To turn off ChatGPT memory for a single Project and wall it off from everything else, open the project, click the three-dot menu, choose Project settings, and set Memory to Project-only. That one switch is how you stop ChatGPT mixing projects, and since an August 2026 update you can flip it on an existing project, not just at the moment you create one. Project-only memory tells ChatGPT to draw context from conversations inside that project alone, and to keep anything it learns there out of the memory it uses in every other chat.
The point is compartmentalization. Your job-hunt project stops leaking into your recipe chats, your therapy-style journaling stays out of your work drafts, and a client project cannot pull in details from a different client. Below is exactly what the setting does, how it differs from saved memories and reference chat history, the step-by-step to switch a project over, and how to review or delete what ChatGPT already remembers.
What project-only memory actually does
Project-only memory builds a two-way wall around one project. When it is enabled, ChatGPT can use other conversations inside that same project for context, but it will not use your saved memories or chats from outside the project to shape its answers. Anything it picks up while you work in the project also stays inside the project rather than updating your account-wide memory.
OpenAI first introduced project-only memory in August 2025, but at launch you could only set it when you created a project, and only on the ChatGPT website and the Windows app, with the other apps following in the weeks after. The August 2026 change is the practical one: for eligible unshared projects you can now switch an existing project between default and project-only memory from inside Project settings, and the control works across ChatGPT plans.
A concrete example makes the wall clearer. Say your global saved memories know you are vegetarian and that you prefer short answers. Inside a default project, ChatGPT can lean on both. Inside a project-only project, it will not: it answers using only what has been said in that project, and if you tell it something new there, like a client's product names or a confidential budget, none of that gets written back to your account-wide memory. The isolation runs in both directions at once.
Two limits are worth knowing before you rely on it. Shared projects stay on project-only memory and cannot be switched back to default, and ChatGPT Work features are not available inside a project that uses project-only memory. So this is a trade: tighter isolation in exchange for a couple of capabilities.
Saved memories, reference chat history, and project-only memory are three different things
Most confusion about ChatGPT memory comes from treating it as one feature. It is three, and they behave differently.
- Saved memories: an explicit list of facts ChatGPT has stored about you, such as your name, job, or preferences. You can open this list, read it, edit it, and delete individual entries. It is the only part of memory you can actually audit.
- Reference chat history: implicit recall, where ChatGPT draws on patterns from your past conversations without keeping a browsable list. There is no fixed set to inspect, and what it holds onto shifts over time based on what it finds useful.
- Project-only memory: a scope setting, not a third store of facts. It decides whether a specific project can reach your global memory at all, in either direction.
Saved memories and reference chat history are each toggled independently under Settings, Personalization, Memory. Project-only memory lives one level down, inside an individual project, and overrides those global settings for that project alone. You can leave global memory fully on for casual use and still keep one sensitive project sealed off.
Saved memories are auditable and reference chat history is not. If you want to know exactly what ChatGPT thinks it knows about you, the saved-memories list is the only part you can read line by line.
How to switch a project to project-only memory, step by step
Here is the direct path to stop ChatGPT mixing one project with the rest of your account. It takes under a minute, and it assumes you already have a project set up. Projects are the folders in your sidebar that group related chats and files together, so if you have been keeping a topic in loose one-off conversations, move those into a project first, then apply the setting below.
- Open the project you want to isolate from your sidebar.
- Click the three-dot menu on the project (More options).
- Choose Project settings.
- Under Memory, select Project-only memory instead of Default memory.
- Confirm, then give it time. Changes to a project's memory setting can take several hours to fully take effect.
The Memory choice can be easy to overlook. In some layouts the option reads like plain text rather than an obvious toggle, so look for Default memory versus Project-only memory under the Memory heading in Project settings rather than a big switch.
If you are starting fresh, you can also set project-only memory at creation time when you make a new project. Either way, the effect is the same: that project draws only on its own conversations, and its context does not flow back into your global memory. Remember that a shared project cannot be moved off project-only memory, so decide before you share.
How to turn off ChatGPT memory entirely
If you want ChatGPT to stop remembering anything across chats, you turn memory off globally rather than per project. Go to Settings, then Personalization, then Memory, and switch off both Reference saved memories and Reference chat history. Each one is a separate toggle, so you can disable just the implicit recall while keeping your explicit saved facts, or shut both down completely.
For a one-off conversation where you do not want any memory involved, start a Temporary Chat. It bypasses memory entirely: ChatGPT will not read from your saved facts or chat history for that session, and it will not add anything new. Think of project-only memory as the durable setting for an ongoing workspace and Temporary Chat as the quick option for a single throwaway question.
