---
id: projects
lang: en
title: "Projects"
seoTitle: "Claude Projects: keep work together"
description: "Claude Projects keep files, instructions, and chats together for ongoing work, so Claude has the full picture every time. What they are and how to set one up."
part: working-smart
partTitle: "Working smart"
partOrder: 4
order: 3
level: Intermediate
durationMinutes: 5
outcome: "Set up a Project so an ongoing piece of work keeps its files, instructions, and history in one place."
summary: "A Project keeps the files, instructions, and chats for one piece of work together, so Claude has the full picture every time instead of starting cold."
draft: false
updated: "2026-07-26"
furtherReading:
  - title: "What are projects?"
    url: "https://support.claude.com/en/articles/9517075-what-are-projects"
    note: "Anthropic's overview: knowledge, instructions, and sharing."
  - title: "How can I create and manage projects?"
    url: "https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects"
    note: "The click-by-click steps to set one up."
  - title: "Retrieval augmented generation (RAG) for projects"
    url: "https://support.claude.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projects"
    note: "How a paid project scales to hold far more knowledge."
  - title: "Project visibility and sharing"
    url: "https://support.claude.com/en/articles/9519189-project-visibility-and-sharing"
    note: "Sharing and permissions on Team and Enterprise plans."
keywords:
  - Claude Projects
  - how to use Claude Projects
  - organise AI work
  - Claude project knowledge
  - reuse context Claude
quiz:
  intro: "Five quick questions. Ace them all and you've got Projects down."
  passScore: 5
  questions:
    - q: "What does a Project keep together in one place?"
      options:
        - "Only your billing details"
        - "The knowledge (files you upload), the instructions (a standing brief), and all the chats for that piece of work"
        - "A list of other people's projects"
        - "Nothing; it's just a folder name"
      answer: 1
      explain: "A Project is a workspace with three parts: its knowledge base, its instructions, and its own chat history. Every chat inside it starts already knowing the background."
    - q: "How does a Project compare to the three places to put context from lesson 2.5?"
      options:
        - "It's the same as a one-off message"
        - "It's the middle scope: broader than a single chat, narrower than your whole profile"
        - "It replaces your account settings"
        - "It only works for coding"
      answer: 1
      explain: "This message, this project, everything: the Project is the middle one. Perfect when a body of work has context that shouldn't leak into every other chat you have."
    - q: "Why can a paid Project hold far more reference material than a single chat could?"
      options:
        - "It has no limits at all"
        - "When the knowledge gets big, Claude switches to retrieval (RAG): it searches your files and pulls in only the relevant parts for each question, instead of loading everything"
        - "It deletes older files automatically"
        - "It makes the model bigger"
      answer: 1
      explain: "This ties straight back to lesson 4.1. Rather than stuffing every file onto the desk, a paid project retrieves only the slices each question needs, so it can hold roughly ten times more without clogging the window."
    - q: "You return to the same kind of work every week: same client, same background, same tone. What's the smart setup?"
      options:
        - "Re-paste the background into a fresh chat each time"
        - "Make a Project: add the files and a short instructions brief once, then just chat"
        - "Keep it all in your head"
        - "Use a different tool each week"
      answer: 1
      explain: "That is exactly what Projects are for. Set up the knowledge and instructions once, and every future chat in that project starts briefed. No more re-pasting."
    - q: "How do you keep a Project useful over time?"
      options:
        - "Never touch it again after setup"
        - "Treat the knowledge like a fridge, not an attic: remove stale files, keep instructions short and current, and use one project per real piece of work"
        - "Add every file you own to a single giant project"
        - "Delete it and start over each month"
      answer: 1
      explain: "Stale files and bloated instructions mislead Claude and waste space. Keep the knowledge fresh and the brief tight, and give each real piece of work its own project."
  pass:
    badge: "Projects unlocked"
    title: "Your work has a home now."
    body: "Full marks. You can set up a Project, feed it knowledge and a brief once, and let every chat start with the full picture. Next: Cowork, where Claude works directly on the files on your own computer."
  retry: "Check the marked answers and give it another go. You've got this."
---

> **Level:** Intermediate · **Reading time:** ~5 min
>
> **What you'll be able to do:** set up a Project so an ongoing piece of work keeps its files, instructions, and history in one place. *(Last checked: July 2026.)*

A one-off chat forgets everything the moment you close it. That is fine for a quick question. It is a pain for work you come back to: the same client, the same report, the same pitch, week after week. You end up **re-pasting the same background every single time.**

A **Project** ends that. Think of it as a home for one piece of work, where the context is always already in the room. Open a new chat inside it and Claude starts out knowing the brief, the files, and the history, before you type a word.

