries. Since June 2026, Plus and Pro accounts default to the dreaming system described below, so the full message now mainly shows up for accounts that opted back into the legacy saved-memories list.
- Trigger: the saved-memories list hit its storage cap.
- Effect: no new saved entries until you clear space.
- Not affected: existing chats and dreaming-synthesized memory continue to work.
The June 2026 dreaming update changes the default
On June 4, 2026, OpenAI began rolling out a new memory system, informally called dreaming, to Plus and Pro users in the United States, with Free and other users to follow. Instead of relying on a manually curated saved-memories list, dreaming synthesizes what ChatGPT remembers about you in a background process that reads across your past conversations and updates a single coherent memory state.
Two practical effects matter for the full message. First, dreaming memory is self-updating, so it revises stored details as time passes rather than piling up permanent one-off entries. OpenAI's own example rewrites "going to Singapore in July" to "you went to Singapore" once the date passes. Second, OpenAI doubled memory capacity for Plus and Pro accounts. For users on this default, hitting a fixed saved-memories cap is far less common.
- Dreaming rolled out June 4, 2026 to Plus and Pro in the US first.
- Memory is synthesized from past chats, not a curated saved list.
- Plus and Pro memory capacity was doubled (2x).
Where to review what ChatGPT remembers
The way you audit memory depends on which system your account uses. On the dreaming default, the main transparency surface is the new memory summary page, where you can read what ChatGPT has synthesized about you and edit, add, or steer details. A separate memory sources view, now available across consumer plans, shows some of the context used to personalize a given response, such as past chats, saved memories, custom instructions, or connected data, so you can see what informed an answer and correct it.
On the legacy saved-memories list, the audit surface is the explicit entry list in settings, which you can view and prune item by item. Reference chat history remains dynamic context drawn from past chats and is not shown as a fixed list. The takeaway: the saved-memories list is no longer the single place to see what ChatGPT knows about you on current default accounts.
- Dreaming default: memory summary page plus memory sources view.
- Legacy mode: the explicit saved-memories list in settings.
- Reference chat history stays dynamic and is not shown as a list.
How to fix it: clear or edit saved memories
If you are on the legacy saved-memories list and see the full message, the direct fix is to remove entries you no longer need. Open Settings, then Personalization, then Manage memories. From there you can delete individual entries or use Clear ChatGPT's memory to remove every saved memory at once. You can also ask ChatGPT in chat to forget a specific detail.
One caveat trips people up. A saved memory lives separately from the chat that produced it. To fully remove information, delete the saved memory in Manage memories and also delete the chat where you shared it. If only the chat is deleted, the saved memory can still be used in future conversations.
- Path: Settings, Personalization, Manage memories.
- Delete entries one by one, or use Clear ChatGPT's memory for a full reset.
- For a complete wipe, remove the saved memory and the source chat.
Why the legacy list fills, and how to avoid it
The fixed-cap full message comes from the saved-memories model, where every confirmation to remember something adds a permanent entry. Casual or one-off facts accumulate alongside genuinely useful ones, and project notes, temporary preferences, and outdated goals all consume the same fixed space. The dreaming system avoids most of this because it synthesizes and revises memory automatically rather than holding a fixed list of confirmations.
If you stay on the legacy list, be selective about what you ask ChatGPT to remember, review the list periodically, and prune stale entries. Switching to the dreaming default on Plus or Pro removes most of the manual upkeep and gives you doubled capacity. For broad recall across many notes and documents, a dedicated memory layer fits better than either ChatGPT mode.
- The full state is a legacy saved-memories behavior, not the dreaming default.
- On the legacy list, prune outdated goals and one-off facts regularly.
- Dreaming synthesizes and revises memory, reducing manual upkeep.
Durable alternatives for long-term recall
ChatGPT memory, whether dreaming-synthesized or a capped saved list, is tuned for assistant personalization rather than holding everything you want to remember long term. If your goal is to capture notes, decisions, and documents and recall them later, an external memory layer is a better fit than a personalization feature.
MemX, by Neural Forge Technologies, is an AI memory app built for exactly that: a personal memory and recall layer where you store information and retrieve it on demand. It does not replace ChatGPT as a chat assistant; it complements it for the personal-memory and recall use case. MemX is private by architecture, with per-user isolation, encryption at rest, Google Cloud KMS, and on-device handling. Use ChatGPT memory for assistant personalization, and a dedicated memory layer when you need durable, searchable recall.
- ChatGPT memory: tuned for assistant personalization.
- External memory layer: built for durable storage and search.
- MemX is private by architecture with per-user isolation and encryption at rest.
Key takeaways
- "Memory full" is a saved-memories cap message, not a sign ChatGPT stopped working or lost your chats.
- Since June 4, 2026, Plus and Pro default to the dreaming system that synthesizes memory from past chats and doubled paid capacity.
- On the dreaming default, the memory summary page plus memory sources are the main places to review and correct what ChatGPT remembers.
- The legacy saved-memories list is now opt-in; fix a full list in Settings, Personalization, Manage memories, and delete both the memory and its source chat for a full wipe.
- For durable, searchable long-term recall across notes and documents, an external memory layer fits better than either ChatGPT mode.
Frequently asked questions
Related reading
Sources
Stop fighting your tools
MemX is an AI memory app: store anything, skip the folders, and find it again by asking in plain English. Private by architecture, with per-user isolation and encryption at rest.
Try MemX Free