Skip to content
Membase
Cross AI Memory: How to Use One Memory Across ChatGPT, Claude, Cursor and More (2026)

Cross AI Memory: How to Use One Memory Across ChatGPT, Claude, Cursor and More (2026)

Research
MembaseMembase Team
View the original post on Medium

You tell ChatGPT about your project on Monday. On Tuesday you open Claude, and it has never heard of it. Your coding assistant knows your codebase but not the decision you made in a chat last week. Every AI you use keeps its own memory, and none of them share it.

Cross AI memory fixes that: one memory that every AI you use can read, so you explain things once.

Quick answer

There are three ways to carry what one AI knows into another:

MethodWhat it doesGood forWhere it stops
Copy and pastePaste notes or instructions into each AIA few facts, onceGoes stale, and you repeat it in every tool
Export and importMove memory or chat history from one assistant to anotherSwitching assistantsA one-time copy, not kept in sync
A shared memory layerEvery AI reads one memory you controlUsing several AIs every dayNeeds a one-time setup per AI

If you use more than one AI every week, a shared memory layer is the only option that stays up to date. The rest of this guide shows all three, then sets up the third with Membase.

Why your AIs don’t share memory

Each assistant builds memory from its own chats and keeps it inside its own product:

None of these memories are visible to the other assistants. That is a sensible default for privacy, but it means your context is scattered across tools.

Option 1: Copy and paste

The simplest approach: keep a short note about yourself and your projects, and paste it into each assistant’s custom instructions or the start of a chat.

It works for a handful of stable facts, like your role or writing style. It breaks down as soon as things change: every update has to be made in every tool, and decisions from real conversations never make it into the note.

Option 2: Export and import

Some assistants can import what another one knows. Gemini, for example, can import a memory summary or your full chat history from ChatGPT or Claude. You can also export your ChatGPT conversations and bring the important parts elsewhere.

Step-by-step guides:

This is the right tool when you are moving from one assistant to another. It is a one-time copy, so anything you learn afterwards stays in the new tool only.

Option 3: One shared memory every AI can read

A shared memory layer sits outside your assistants. You keep your memory in one place, and each AI connects to it and reads it when it needs context.

Membase is a memory layer built for this. Your memory lives in your Membase account as named Memories, for example “Work”, “Ledger project” or “Writing”. Each Memory learns from what you add to it, and every AI you connect reads the Memories you switch on for it. Every answer it finds names the Memory it came from.

Step 1: Create a Memory and add your material

Sign in at app.membase.com, create a Memory, and add what it should know: notes, files, or your past conversations. To bring in conversations you have with assistants in your browser, use the Membase browser extension. It captures those conversations as they happen and adds them to a Memory as a source.

Membase learns new material in the background, so give it a short while before you test.

Step 2: Connect each AI

Every client below connects to the same address:

https://api.app.membase.com/mcp-http
AIHow to connect
ChatGPTPlugins > + > Add custom MCP server. Paste the address, choose OAuth, then approve access in the browser. In a new chat, type @ and select Membase.
Claude (web and desktop)Customize > Connectors > add Custom > Web. Paste the address and approve access. In a chat, open + > Connectors and switch Membase on.
Claude Codeclaude mcp add --transport http membase https://api.app.membase.com/mcp-http
CursorAdd the address to .cursor/mcp.json (one project) or ~/.cursor/mcp.json (all projects), enable Membase and approve access.
Codexcodex mcp add membase --url https://api.app.membase.com/mcp-http, then codex mcp login membase
GrokAdd a custom connector on grok.com with the same address.
Other MCP clientsVS Code, Windsurf and any client that supports remote MCP servers use the same address.

When you connect, Membase opens a consent screen in your browser. You tick which Memories that AI may use, and whether it may read your profile. Nothing else is shared.

Full setup for each client, including team and workspace plans, is at docs.membase.com/connect.

Step 3: Check that it works

In any connected AI, ask:

Which Membase memories can you use?

It should name the Memories you ticked. Then ask about a specific fact you added, and check the answer. If the list is empty, no Memory is switched on for that connection yet.

Step 4 (optional): Let an AI save to memory

Connections through the consent screen are read-only by design: an AI can search your memory but not change it. For an AI that should also save, such as Claude Code, Codex or Cursor’s agent, create a developer key under Connect > Developer keys and give it to that AI. You choose the key’s access level and which Memories it can reach, and you can revoke it at any time.

What about Gemini, DeepSeek and other assistants?

Gemini and DeepSeek are not on Membase’s list of connected clients today, so they cannot read your Membase memory directly. You can still bring your conversations with them into your memory through the Membase browser extension, and then read that memory from ChatGPT, Claude or your coding tools. Which assistants the extension captures is a setting in the extension.

For details on each assistant’s own memory, see Gemini Memory: How It Works and Does DeepSeek or Grok Have Memory?

Which method should you use?

  • You only need a few stable facts in one or two tools: copy and paste.
  • You are moving from one assistant to another: export and import.
  • You use several AIs every week and want them all to know the same things: a shared memory layer.

Many people combine them: each assistant’s built-in memory for small preferences, and one shared memory for projects, decisions and anything you don’t want to explain twice.

You stay in control

  • You decide which Memories each AI can use, per connection.
  • Connections through consent can only read.
  • You can change or revoke any connection on the Connect page, and the change applies on that AI’s next request.

Get started

For developers

Videos

Help

Frequently Asked Questions

What is cross AI memory?
Cross AI memory is one memory that several AI assistants can read, so what you tell one AI is available in the others. It usually sits outside the assistants, in a memory layer each of them connects to.
Can ChatGPT and Claude share memory?
Not on their own. Each keeps its memory inside its own product. To share, connect both to the same memory layer. With Membase, ChatGPT connects as a custom MCP plugin and Claude as a custom connector, both to the same address.
Can I sync memory across ChatGPT, Claude and Gemini?
ChatGPT and Claude can both read one Membase memory. Gemini cannot read Membase directly today, but the Membase browser extension can bring your Gemini conversations into your memory, so ChatGPT and Claude can use what you discussed there.
Is a shared AI memory private?
You choose which Memories each AI can use, and connections through consent are read-only. You can revoke any connection at any time.
Do I need to be a developer to set this up?
No. ChatGPT and Claude connect through their settings and a consent screen in your browser. Developer keys are only needed when you want an AI to save to memory.
What does Membase cost?
Membase is free during the beta. Paid plans are coming soon.
Back to all posts