How to Use Claude Projects
Claude Projects keep files and instructions across chats. Free accounts get up to 5 projects; paid plans add RAG that expands knowledge up to 10x.
On this page5 sections
A Project is a self-contained workspace with its own chat history and its own knowledge base, so you stop pasting the same background into every conversation. Projects are available to all users, including free accounts — free is capped at five projects, paid plans go higher.
1. Create one, scoped to a single context
Open Projects → New project and name it. One project per client, per app, per manuscript. A project covering three unrelated things retrieves the wrong document at the wrong moment, which is worse than no project.
2. Upload the knowledge
Add the documents, specs, code or examples Claude should always have: the brand guide, the architecture notes, a writing sample to match. They stay available in every chat inside the project.
On paid plans (Pro, Max, Team, Enterprise), enhanced project knowledge with RAG kicks in automatically as you approach the context limit, expanding capacity by up to 10x. On free accounts the knowledge has to fit the context window, so be selective.
3. Write custom instructions once
Always answer in British English. Be concise — answer first, then a short why.
When writing code, use TypeScript with strict mode.
If the project files don't cover something, say so instead of guessing.
What it does: standing guidance every chat in the project inherits, including the instruction that most reduces invented answers — permission to say “not in the files”.
4. Share it, if your plan allows
Team and Enterprise plans can share a project with named people or the whole organization, with view and edit permission levels, shared tabs and email notifications.
Verify it worked
Start a fresh chat inside the project and ask something only the uploaded documents answer. If you get a plausible generic answer instead, the file isn’t being retrieved — check it’s in this project, and cut it down if it’s enormous.
This is RAG with the machinery hidden: Claude searches your knowledge, pastes the best passages into the prompt, then answers. Which is why the answer changes the moment you update a document, and why quality is set by retrieval rather than by the model — see RAG explained simply.
ChatGPT’s equivalent is ChatGPT Projects; for output you want to keep and iterate on, use Artifacts.
Source: What are Projects?