Claude Projects and ChatGPT Projects Don't Scale to a Team
Your team tried to share AI context the obvious way. Someone set up a Claude Project or a ChatGPT Project, dropped in the docs, and pointed people at it. It half-works. Then it hits a wall, and everyone quietly goes back to pasting context into their own chats.
Quick answer
As of August 2026, assistant memory in Claude and ChatGPT is per person: there is no team-wide memory your colleagues share. Projects can be shared on Team and Enterprise plans, but a shared Project is upload-based, scoped to that project, and works with one assistant. The working fix is a shared knowledge base with a built-in MCP server, like Hjarni Teams, where every teammate's own Claude or ChatGPT reads and writes the same notes.
Projects are good at what they are: context for one focused piece of work. As a team's shared brain, even the shareable ones run into three limits.
Three walls
They live inside one assistant. A Claude Project helps the people on Claude. A ChatGPT Project helps the people on ChatGPT. Whichever you pick, the teammates using the other one are left out, and most teams use both.
They're built around uploads, not a living knowledge base. A Project is a set of instructions and files you attach and then read from. It isn't a shared body of notes your team edits as decisions change, with assistants that read and write back to it. It's a reference drop, and it drifts out of date.
They're scoped to one project. Context in one Project doesn't carry to the next. Your team rebuilds the same background every time it starts something new.
None of this is a knock on Projects. They were built around one focused task and a set of files, not a team and a growing body of knowledge.
The missing feature: shared team memory
There's a name for what teams keep asking for here: shared team memory. Sharing a Project shares the files and instructions you put in it. The assistant's memory is a different thing, and it stays personal: Claude remembers things about you, but that memory is yours alone. It doesn't reach your teammates, and their assistants don't learn from what yours knows. Users have been asking Anthropic for a team-wide version; as of August 2026 neither Claude nor ChatGPT has one.
You don't have to wait for the feature. The shape of the fix already exists, and it's arguably better than an in-app version would be, because it isn't locked to either assistant.
One place, outside any single assistant, so it serves the Claude users and the ChatGPT users alike. Writable, so decisions and runbooks stay current instead of frozen at upload time. And readable by every teammate's assistant over MCP.
That's a shared knowledge base with a built-in MCP server, not a folder of files inside one person's Project. Every teammate connects their own AI, and it reads and writes the same notes the rest of the team does. Write a decision down once and every assistant on the team can reach it, in the next session and the one after that.
The full version of this argument is in the team knowledge base for Claude and ChatGPT. The short version: a Project is a person's working context. A team needs a shared, living one. The same holds when you hand a project off for good: a consultant handing an AI project to a client needs the client to inherit a source of truth they can edit, with a trail of what changed, not a Project locked in one account.
When Projects are still the right tool
Keep using a Project for what it's good at. A single person, a single deep task, a set of reference files you want that one assistant to focus on. That's the job it does well.
The moment the context needs to be shared, stay current, and serve more than one person's AI, it has outgrown a Project.
Moving your team off Project silos
- Create a team and a few shared folders for decisions, runbooks, and conventions. The set up your team guide is the quickstart.
- Move the reference material out of the scattered Projects and into the shared base, once, where the whole team maintains it together.
- Each teammate connects their own Claude or ChatGPT. A team is free for its first 25 team notes, so you can move everything over before you spend anything.
The point
Projects gave your team a taste of what a working AI context feels like. The problem was never the idea. It was that the context stayed boxed inside one project, in one assistant, and couldn't grow.
Give your team a shared one every assistant can read. See Hjarni for teams, or Hjarni for engineering managers for how a team puts it to work.
One knowledge base your whole team's AI reads. Free.
Start a shared brain your teammates' Claude and ChatGPT read and write. Free for your first 25 team notes.
One knowledge base your whole team's AI reads. Free.
Common questions
FAQ
Can you share a Claude Project with your team?
You can share a Project within a Claude Team or Enterprise workspace, but it stays inside Claude and is built around uploaded files and instructions for a focused task. It is not a shared, writable knowledge base that both Claude and ChatGPT read, which is what a team's shared context needs.
Do ChatGPT Projects work for teams?
A ChatGPT Project helps the people on ChatGPT, not the teammates using Claude, and it is scoped to that one project. For shared team context you want something outside any single assistant that both clients can read.
What is the alternative to Claude Projects for a team?
A shared knowledge base with a built-in MCP server. It lives outside any one assistant, so every teammate's Claude or ChatGPT reads the same notes, and the team edits it as decisions change. Hjarni is built for this.
When should we still use Claude or ChatGPT Projects?
For one person and one focused task with a set of reference files. That is what Projects do well. The moment the context needs to be shared, stay current, and serve both Claude and ChatGPT users, it has outgrown a Project.
Does Claude have shared team memory?
Not as of August 2026. Claude's memory is per person, and while Projects can be shared on Team and Enterprise plans, they are upload-based, project-scoped, and Claude-only. The working answer today is a shared knowledge base over MCP: every teammate's Claude reads and writes the same notes, and so does every teammate's ChatGPT.
How do we give our whole team's AI one shared memory?
Put the team's context in a shared knowledge base with a built-in MCP server, like Hjarni Teams. Each teammate connects the assistant they already use. Decisions, runbooks, and conventions live in shared folders that every assistant reads, and updates land for everyone at once.