How can a team share one AI memory across ChatGPT, Claude and Cursor?

Short answer

Put one memory store outside the tools, connect ChatGPT, Claude, Cursor and your own agents to it, and give each person a private scope next to a shared one with clear read and write rights. The built-in memory of ChatGPT and Claude belongs to one person, and shared projects or instruction files such as CLAUDE.md and AGENTS.md only share context inside the tools that load them.

What each tool shares today

The setup that works across tools has four parts: one memory store outside the tools, every client the team uses connected to it, a personal scope next to a shared one, and read and write rights per person. Below is why the built-in options do not get you there, then the steps.

Every AI tool your team uses has some way to share context, but each one stops at its own edge. As of November 2026:

Tool What is personal What can be shared Limit
ChatGPT (Business, Enterprise, Edu) Memory, chat history Shared projects: chats, files and instructions Shared projects use project-only memory; up to 100 collaborators and 40 files per workspace project
Claude (Team, Enterprise) Memory, chats Shared projects: knowledge base and instructions Chats in a project stay private unless shared one by one
Claude Code ~/.claude/CLAUDE.md, CLAUDE.local.md, auto memory ./CLAUDE.md in the repository, managed policy CLAUDE.md; it can also read a repository’s AGENTS.md, which other coding tools use too Shared files only help people working in that repository; auto memory stays personal
Cursor Its own settings Project rules, AGENTS.md and team rules (Team and Enterprise) Each tool has its own loading rules; a shared file only holds what someone deliberately puts in it

Some details that matter:

  1. ChatGPT shared projects let ChatGPT draw on everything in the project, including chats, uploaded files and custom instructions, so people can pick up where others left off. Business, Enterprise and Edu users can invite individuals or workspace groups. A shared project is always set to project-only memory and cannot access a member’s personal memories.
  2. Claude projects on Team and Enterprise can be public to the organization or shared with specific people. Sharing gives access to the project and its knowledge base; your chats in the project remain private.
  3. Claude memory is per person. Owners can switch memory on or off for the whole organization but cannot see or edit individual memories.
  4. Claude Code reads CLAUDE.md files at several levels. A ./CLAUDE.md checked into the repository is shared with the team through version control, and a managed policy file applies to the whole organization. Anthropic notes that Claude treats these as context, not enforced configuration.

The misunderstanding: “we have a team plan, so we share memory”

A Team or Business plan gives you shared billing, admin controls and shared projects. It does not merge what each person’s assistant has learned. If your account manager taught Claude how a client likes proposals, the colleague who takes over the client starts from zero, even on the same plan. And a shared project only helps the people who open that project in that tool.

How to set it up, from simple to complete

  1. Write the stable rules down once. Tone of voice, client exceptions, how a quote is built. Put them in shared project instructions and, for code, a ./CLAUDE.md in the repository. Our guide to company context has a template.
  2. Use shared projects per client or process in the tool most of the team uses, so files and instructions live in one place.
  3. Agree on where corrections go. When someone corrects the AI, the rule should land in the shared instructions, not only in their own memory. This is the weak spot, because it depends on people remembering to do it.
  4. Connect every tool to the same memory. Once the team uses more than one assistant, you keep one memory with a layer outside the tools that each of them can read and write, for example through MCP:
    1. Pick one store outside the tools.
    2. Set per person what they may read and what they may add, and keep a personal scope apart from the shared one.
    3. Connect the two tools the team uses most, for example Claude and Cursor.
    4. Test it: person A gives a correction in one tool, person B asks about the same topic in the other. If the correction comes back and A’s private notes stay hidden from B, it works.

What this means for a team

Without step 4, you maintain the same knowledge in several places: ChatGPT project instructions, Claude project knowledge, a CLAUDE.md, Cursor rules. They drift apart. Switching tools costs context, and a new colleague inherits nothing of what the others built up only in their personal memory. For ChatGPT specifically, see Does ChatGPT Business share memory across a team?. Read more in What are the types of memory in AI agents? and What is an MCP memory server?

Where Bombos fits

This is the gap Bombos is built for. It is a plugin that gives Claude, Claude Code, ChatGPT, Copilot, Grok, Codex, Cursor and your own agents one shared memory, with team knowledge, rules the team has learned, and what already sits in mail, documents and repositories. Every answer shows what it is based on, and it runs in our cloud or on your own server. One person can try it first for their own agents and bring the team in when it works.

Related questions

Can a ChatGPT Business admin see or share members' memories?

Manage: yes, OpenAI says workspace owners and admins can manage supported memory settings and role permissions. Share: no, a shared project deliberately cannot access a member's personal memories. OpenAI does not say that admins can view an individual member's memories.

Can a Claude Team owner see what Claude remembers about each member?

Not in the interface: owners can turn memory on or off for the whole organization, but they cannot view or edit individual memories. Anthropic does note that memory entries are included in standard conversation history exports, even when memory is turned off.

Is a shared Google Drive or wiki enough?

It is a good source, but only if every tool reads it and someone keeps it current. A wiki does not capture the corrections people give their AI every day.

What about the MCP memory server?

The reference server stores a knowledge graph in a local file. It works across MCP clients on one machine, but it has no users, permissions or sync, so it is not a team memory by itself.

Sources

  1. Projects in ChatGPT (OpenAI Help Center)
  2. Manage project visibility and sharing (Claude Help Center)
  3. Use Claude's chat search and memory (Claude Help Center)
  4. How Claude remembers your project (Claude Code docs)
  5. Rules (Cursor docs)
  6. Knowledge Graph Memory Server (Model Context Protocol)
The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.