How to share memory between ChatGPT and Claude
Every major AI assistant now remembers things about you — but each lab keeps that memory inside its own product. ChatGPT cannot read what Claude knows about you, Claude cannot read what ChatGPT has saved, and neither will show its memory to any other tool. Today you have three options: live with the silos, move your context by hand, or keep one memory you own and let every tool read that.
What each lab gives you
| Product | What it remembers | Where it works |
|---|---|---|
| ChatGPT memory | "Saved memories" you can inspect and delete one by one, plus "reference chat history" that lets ChatGPT pick up context from past conversations. Managed in Settings → Personalization → Memory. | ChatGPT only. Not exposed over the OpenAI API, not readable by other tools. |
| Claude memory & projects | A rolling memory Claude maintains about you (editable, with incognito chats excluded), plus Projects that carry their own instructions and knowledge. | Claude apps only. Memory is per-product and stays inside Anthropic. |
| Gemini personal context | "Personal context" drawn from your past Gemini chats, plus "Saved info" facts you set explicitly. | Gemini and Google AI surfaces only. |
What none of them do:
- Portability — no lab lets another product read its memory. Switch assistants and you start from zero.
- User ownership — the memory lives in the lab's account system, in the lab's format. Some of it (ChatGPT's chat-history references, Gemini's personal context) is not inspectable as plain entries at all.
- Cross-tool reads — none of them expose memory over an API that your other tools, editors, or agents can pull from.
The manual way: export and re-enter
You can do a real sync today with nothing more than copy and paste. It is worth doing even if you never touch another product — seeing what an assistant has stored about you is useful on its own.
- Inspect ChatGPT's saved memories. Settings → Personalization → Memory → Manage. Read the list and copy anything worth keeping — this is the most explicit, portable layer ChatGPT has.
- Copy ChatGPT's custom instructions. Settings → Personalization → Custom instructions. Two free-text boxes ("what would you like ChatGPT to know about you" and "how would you like it to respond") — these transfer almost verbatim into any other tool.
- Pull the full ChatGPT export. Settings → Data Controls → Export data. OpenAI emails you an archive of your account data, memories and chat history included. Useful when you want everything, not just the curated list.
- Copy Claude's side. In Claude's settings, view and edit the memory summaries it keeps; open each Project and copy its custom instructions and any knowledge documents you added.
- Check Gemini. Saved info lives in Gemini's settings; past-chat activity is at myactivity.google.com/product/gemini.
- Paste it into the other tool. The combined text goes into Claude's project instructions, ChatGPT's custom instructions, or Gemini's saved info — whichever you use.
The limitation is drift: this is a snapshot. The next conversation teaches ChatGPT something Claude will never see, and a week later the copies disagree. Keeping silos in sync by hand means repeating the export forever.
The portable way: one vault every tool reads
MyMemory replaces the loop with a single vault you own. Take the text you exported above — custom instructions, saved memories, project notes — and import it once at myclawn.com/mymemory. A distiller splits it into atomic entries (directives, facts, preferences, notes) that wait in a review queue until you approve them. From then on, any tool that can make an HTTP request reads the same compiled context:
curl -H "Authorization: Bearer mm_28dbde38035867722fed80de7a99a48f" \
https://www.myclawn.com/api/mymemory/contextThat is a live, read-only key on a public demo vault — run it. The response is { context_block, entries }: context_block is paste-ready text for a system prompt, ChatGPT custom instructions, or a Claude Project; entries is the raw list for structured use. MCP-capable agents (Claude Code, Claude Desktop, others) read the same vault through the mymemory-mcp server instead — setup is in the MyMemory docs.
After conversations, agents propose new memories into your review queue — nothing becomes memory until you promote it. So instead of three silos drifting apart, you have one explicit, inspectable memory that every assistant shares.
FAQ
Can ChatGPT read Claude's memory, or vice versa?
No. Each lab's memory is private to its own product. There is no setting in ChatGPT, Claude, or Gemini that lets another assistant read what it has remembered about you, and none of them expose memory over an API for other tools to consume.
Can I see everything ChatGPT remembers about me?
Partially. Saved memories are inspectable at Settings → Personalization → Memory → Manage, where you can read and delete individual entries. The "reference chat history" layer is not fully inspectable — the closest thing to a complete picture is the data export at Settings → Data Controls → Export data.
Does MyMemory write into ChatGPT's or Claude's built-in memory?
No. MyMemory is a separate vault that you own. Tools read it over HTTP or MCP, and agents can only propose new entries into a review queue you approve. It does not modify, sync with, or depend on any lab's built-in memory feature.
Is the demo API key on this page safe to use?
Yes. It is a read-only key on a public demo vault: it can fetch the context block but cannot write or propose anything. For your own data, create a vault at myclawn.com/mymemory and issue your own keys.
What about Gemini, Copilot, and other assistants?
The same pattern holds: every assistant keeps its own memory and none read each other's. Any tool that can make an HTTP request or run an MCP server can read the same MyMemory vault, which is what makes memory portable across all of them at once.