A Claude Project remembers four things, and two walls decide who can use them. The four are its instructions, its knowledge files, its own chat history and its own memory. Here is our own proposal for reading that, the two-wall test: for anything you want Claude to know, ask which side of the project wall it is on, and which side of the Claude wall. The first wall keeps one project's memory out of another. The second keeps all of it inside Claude.
| What | Who writes it | Which chats can use it | Can another assistant read it |
|---|---|---|---|
| Project instructions | You | Every chat in that project | No |
| Project knowledge | You, by adding files, text or code | Every chat in that project. On paid plans a large knowledge base is searched instead of loaded whole | No |
| Project chat history | You and Claude, as you work | That project's own history | No |
| Project memory | Claude, from the work you do in the project | That project only. Other projects do not see it | No |
| Account memory | Claude, from chats outside projects | Chats outside projects, and inside a project unless you turn off Use account memory | No |
Anthropic's projects page describes a project as a self-contained workspace with its own chat history and knowledge base. Its memory page adds that each project has its own memory. In a project, Claude uses that project's memory and your memory from chats outside projects, and it does not use memory from other projects.
That wall is the point of a project. If you keep one project per client, what Claude learned about one client stays out of the next client's chats. You can raise the wall further: the memory page says turning off Use account memory on the project's page keeps the project separate from your account memory too.
The cost shows up when a fact belongs to more than one project. Your writing style, the name of your company, a decision that affects two pieces of work: each has to be put in both projects, or left in account memory where every project that allows it can see it.
Everything in the table sits inside your Claude account. The help pages we read describe no way for ChatGPT, Cursor or another assistant to read a project's instructions, files, history or memory. Two things do cross the wall, both as copies and not as a live connection. The memory page says all memory data is included in data exports. It also describes moving memory between Claude and other AI services, and calls that feature experimental and still in active development.
So if you also use ChatGPT or a coding agent, a project's memory does not follow you there. The other assistant starts from nothing, or from whatever you paste in.
We run ModelBrain's own work from a Claude Project. On November 14, 2026 it held 388 documents: plans, test results and drafts. Claude can use all of them in that project. None of them is visible to ChatGPT or Cursor, and none is visible to our other Claude projects.
That is why the facts and decisions we want every assistant to have do not live only in the project. We save them to ModelBrain, which sits outside both walls, and each assistant we connect can recall them. The project stays the place for documents, and the memory outside it is the place for what has to travel.
ModelBrain is our product: a memory kept on your own computer that Claude Desktop, Claude Code, Cursor and ChatGPT can all use. It does not replace a project. It holds no project files and sets no instructions. It keeps the facts and decisions you or your assistants save, in partitions that do the same job as the project wall, so one client's memory stays apart from another's while still being reachable from every assistant. Setup for Claude is in the Claude Desktop guide.
We read Anthropic's two help pages on November 14, 2026 and report what they say. They show relative update times, not dates. We did not measure how much a project's knowledge base can hold or test how project memory decides what to keep. Claude changes often, so check the help pages for your plan before relying on a detail.
Does a Claude Project remember previous chats? Yes, in two ways. A project keeps its own chat history, and Anthropic's memory help page says each project has its own memory that Claude builds from the work you do there. Inside a project, Claude uses that project's memory and, unless you turn it off, your account memory from chats outside projects.
Do Claude Projects share memory with each other? No. Anthropic's help page says that in a project Claude does not use memory from other projects. Instructions and knowledge files also belong to one project. To use the same fact in two projects you have to put it in both, or in your account memory, or somewhere outside Claude.
Can ChatGPT or Cursor read a Claude Project? Not directly. A project lives inside your Claude account, and the help pages we read on November 14, 2026 describe no way for another assistant to read it. They do say memory is included in data exports and that an experimental feature moves memory between Claude and other AI services.
Are Claude Projects free? Projects are available on every plan, including free accounts, which can create up to five. The larger knowledge capacity that uses retrieval is for paid plans, and sharing a project with teammates is for Team and Enterprise plans, according to Anthropic's projects help page.
How do I stop a project from using my account memory? Anthropic's memory help page says to turn off Use account memory on the project's page. The project then works only from its own memory, instructions and knowledge.
Last checked November 14, 2026.