What you will be able to do
- Explain what a Claude Project holds that a standalone chat does not
- Set up a project in the right order, knowing which details Claude can and cannot see
- Decide whether a requirement belongs in project knowledge, project instructions, account-wide instructions or a skill
- Describe how a project behaves when its knowledge base grows past the context window
Key concept
Claude Project — A workspace for one piece of recurring work. It keeps its own chats and its own knowledge base, and its instructions apply to every chat inside it, so you set the context up once instead of pasting it into each new chat.
1.What a project adds to a plain chat
A standalone chat starts empty. Any background Claude needs, such as the brand guide, the client brief or last quarter's figures, has to be supplied again each time. A Claude Project fixes this by giving a piece of recurring work its own workspace, with its own chat history and its own knowledge base. Inside it you can upload documents, provide context and hold focused chats with Claude.
You configure a project in two places, and this page covers both. Project knowledge is what Claude reads: the documents, text, code and other files you upload. Project instructions are how Claude should behave: the tone to use, the role or perspective to take, the rules to follow. A simple way to tell them apart: knowledge answers "what does Claude need to know?" and instructions answer "how should Claude respond?" Most set-up decisions come down to putting a requirement in the right one of the two.
Sources1
2.Setting up a project, step by step
The documented set-up runs in this order:
1. Open Projects from the left side of your account, or go to claude.ai/projects. 2. Click + New Project in the upper right corner. 3. Give the project a name and a description. 4. On a Team or Enterprise plan, choose the visibility. Keep it private means only you and the members you invite can use it. Share with your broader organization means everyone in your organisation can view and use it, where administrators allow that. 5. Add material to the project knowledge base with the + button: documents, text files or code snippets. 6. Click Set project instructions, write how Claude should behave, and click Save instructions. 7. Start chats inside the project.
Step 3 catches people out. Claude has no access to the project's name or description. They are labels for people. A description such as "Formal tone, answer as a financial controller" changes nothing about how Claude responds. That direction only takes effect in the project instructions, which Claude applies to every chat in the project.
What is the primary purpose of the knowledge base within a Claude Project?
Correct answer: A — It gives Claude uploaded documents, text, and code as shared context for every chat inside the project
- A. Correct. The project knowledge base holds documents, text, code, and other files that Claude uses as background context across all chats within that project.
- B. Incorrect. The knowledge base is not a prompt history log; it is a repository of uploaded reference material scoped to one project.
- C. Incorrect. Files uploaded to a project's knowledge base are not automatically published outside the project; access depends on project sharing settings.
- D. Incorrect. The knowledge base supplies context to Claude within the project; it does not spin up a separate chatbot.
Sources2
3.Project knowledge: the only thing shared across chats
The project knowledge base sits on the right side of the project's main page. Anything you upload there is used in every chat in that project, and Claude uses it as context for each one. That is why you upload a brand guide, past newsletter issues or a client's reference documents once, not per conversation.
The documentation also states the limit plainly: chats in a project do not share context with each other. A decision you reach in one chat, or a file you attach to that chat alone, is not visible to the next chat. It only carries over if you add it to the project knowledge base. A project groups chats around shared knowledge and instructions. It does not turn them into one long conversation.
Neither. The style guide was attached to one chat and the brief was pasted into another, so both stayed in their own chats. To make them available in every chat, upload both to the project knowledge base.
Sources2
4.When the knowledge base outgrows the context window
Project knowledge used to be capped by the context window, and once you reached it you couldn't add more. Now, when a project's knowledge approaches or exceeds that limit, Claude automatically switches to retrieval augmented generation (RAG). This expands the project's capacity by up to 10x. Claude no longer loads all the project content at once. It uses a project knowledge search tool to find and retrieve only the passages relevant to your question.
You don't configure anything for this. RAG turns on automatically when it is needed, and a visual indicator shows that the project is RAG-enabled. If the knowledge later drops back below the threshold, Claude can switch back to in-context processing. According to the documentation, response quality stays consistent with in-context processing, and RAG works with web search, extended thinking and Research. The only visible difference is that you will see Claude using the search tool.
Once Claude is retrieving instead of reading everything, you can help it find the right material. Give files clear names, keep related documents in the same project so Claude can draw connections between them, and name a specific document in your question when you know where the answer is.
Sources3
5.Project instructions compared with other ways to steer Claude
Project instructions are the lasting version of the direction you would otherwise repeat in every prompt. A one-off prompt shapes one reply. Project instructions apply to all chats in the project, for example telling Claude to use a more formal tone or to answer from the perspective of a specific role or industry. According to the personalisation guidance, they suit project-specific context, guidelines for a particular workflow, requirements for a specific set of tasks, and roles Claude should adopt. They are most useful for focused work where Claude has to keep the same context across many conversations.
