Projects: where your Claude work stops evaporating
You have a quarterly marketing campaign that has been eating chat threads for two weeks. Every time you open a new conversation you paste the same brand guidelines, re-explain the target audience, and remind Claude that you work in B2B SaaS — not DTC, not consumer, not retail. Most of your first prompt is context Claude should already have. That is exactly the problem Projects are designed to solve.
A Project is a self-contained Claude workspace with its own chat history, knowledge base, and custom instructions. Think of it as a dedicated environment for a specific stream of work — one where Claude already knows who you are, what you are working on, and how you want it to respond, before you type a single word.
When a project is worth creating
A project is overkill for a one-off question. It is exactly right when you have ongoing work that keeps pulling in the same context. Three signals that you should be working in a project instead of a fresh chat:
- Reference materials you use repeatedly. Meeting notes, survey results, historical reports, brand guidelines, client briefs. Things you would otherwise re-upload every time.
- Consistent rules for how Claude should respond. Always formal. Always cite sources. Always follow our template. Always assume the reader is a procurement lead.
- Team collaboration on shared context. Multiple people working from the same foundation, so nobody has to be the keeper of "the right version".
Projects are a unit of continuity. If you are going to touch the same body of work more than twice, make a project.
The three pieces of a project
A project has three components, and getting each right is what makes Claude feel like a real collaborator instead of a smart chat box.
1. A name, a description, and a visibility setting
The name is what you will see in the sidebar. Make it descriptive — Q4 Marketing Campaign, Client: Osprey Launch, Product Documentation — not Project 1. The description is a short note about what you are working on. Claude does not read it directly, but it helps you and your teammates remember what this workspace is for.
Visibility is either private (just you) or shared with your organisation (for Team and Enterprise users). You will come back to sharing in a moment.
2. Project instructions — the standing brief
Project instructions tell Claude how to behave across every conversation in this project. They are your standing brief. Good instructions usually cover four things:
- Context: "This project is for creating marketing content for our B2B SaaS product."
- Process: "First consider a blog structure that will engage this audience, then write the draft."
- Tone and style: "Use a professional but conversational tone. Avoid jargon where possible."
- Specific requirements: "Always include a call-to-action at the end of marketing copy."
Think of project instructions as programming Claude's behaviour for this stream of work. You write them once, save them, and every new chat inside the project inherits them — on top of your global preferences and any style you have selected.
Pro tip. You can use instructions to automate small recurring workflows. "When I upload a meeting transcript, create a structured summary using this template." That turns the project into a one-line command surface for something you used to do manually every week.
3. A knowledge base — your uploaded reference material
The knowledge base is where you upload documents Claude should reference across all conversations. Open the files menu on the right side of the project page and click the plus button. Claude accepts PDFs, DOCX, CSVs, TXT, HTML, and more, and can connect directly to Google Drive.
What to upload:
- Reference documents — brand guidelines, style guides, templates.
- Background material — research reports, meeting notes, requirements docs.
- Examples of work you want Claude to emulate.
- Technical documentation or specifications.
Pro tip. Name your files descriptively. Claude uses file names to pick the right reference. "Q4-2025-Brand-Guidelines.pdf" is vastly more useful than "doc1.pdf".
How projects handle large knowledge bases
You might wonder what happens when you dump a hundred documents into a project. Claude handles it automatically through Retrieval Augmented Generation, usually called RAG mode. When your project's knowledge base gets large enough that it cannot all fit in the context window at once, Claude switches from loading everything to intelligently retrieving only the most relevant chunks for each question.
You will see a visual indicator when a project is RAG-enabled, but the experience feels the same — you still upload documents, still chat with Claude, still get context-aware responses. The difference is capacity: RAG expands what a project can hold by up to 10x while keeping response quality steady. You never have to think about when RAG kicks in; Claude makes that call.
Sharing and collaboration
For Claude for Work users (Team and Enterprise plans), projects get a second life as a collaboration surface. You can share a project with specific teammates or make it visible to your entire organisation.
There are three permission levels:
- Can view — members can see project contents, access the knowledge base, and chat, but cannot change anything. Read-only with discussion rights.
- Can edit — members have full collaboration power. They can modify instructions, update knowledge, manage members, and contribute to the project.
- Owner — project creators control everything, including who sees the project.
A brand team can create one Brand Voice project with voice and tone guidelines, then share it with everyone in the company. A client account team can keep a single Client: Osprey hub with brand guidelines, past deliverables, and communication history — and every new team member walks into the same context. Projects stop being one person's private workspace and start being the team's institutional memory.
A quick mental model — projects as continuity
Every project has three things Claude carries across every conversation: your instructions (how to act), your knowledge base (what to know), and your chat history (what you have already decided). Together they form continuity — a place where your work accumulates instead of evaporating at the end of each chat.
If you only remember one thing from this lesson, remember this: if you find yourself re-pasting the same context, you should be in a project instead.
Before you move on
Look at your last week of Claude conversations. Find one where you re-explained the same context more than once. That is your first real project.
Key Takeaways
- 1A Project is a self-contained Claude workspace with its own chat history, knowledge base, and custom instructions — use one whenever you have ongoing work that keeps reloading the same context.
- 2Project instructions act as a standing brief: context, process, tone, and specific requirements that apply to every conversation inside the project.
- 3The knowledge base is where reference material lives — descriptive file names matter because Claude uses them to decide which documents to pull into a given answer.
- 4Projects automatically scale through RAG mode when the knowledge base grows beyond the context window, expanding capacity by up to 10x while keeping response quality steady.
- 5On Team and Enterprise plans, projects become shared hubs — view, edit, and owner permissions turn a personal workspace into an institutional memory for the whole team.