You open Claude, explain your role, paste in the same background, the same goals, the same do's and don'ts. The answer comes back good. The next morning you start a fresh chat and type all of it again. A Claude Project ends that re-explaining loop. In about fifteen minutes you can set up a workspace that already knows who you are and what you're working on before you type a word. That small act of setting up standing context is what The 7 Levels of AI Proficiency calls context engineering, and it's the difference between using a tool and building one around your work.
What a Claude Project is
A project is a self-contained workspace with its own conversations, its own knowledge base, and its own memory. Every conversation you start inside it stays grouped with the same body of work.
Three things make a project useful, and Claude can draw on all three for the chats inside it.
The first is project instructions. These tailor Claude's responses. You can tell it to use a more formal tone, or to answer from the perspective of a specific role or industry. Claude uses those instructions for all the chats within the project, so you write them once.
The second is the knowledge base. You upload documents, text, code, or other files, and Claude uses them as context for the conversations in that project. Your style guide, a few past examples, a one-page brief on the work. Whatever a good colleague would want to read before helping you.
The third is project memory. When memory is turned on, Claude keeps a running summary of the work and conversations inside that project. That summary stays separate from your other projects and from your chats outside the project. You can pause or reset Claude's memory whenever you want from Settings.
That's the whole idea. Standing instructions, a small shelf of reference files, and a memory of the work so far, available on every chat without you pasting it again.
When it's worth doing
Not every task needs a project. A one-off question is fine in a regular chat. The setup pays off when you find yourself typing the same context over and over.
Good first candidates:
- A role you write in often. The newsletter you send every week, the client updates, the board summaries.
- An ongoing project with reference docs. A launch, a course, a research thread where the same few files keep coming up.
- A standard you reuse. A brand voice, a format, a checklist you want applied the same way each time.
If you've explained the same thing to Claude more than twice, that explanation belongs in a project.
The method: build it in five steps
Here's the path from empty to working. None of it requires code.
-
Open Projects
In the left sidebar, click "Projects," or go straight to claude.ai/projects.
-
Start a new one
Click "+ New Project" in the upper right. Give it a name and a description. One thing to know here: Claude does not have access to the name or description. They're labels for you, not context Claude reads. The real context comes in the next two steps.
-
Set the instructions
Click "Set project instructions." Write how you want Claude to behave and respond across this workspace, then click "Save instructions." Keep it specific. Who you are, what the project is for, the tone you want, the rules you'd like followed.
-
Add your knowledge
Click the "+" button to add content, and upload your documents, text files, or code. Start with two or three files you'd hand a new teammate.
-
Open a chat and work
Start a conversation inside the project. Claude already carries your instructions and your files, so you can skip the setup speech and get to the real request.
The first three steps take the longest the first time. After that, you're mostly adding a file now and then and refining the instructions as you learn what works.
Do it now: draft your instructions with Claude
You don't have to write the instructions from a blank page. Claude can interview you and draft them. Paste this into any chat, answer the questions, and you'll get a clean block to drop into the project-instructions box.
I want to set up instructions for a Claude Project. Before you write anything, interview me. Ask one or two questions at a time: - who I am and the role I want you to take - what this project is for and the kind of work we'll do here - the tone and format I want in your responses - any rules you'd like followed as closely as possible on every chat After I answer, draft a clear set of project instructions I can paste in. Write them as plain sections: Who you are, What this project is for, Tone and format, Rules to follow as closely as possible. Keep it tight and specific to my answers. Phrase the rules as strong guidance, not as guaranteed behavior.
Expect a short back-and-forth, then a draft. Read it, cut what doesn't fit, and paste the rest into "Set project instructions." You can always come back and edit. Instructions guide Claude's responses across the project; treat them as a strong steer rather than a hard rule engine, and you'll set the right expectation.
A starter file to download
If you'd rather fill in a template than start from scratch, grab the companion. It's a project-instructions skeleton with the parts that count already labeled: Who you are, What this project is for, Tone and format, Rules that always apply, plus a short worked example. Rename it, fill in your answers, and paste it into the instructions box.
