How to Use Claude Projects for Long-Running Work
A practical guide to Claude Projects: how to set up a knowledge base and custom instructions so you stop re-explaining context in every chat.
If you use Claude for anything ongoing — a codebase, a book, a research topic, a specific client — you've probably noticed the tax: every new chat starts from zero. You paste the same background, re-explain the same rules, and only then get to the actual question. Claude Projects exists to kill that tax. This guide shows you how to set one up so Claude already knows your context before you type a word.
What a Claude Project Actually Is
A Project is a container that bundles three things in one place: a persistent knowledge base you upload once, a set of custom instructions Claude follows in every chat, and all the conversations you have inside it. Each Project gets a 200K-token context window — about a 500-page book — so Claude can ground its answers in your material instead of guessing. Projects are available on the Claude Pro and Team plans.
When to Use a Project vs a Regular Chat
Not everything needs a Project. The deciding question is simple: are you repeating context?
| Situation | Use a Project? |
|---|---|
| One-off question, no shared background | No — a regular chat is faster |
| Same codebase or docs referenced across many chats | Yes |
| A client or brand with consistent tone and rules | Yes |
| A book, thesis, or research area you work on for weeks | Yes |
| Quick brainstorm you'll never revisit | No |
Knowledge vs Custom Instructions: What Goes Where
The most common mistake is dumping everything into one box. Projects split context into two distinct slots, and using them correctly is what makes Claude feel like it «gets» your work.
| Slot | What belongs here | Example |
|---|---|---|
| Knowledge (uploaded files) | Reference material Claude should read from — facts, code, documents | Your API docs, a style guide PDF, the current draft of your book |
| Custom Instructions (text) | How Claude should behave — role, tone, format, rules | "You are my senior editor. Be blunt. Always suggest a shorter version." |
Setting Up Your First Project — Step by Step
- In Claude, create a new Project and give it a clear name (e.g. «Acme Client» or «My Novel»).
- Add custom instructions: define Claude's role, tone, and any hard rules it must follow every time.
- Upload your knowledge: the documents, code, or notes Claude should reference. Prefer clean, current files over outdated ones.
- Start a chat inside the Project and test it — ask something that requires the uploaded context to answer correctly.
- Refine: if answers miss the mark, tighten the instructions or replace stale knowledge files, then re-test.
The Efficiency Payoff
The value is concrete, not abstract. Say you normally paste ~1,500 words of background at the start of each session, and you start five Claude sessions a day on the same project. That's roughly 7,500 words of re-typing or re-pasting daily — and every one of those tokens eats into the context window and your attention. With a Project, you set that context once and reclaim it on every future chat.
| Without a Project | With a Project | |
|---|---|---|
| Context setup per chat | Re-paste every time | Once, reused automatically |
| 5 chats/day background | ~7,500 words re-pasted | 0 re-pasted |
| Consistency of rules | Varies per chat | Identical every chat |
Common Mistakes to Avoid
- Uploading stale files — Claude will faithfully use outdated context. Keep the knowledge base current.
- Writing vague instructions like «be helpful.» Be specific about role, tone, and format.
- Overloading one Project with unrelated topics — split distinct workstreams into separate Projects.
- Forgetting the 200K window is shared: a huge knowledge base leaves less room for long conversations.
Bottom Line
Claude Projects turns Claude from a stateless chatbot into something that remembers your work. If you catch yourself pasting the same context twice, that's your signal to make a Project. Set the knowledge and instructions once, and every future chat starts where you left off. New to Claude overall? Start with our full Claude overview to see models, pricing, and what it's best at before diving into Projects.
Frequently Asked Questions
- Do I need a paid plan to use Claude Projects?
- Yes. Projects are available on Claude Pro and Team plans, not the free tier.
- How much can I upload to a Project's knowledge base?
- Each Project has a 200K-token context window — roughly a 500-page book — shared across your uploaded knowledge and the conversation.
- Is my Project data used to train Claude?
- No. Anthropic states that data shared in Projects is not used to train its models without your explicit consent.
- When should I use a Project instead of a normal chat?
- Use a Project whenever you keep re-pasting the same background — a codebase, a style guide, a client's details — across multiple chats. For a one-off question, a normal chat is fine.