> **A quick heads-up on the word "project."** This lesson is about Projects in Claude's **Chat**. Claude also has **Cowork**, now built right into the same Claude app: you pick Chat or Cowork from the same message box. Cowork has its *own* Projects, which live on your computer and organise the work Claude does on your files. Those are the next lesson (4.4). Same word, two homes, so it pays to know which one you are in.

## What a Project holds

A Project keeps three things together:

- **Knowledge.** Files you upload once: the brief, the brand guide, past examples, the data. Claude reads from these in *every* chat in the project.
- **Instructions.** A standing brief for this work: who it's for, the goal, the tone, what to avoid. Set once, applied to every chat inside.
- **Chats.** Its own history. All the conversations about this work live here, not scattered across your sidebar.

<figure class="lesson-svg" aria-labelledby="proj-title">
<svg viewBox="0 0 640 250" width="100%" height="auto" role="img" xmlns="http://www.w3.org/2000/svg" font-family="Inter, system-ui, sans-serif">
<title id="proj-title">What a Project keeps together</title>
<desc>A Project box holds three parts: knowledge (uploaded files), instructions (a standing brief), and the project's own chats. All three feed into every new chat, which starts already briefed.</desc>
<text x="0" y="24" font-size="20" font-weight="800" fill="var(--ink,#1F2328)">One home for one piece of work</text>
<rect x="0" y="40" width="380" height="190" rx="14" fill="var(--part-tint,#EDF2FC)" stroke="var(--part,#3E74E0)"/>
<text x="20" y="66" font-size="14" font-weight="800" fill="var(--part-ink,#2C55B8)">Your Project</text>
<rect x="20" y="78" width="340" height="40" rx="9" fill="#fff" stroke="var(--part,#3E74E0)"/>
<text x="34" y="94" font-size="13" font-weight="700" fill="var(--ink,#1F2328)">Knowledge</text>
<text x="34" y="110" font-size="11.5" fill="var(--med-gray,#5B6169)">Files you upload once: brief, brand guide, data.</text>
<rect x="20" y="124" width="340" height="40" rx="9" fill="#fff" stroke="var(--part,#3E74E0)"/>
<text x="34" y="140" font-size="13" font-weight="700" fill="var(--ink,#1F2328)">Instructions</text>
<text x="34" y="156" font-size="11.5" fill="var(--med-gray,#5B6169)">A standing brief: who, goal, tone, watch-outs.</text>
<rect x="20" y="170" width="340" height="40" rx="9" fill="#fff" stroke="var(--part,#3E74E0)"/>
<text x="34" y="186" font-size="13" font-weight="700" fill="var(--ink,#1F2328)">Chats</text>
<text x="34" y="202" font-size="11.5" fill="var(--med-gray,#5B6169)">Every conversation about this work, in one place.</text>
<path d="M388 135 L452 135" stroke="var(--part,#3E74E0)" stroke-width="2.5" fill="none" marker-end="url(#ar)"/>
<defs><marker id="ar" markerWidth="9" markerHeight="9" refX="7" refY="4.5" orient="auto"><path d="M0 0 L9 4.5 L0 9 z" fill="var(--part,#3E74E0)"/></marker></defs>
<rect x="460" y="96" width="180" height="78" rx="12" fill="var(--part,#3E74E0)"/>
<text x="550" y="128" font-size="14" font-weight="800" fill="#fff" text-anchor="middle">Every new chat</text>
<text x="550" y="150" font-size="13" fill="#fff" text-anchor="middle">starts briefed.</text>
</svg>
</figure>

## Why it beats a one-off chat

Back in lesson 2.5 you met **three places to put context**: this message, this project, everything (your profile). The Project is the **middle scope**, and it is the one people underuse. Broader than a single chat, so you do not repeat yourself. Narrower than your profile, so this work's context does not bleed into every unrelated chat you have. When a body of work has its own world, the Project is where that world lives.

Good news on access: **Projects are free for everyone.** Free accounts can keep up to five; paid plans lift that and add the scaling trick below.

## Setting one up

It takes a couple of minutes:

1. **Make a project** and give it a clear name ("Acme account", "Q3 board report").
2. **Add the knowledge.** Drop in the documents this work leans on. Same uploads as any chat, up to 30 MB a file, and the lean-format tip from 2.5 still pays: prefer `.md` or `.txt` for big reference material.
3. **Write the instructions.** The same five-line brief from 2.5 (who, goal, audience, style, watch-outs), scoped to this work.
4. **Start chatting.** That's it. Every chat now begins with the full picture.

## It scales past a single window

Here is where a Project quietly outgrows a normal chat, and it connects straight to lesson 4.1. In an ordinary chat, everything you add competes for the same desk space. Pile on enough files and the window fills.

