Can Claude and ChatGPT share one memory?

Neither product can read the other’s memory. Claude’s memory is built from your Claude conversations and lives in your Claude account, ChatGPT’s memory is built from your ChatGPT conversations and lives in your OpenAI account, and there is no setting on either side that points one at the other. If you use both, anything you want both of them to know has to live in a store outside both products, such as a Wire container.

Why this happens

Memory is built as a retention feature, so it is scoped to one account inside one vendor by design. OpenAI’s own documentation says memory is tied to each individual account and is not transferable to other users, even within the same Business workspace, and that other members of a workspace cannot see your memories. Anthropic scopes Claude’s the same way: a Team or Enterprise owner can turn memory on for the organization but cannot view or edit any individual member’s memories. The boundary is per account first and per vendor second, which catches one person holding two subscriptions as squarely as two people on different tools.

Recent changes inside each vendor have not softened that boundary. On August 25, 2026 Anthropic merged the memory used by Claude chat and Claude Cowork, so what Claude learns in a chat is available when you hand it a task in Cowork in the cloud, and back again. That is one vendor connecting its own two products because the gap between them was costing it something. The gap facing a competitor’s product is unchanged, which is the same reason memory lock-in stops working as a moat once a vendor needs you to adopt a second product.

Both vendors offer a way to move your data out: an export or an import, one direction, at one moment. That satisfies data portability and changes nothing about running two assistants at once, because as soon as the copy finishes you have two stores that disagree. The property you want is context portability, which a one-time copy does not give you.

What each product’s memory actually holds

Neither memory holds what you are working on, it holds you. Claude’s spans chat and Cowork when Cowork runs in the cloud, and is readable, editable, and deletable by the person it belongs to. ChatGPT’s is described by OpenAI as a continually updated synthesis of context from your past chats, broader than the list of items the settings page shows. Both are good at preferences, style, and the kinds of things you tend to ask about. Whether either one is enough on its own is a different decision, answered per product in Claude’s memory for a team and ChatGPT’s memory for a team.

Neither holds the specifics of what you are planning. Say it is a ten-day trip in April: the dates, the $4,200 ceiling, the guesthouse you ruled out on Sunday because it is a fourth-floor walk-up and your partner’s knee will not take it, the shellfish allergy that shapes every restaurant list. You settle that with Claude on Sunday, ask ChatGPT about neighborhoods on Wednesday, and it hands the walk-up straight back to you, because it never saw the reason. When people ask whether the two assistants can share memory, that is usually what they mean, and neither product has anywhere to put it.

Your options

One store both assistants connect to is the only option here where a fact written once is readable by both and stays readable as it changes. This is what a Wire container is: one place outside both products that each assistant connects to over MCP, searches before it answers, and writes back to when something is settled, so the ruled-out guesthouse comes back with the stairs attached. The cost is the ChatGPT side, where a custom connector needs developer mode, which OpenAI lists for Pro, Plus, Business, Enterprise, and Education accounts on the web. Both assistants also need standing instructions, because left alone each writes to its own memory.

Claude’s memory import from ChatGPT does one narrow job well. Anthropic’s flow gives you a prompt to paste into ChatGPT, which produces a summary of your instructions, preferences, and the tools you use, and you paste that into Claude’s memory settings. It is available on Free, Pro, Max, and Team. It runs once, from ChatGPT into Claude and not back, and Anthropic labels it experimental. It carries what an assistant knows about you, not what you decided on Sunday.

A shared document both are told to read is the cheapest option, and it is enough in plenty of cases. A Notion page, a Google Doc, or a note on your phone, attached at the start of a conversation, gives both assistants the same ground truth. It works on any plan and on mobile. It fails in the two ways documents always fail. It goes stale, because updating it competes with the thing you are actually doing. And only you can write to it, so nothing either assistant works out in a conversation gets back into it.

Re-explaining it each time is what most people are doing right now, and it is defensible when the context fits in a paragraph. It stops being defensible once you switch between the two assistants regularly, because then the cost is no longer the typing. You describe it differently on Wednesday than you did on Sunday, get an answer that contradicts what you settled, and spend longer untangling that than a lookup would have taken.

How to decide

How often does what you are planning change? If it is settled and you only need both assistants to know the rules, a document is fine and the setup cost of anything else is not worth paying. If it changes as you go, which is the usual case while you are still deciding things, a document is out of date by the time you attach it, and you want a store both assistants write to.

Are you genuinely using both assistants on it? One assistant and a good note gets you a long way, and a one-time memory import covers the preferences. The case for a shared store starts when you are switching, because switching is when one of them can hand you back something you already ruled out. The same test one step up settles whether anyone else needs in: if someone is planning alongside you, they read the same container from whichever assistant they use, which is the team context case rather than a different build.

Can you add a connector on the ChatGPT side? This is usually what decides it. A custom MCP connector needs developer mode on a paid plan, on the web client. If your ChatGPT account is on Free, or you only ever open it on your phone, the honest recommendation is the shared document until that changes. Answer yes here and yes to the two questions above, and the option to take is the first row, which is Wire.

What to do next

If what you are planning changes as you go and you use both assistants on it, the next step is to set up one store and connect both of them to it.

If you are not ready for that, do the cheap option properly: one document holding the decisions and their reasons rather than a description of the thing. “Not the Alfama guesthouse, fourth-floor walk-up, bad knee” is still usable in three weeks; “looking at Alfama” is not. Attach it on both sides, and update it in the sitting where you decide something rather than promising to do it later. For the one-assistant case, moving your AI context between tools covers what the vendor import flows do and do not carry.

Wire is our version of the shared store. Both assistants connect to one container over MCP, read the same entries, and write new ones back, which is the AI portability case. It earns its place under the conditions above and not otherwise. If what you are planning is small and settled and one assistant does all of it, a document will beat any tool on this list, including ours.

Sources: Claude’s chat search and memory · Import and export your memory from Claude · Custom connectors using remote MCP · ChatGPT developer mode · OpenAI Memory FAQ (Business)