Two neighbouring features also steer Claude, and scenario questions depend on telling them apart. Account-wide Instructions for Claude, found under Settings, apply to all of your conversations, not just one project. Skills add specific behaviours or capabilities that you turn on when you need them.
| Feature | Where it applies | Use it for |
|---|---|---|
| Instructions for Claude (Settings) | All of your conversations with Claude | Preferred approaches or methods, common terms, typical scenarios, general communication instructions |
| Project instructions | Only chats within that project | Project-specific context, guidelines for a workflow, requirements for a set of tasks, roles or perspectives |
| Skills | Your conversations, activated on demand | Repeatable behaviours, tone and format of responses, specialised capabilities for particular tasks |
For an example of well-written instructions, see the new version of projects, currently in beta and rolling out first in Claude Code. Existing projects in chat and Cowork keep working as they do today. In the beta, project instructions are text sent to each new thread, up to 16,000 characters, and the guidance says to cover what the project is for, where the work happens, how a thread checks its work, what to do when something is missing, and what needs your approval first.
This project holds p95 latency for the payments API under 200 ms: profiling, query and caching fixes, and the dependency upgrades that come with them, in the payments-api repository.
- Branch from main and open one draft pull request per thread.
- Before you call work done, run `make test` and `make lint` and paste the summary lines in your final message.
- If you can't reach something you need, such as a repository, a secret, an API, or a connector, say exactly what's missing in your first message and stop. Don't substitute, mock, or guess.
- Don't merge, force-push, or change CI configuration without asking me in the thread.Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Putting the project's purpose or tone in its name and description is enough for Claude to follow it.Why is that wrong?
Claude cannot see a project's name or description. Direction only takes effect when it is written in the project instructions.
Covered in Setting up a project, step by step
2.Everything discussed or attached in one chat of a project is automatically available in the project's other chats.Why is that wrong?
Chats in a project do not share context. Only material added to the project knowledge base carries across chats.
Covered in Project knowledge: the only thing shared across chats
3.Once a knowledge base gets too big, you have to turn on RAG mode yourself or split the material into another project.Why is that wrong?
RAG turns on automatically as project knowledge nears the context window limit. It needs no setup and can switch back if the knowledge shrinks again.
Covered in When the knowledge base outgrows the context window
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“Projects allow you to create self-contained workspaces with their own chat histories and knowledge bases.”
↩︎ What a project adds to a plain chat“For example, instructing Claude to use a more formal tone or answer questions from the perspective of a specific role or industry.”
↩︎ Project instructions compared with other ways to steer Claude“Projects allow you to create self-contained workspaces with their own chat histories and knowledge bases.”
↩︎ Key concept - 2.
“Click "+ New Project" in the upper right corner.”
↩︎ Setting up a project, step by step“Keep it private: Only you and members you invite can view or use this project.”
↩︎ Setting up a project, step by step“Anything you upload to this space will be used across all of your chats within that project.”
↩︎ Project knowledge: the only thing shared across chats“Note: Context is not shared across chats within a project unless the information is added into the project knowledge base.”
↩︎ Project knowledge: the only thing shared across chats“Give your project a name and description (note that Claude will not have access to these details).”
↩︎ Exam trap 1“Note: Context is not shared across chats within a project unless the information is added into the project knowledge base.”
↩︎ Exam trap 2 - 3.https://support.claude.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projectsOfficial docs
“Claude will automatically enable RAG mode to expand your project's capacity by up to 10x while maintaining quality responses.”
↩︎ When the knowledge base outgrows the context window“Claude intelligently searches and retrieves only the most relevant information needed to answer your questions.”
↩︎ When the knowledge base outgrows the context window“If your project knowledge later drops below the context window threshold, Claude can automatically convert back to context-based processing.”
↩︎ When the knowledge base outgrows the context window“Well-named files help Claude understand and retrieve the right information more effectively.”
↩︎ When the knowledge base outgrows the context window“Yes. RAG works with all Claude features, including web search, extended thinking, and Research.”
↩︎ When the knowledge base outgrows the context window“No, RAG activates automatically when needed. No setup or configuration is required.”
↩︎ Exam trap 3 - 4.https://support.claude.com/en/articles/10185728-understanding-claude-s-personalization-featuresOfficial docs
“These instructions only apply to chats within that project.”
↩︎ Project instructions compared with other ways to steer Claude“define repeatable behaviors you activate on demand”
↩︎ Project instructions compared with other ways to steer Claude - 5.https://code.claude.com/docs/en/claude-projectsOfficial docs
“Text sent to each new thread and to Claude in the project conversation, up to 16,000 characters.”
↩︎ Project instructions compared with other ways to steer Claude