Project instructions starter (.md)
A labeled skeleton with a worked example. No signup.
Common mistakes to skip
A few easy things to step around on your first build.
Hiding context in the name or description. It's a natural assumption that a good project name does some work. It doesn't. Claude reads the instructions and the files, not the label. Put anything you want Claude to know in the instructions box.
Treating instructions like a contract. Instructions shape how Claude responds, and they're strong, but they aren't a switch that forces exact behavior every time. Write them as clear guidance, check the output, and tighten the wording where you see drift.
Assuming you need to pay to get value. Projects are free, and free accounts can create up to five, which is plenty to start. Memory runs on every plan too. When a project's knowledge grows past the context limit, Claude automatically switches to retrieval (RAG) that expands its capacity up to ten times. One honest note here: Anthropic's own pages disagree on whether free plans get that automatic retrieval. Its dedicated RAG page says all plans, free included; its older projects overview still says paid-only. The dedicated page is the more specific source, so check your own account before you count on it. The feature clearly reserved for paid plans is searching and citing your past chats across conversations. Either way, you can build a full project, memory and all, on a free account.
Where this fits in the bigger picture
You came here to stop re-explaining yourself. What you built is bigger than that. You designed the standing context a workspace inherits before any chat begins, which is the same skill as writing instructions for any AI system you want to reuse. The job moved from typing a good prompt once to shaping an environment that produces good answers again and again.
That change, from using a tool to building a system around it, is one of the steps The 7 Levels of AI Proficiency maps out. Early on, the skill is asking AI for things well. Later, the skill is engineering the context so the tool already knows your world. A Claude Project is a clean, low-stakes place to practice that second skill. One workspace, a handful of files, instructions you refine over a few weeks.
If you want to see where you are on that climb and what to practice next, the free 7 Levels of AI Proficiency assessment takes about ten minutes and points you to your next step. Take the assessment.
Related reading: Level 4: The Commander.
Sources
- Claude Help Center, "What are projects?", support.claude.com (Accessed June 19, 2026).
- Claude Help Center, "How can I create and manage projects?", support.claude.com (Accessed June 19, 2026).
- Claude Help Center, "Use Claude's chat search and memory to build on previous context", support.claude.com (Accessed June 19, 2026).
- Claude Help Center, "Retrieval augmented generation (RAG) for projects", support.claude.com (Accessed June 19, 2026).
- The 7 Levels of AI Proficiency assessment, launchready.ai/assessment.
Frequently Asked Questions
What is a Claude Project?
It's a self-contained workspace in Claude with its own conversations, knowledge base, and memory. You set project instructions once and upload reference files, and Claude uses both, plus a memory of the work when memory is on, as context for every chat inside that project, so you stop repeating your background each time.
Do Claude Projects cost money?
Projects are available to all users, including free accounts, which can create up to five. Memory works on any plan. The feature clearly reserved for paid plans is searching and citing your past chats across conversations (Pro, Max, Team, or Enterprise). When a project's files grow large, retrieval (RAG) expands its capacity up to ten times; Anthropic's own pages disagree on whether free accounts get that automatically, so check yours.
Does Claude read the project name and description?
No. The name and description are labels for you. Claude does not have access to them. The context Claude uses comes from your project instructions, the files you upload to the knowledge base, and, when memory is on, the project's memory of earlier conversations.
How is a project different from a single chat?
A project gives you a focused workspace for one body of work. It carries project-specific instructions, your uploaded files, related conversations, and its own separate memory. A standalone chat can tap Claude's general memory too, when that's turned on, but it doesn't inherit a project's dedicated instructions and knowledge base.
How long should project instructions be?
Long enough to cover who you are, what the project is for, the tone you want, and the rules you'd like followed. Specific beats long. A focused half-page usually does more than a vague three pages.
Do I need to know how to code to build one?
No. Building a project is clicking through the setup, writing instructions in plain language, and uploading files. You can upload code if your work involves it, but the workspace itself takes no coding to create.
Find your AI Proficiency level
The free 7 Levels assessment places you across seven stages of AI capability. Under ten minutes. Research-backed scoring.