On a **paid plan**, a Project handles this for you. When the knowledge base gets large, Claude switches to **retrieval** (the technique is called RAG): instead of loading every file onto the desk at once, it *searches* your knowledge and pulls in only the parts that matter for the question in front of it. That lets a project hold far more reference material, up to roughly ten times more, without the window ever clogging.

<figure class="lesson-svg" aria-labelledby="rag-title">
<svg viewBox="0 0 640 210" width="100%" height="auto" role="img" xmlns="http://www.w3.org/2000/svg" font-family="Inter, system-ui, sans-serif">
<title id="rag-title">Retrieval: pull only what the question needs</title>
<desc>A large knowledge base sits on the left. For a given question, Claude retrieves only the two or three relevant files and loads just those into the context window, leaving the rest on the shelf.</desc>
<text x="0" y="24" font-size="20" font-weight="800" fill="var(--ink,#1F2328)">Big knowledge, small footprint</text>
<text x="0" y="52" font-size="13" font-weight="700" fill="var(--part-ink,#2C55B8)">Project knowledge</text>
<g>
<rect x="0" y="62" width="150" height="22" rx="5" fill="#fff" stroke="var(--part,#3E74E0)"/>
<rect x="0" y="90" width="150" height="22" rx="5" fill="var(--part,#3E74E0)"/>
<rect x="0" y="118" width="150" height="22" rx="5" fill="#fff" stroke="var(--part,#3E74E0)"/>
<rect x="0" y="146" width="150" height="22" rx="5" fill="var(--part,#3E74E0)"/>
<rect x="0" y="174" width="150" height="22" rx="5" fill="#fff" stroke="var(--part,#3E74E0)"/>
</g>
<text x="180" y="120" font-size="13" fill="var(--med-gray,#5B6169)">Question comes in</text>
<path d="M300 115 L372 115" stroke="var(--part,#3E74E0)" stroke-width="2.5" fill="none" marker-end="url(#ar2)"/>
<defs><marker id="ar2" markerWidth="9" markerHeight="9" refX="7" refY="4.5" orient="auto"><path d="M0 0 L9 4.5 L0 9 z" fill="var(--part,#3E74E0)"/></marker></defs>
<rect x="384" y="70" width="256" height="96" rx="12" fill="var(--part-tint,#EDF2FC)" stroke="var(--part,#3E74E0)"/>
<text x="512" y="96" font-size="13" font-weight="800" fill="var(--part-ink,#2C55B8)" text-anchor="middle">Only the relevant</text>
<text x="512" y="114" font-size="13" font-weight="800" fill="var(--part-ink,#2C55B8)" text-anchor="middle">files loaded</text>
<rect x="470" y="128" width="84" height="20" rx="5" fill="var(--part,#3E74E0)"/>
<rect x="470" y="152" width="84" height="10" rx="3" fill="var(--part,#3E74E0)"/>
<text x="0" y="208" font-size="12" font-style="italic" fill="var(--med-gray,#5B6169)">Paid plans only. On Free, keep a project's knowledge modest so it fits the window.</text>
</svg>
</figure>

On a **free plan** there is no retrieval, so keep a project's knowledge modest and it works fine. Either way, the habit is the same: give the project what the work needs, and keep it current.

## Keep it tidy

A project's knowledge is a **fridge, not an attic.** Things go off. Remove files that are out of date, or Claude will happily quote last quarter's numbers. Keep the instructions **short and current**, a bloated brief works worse than a tight one. And give **each real piece of work its own project** rather than dumping everything into one; a project about "everything" is a project about nothing.

## Sharing with a team

On **Team and Enterprise** plans, a project can be **shared** with colleagues, as view-only or full edit. Now the whole team works from the same knowledge and the same brief, which is the closest thing to a shared playbook Claude offers. On Free, Pro, and Max, a project stays yours alone.

## Where this goes next

Put lessons 4.2 and 4.3 together and you have the power-user setup: **Skills** are *how* a job gets done your way; **Projects** are *where* a body of work lives, with its context on tap. Next, in lesson 4.4, we leave the chat window entirely: **Cowork** lets Claude work directly on the files sitting on your own computer.

## Try it

Pick something you come back to often: a client, a recurring report, an ongoing pitch. Make a Project for it. Add two or three files it always needs, and write a five-line instructions brief.

Now open a fresh chat inside the project and ask for something real, without pasting any background. Watch it answer as if it already knew the whole story. **It does, because you told it once instead of every time.** That is the whole point of a Project.

*(General exercise. Role and industry versions come once accounts are in.)*

## Key terms

[Project](/glossary#project), [Context](/glossary#context), [Context window](/glossary#context-window), [Custom instructions](/glossary#custom-instructions)