When to use each mode
Match the mode to the stakes. Default memory is convenient for everyday chatting where cross-context recall helps more than it hurts. Project-only memory is the right call whenever bleed-through would be awkward or risky. The test is simple: if a detail surfacing in an unrelated chat would embarrass you or breach a confidence, that work belongs in a walled-off project.
- Keep default memory for casual, personal use where you want ChatGPT to remember your general preferences.
- Use project-only memory for client or freelance work, so one client's details never surface in another's project.
- Use project-only memory for sensitive topics like health, finances, or a job search you would rather keep out of unrelated chats.
- Use Temporary Chat for a single sensitive question you do not want stored at all.
- Turn memory off globally if you simply prefer ChatGPT to treat every conversation as a blank slate.
How to review and delete what ChatGPT remembers
You can review and clear ChatGPT's memory, but only the part that keeps a list. Open Settings, Personalization, Memory, and use Manage memories to see your saved memories. From there you can delete individual entries or clear them all. This is the auditable layer, so it is where you go to check and correct what ChatGPT believes about you.
Reference chat history is the harder part, because there is no browsable list of what it has absorbed. You cannot open it, read it, and prune specific items the way you can with saved memories. The realistic controls there are to turn reference chat history off, use Temporary Chat, or wall a project off with project-only memory so that history cannot reach it in the first place.
Do a quick sanity check after you change a project's setting. Because the switch can take several hours to take effect, do not assume isolation is instant. A simple way to test it later is to ask the project a question that only your global memory could answer, such as a personal preference you never mentioned in that project. If project-only memory is active, ChatGPT should not know it. If it still does, wait and try again before you trust the project with anything sensitive.
Compartmentalized memory you actually own
Project-only memory is a good instinct pointed at a limitation. It exists because default ChatGPT memory is one shared pool, and the moment you have anything you would rather keep separate, you have to reach for a switch that walls part of it off. Isolation is the exception you opt into, not the default, and the implicit half of that memory stays something you cannot fully read back.
MemX is built the other way around. It is a consumer memory app where your documents, photos, voice notes, and messages are stored as your own memory, and each item is a distinct thing you can see, search, and remove, not one opaque blob that quietly blends everything together. You ask a question and get an answer with the source it came from, so recall is always something you can check. Compartmentalizing is the model, not a setting you have to remember to flip.
It is private by architecture: per-user isolation, encryption at rest, and processing kept on-device where possible, with no training on your data. That is a deliberately narrower promise than end-to-end encryption or zero-knowledge, and MemX does not claim those. The honest pitch is simpler: memory you can see, isolate, and delete, so keeping two parts of your life apart does not depend on finding the right menu.
| What you get | ChatGPT project-only memory | MemX private memory |
|---|---|---|
| Isolation model | Opt-in per project; default memory is one shared pool | Each item is separate by default; you search across what you choose |
| What you can audit | Saved memories are viewable; reference chat history is not | Every stored item is visible, searchable, and removable |
| Answers with sources | Recall is implicit; no source shown for what it remembered | Answers return the source item they came from |
| Data ownership | Stored in your ChatGPT account under OpenAI's controls | User-owned, per-user isolation, encryption at rest, no training on your data |
01How do I turn off ChatGPT memory?
Go to Settings, then Personalization, then Memory, and switch off both Reference saved memories and Reference chat history. They are separate toggles, so you can disable one or both. For a single session with no memory at all, start a Temporary Chat instead.
02What is the difference between saved memories and reference chat history?
Saved memories are an explicit, editable list of facts ChatGPT stored about you, and you can read and delete them. Reference chat history is implicit recall from past chats with no browsable list, so you cannot inspect specific items, only toggle it off.
03How do I stop ChatGPT mixing my projects?
Open the project, click the three-dot menu, choose Project settings, and set Memory to Project-only. That project will then use only its own conversations for context and will not reach your global memory or push anything back into it.
04Can I change an existing project to project-only memory?
Yes. Since an August 2026 update, eligible unshared projects can be switched between default and project-only memory in Project settings. Shared projects stay on project-only memory and cannot be changed, and the switch can take several hours to take full effect.
05Does project-only memory delete my existing memories?
No. It changes scope, not storage. Your global saved memories stay intact; the project simply cannot use them, and what happens inside the project stays out of your account-wide memory. To remove saved facts, use Manage memories under Settings, Personalization, Memory.
