Claude Projects are self-contained workspaces in Claude with their own chat history, knowledge files and instructions, so every chat inside starts with the context you set once. This guide shows how to create a project, what to put in its knowledge and instructions, how memory and sharing work, and five project setups you can copy.
- A project bundles knowledge files, custom instructions and memory; every chat in it starts with that context.
- Projects are available on all plans, including Free (up to five projects); paid plans switch to retrieval (RAG) when knowledge nears the context limit.
- Put stable reference material in knowledge and behavior rules in instructions; chats in the same project do not see each other unless you add content to knowledge.
- Team and Enterprise plans can share projects with Can view or Can edit access.
- A newer, beta version of Projects in Claude Code coordinates parallel cloud sessions on your repositories.
What are Claude Projects?
Claude Projects are workspaces in Claude that have their own chat history, knowledge base and instructions. Anthropic describes them as self-contained workspaces. You upload reference material and write instructions once, and every new chat in the project starts with them. That saves you from pasting the same documents and explaining the same context in every conversation.
A project has four parts:
| Part | What it holds | Example |
|---|---|---|
| Knowledge | Files, text and code Claude uses as context | Style guide, product specs, past reports |
| Instructions | Rules for how Claude behaves in this project | "Answer as a senior tax accountant; cite the section of the policy you used" |
| Chats | Conversations that each start with knowledge and instructions | "Draft this month's investor update" |
| Memory | Notes Claude keeps about the project (when memory is on) | Decisions and preferences from earlier chats |
Projects are available to all users, including the Free plan, which allows up to five projects. Paid plans add retrieval mode for large knowledge bases, and Team and Enterprise add sharing. Details below are current as of October 2026; check the linked help pages for changes.
How do you create a Claude Project?
Creating a project takes a minute. The steps from Anthropic's create and manage projects article:
- Open Projects in the left sidebar, or go to claude.ai/projects.
- Click + New Project in the upper right.
- Enter a name and description. These help you find the project; Claude does not see them.
- On Team or Enterprise plans, choose whether the project is private or shared with your organization.
- In the project, click + in the knowledge area on the right and add files, text or code.
- Click Set project instructions, write how Claude should behave, and save.
- Start a chat from the project page.
You can star a project to pin it in the sidebar and archive it when you are done. Archived projects keep their chats, knowledge and members, and you can unarchive them later. To delete a project, unarchive it first.
What should you put in project knowledge?
Put stable reference material in project knowledge: the documents you would otherwise paste into every chat. Claude uses that content as context in every conversation in the project.
Good project knowledge:
- Style and voice guides, with two or three examples of finished work.
- Reference documents: product specs, policies, price lists, research papers, contracts you review against.
- Templates for recurring outputs: report formats, email structures, meeting notes.
- Past examples of good output, labeled as such ("Example of a strong weekly update").
- Glossaries of internal terms and acronyms.
Poor project knowledge:
- Drafts and outdated versions that contradict current documents.
- Large dumps of loosely related files added for completeness.
- Data you are not allowed to share with an AI tool under your company's policy.
Name files descriptively (pricing-2026-q4.pdf, not final_v3.pdf) and mention them by name in your instructions. Claude finds and cites the right file more reliably when it knows what each one is for.
How big can project knowledge be?
Project knowledge is bounded by Claude's context window. On paid plans, when knowledge approaches that limit, Claude automatically switches to RAG mode: instead of loading everything into context, it retrieves the relevant parts for each question. Anthropic says this expands capacity by up to 10 times while keeping response quality. If you want to understand how retrieval works and where it can miss things, see retrieval augmented generation explained.
Adding code from GitHub
You can add files from a GitHub repository to project knowledge. Click + in the knowledge section, choose GitHub, search your repositories or paste a URL, and pick specific files and folders. The GitHub integration syncs file names and contents from one branch (not commit history or pull requests), and the Sync icon refreshes to the latest version. Private repositories need the Claude GitHub App to have access.
Select only the folders relevant to your questions. The whole selection has to fit within the context window.
How do you write good project instructions?
Project instructions tell Claude how to behave in every chat in the project. Keep them short and specific: who Claude should act as, who the audience is, what format to use, and what to do when information is missing.
A template you can adapt:
Role: You are an editor for our B2B newsletter, written for operations managers.Use the knowledge files:- style-guide.md for tone and banned phrases- newsletter-examples.md for structure (match it)- product-facts-2026.md for any product claim. Never state a product fact that is not in this file.Default output: a draft of 400 to 600 words, then a list of claims I shoulddouble-check.If my request lacks a key detail (audience segment, main announcement), ask onequestion before drafting.Plain US English, short paragraphs, no exclamation marks.What makes this work:
- It names the files and says what each one is for.
- It sets a rule for facts ("never state a product fact that is not in this file"), which reduces made-up details.
- It defines the default output, so short prompts like "draft this week's issue" produce the right shape.
- It says what to do when information is missing, instead of letting Claude guess.
Instructions and knowledge together are a form of context engineering: giving the model the right information and rules up front matters more than clever wording in each prompt.
How do memory and chat history work in Claude Projects?
Chats in a project do not see each other. Each chat starts with the project's knowledge and instructions, but what you discussed in one chat is not available in another unless you add it to the project knowledge.
Memory changes this partly. According to Anthropic's memory help article, memory is on by default for Free, Pro and Max users and available on Team and Enterprise when an owner enables it. Each project has its own memory, separate from other projects. In a project, Claude uses that project's memory plus memory from chats outside projects, but not memory from other projects. To keep a project fully separate from the rest of your account, turn off Use account memory in the project's memory settings.
Practical habits:
- When a chat produces a decision or summary you will need again, save it as a knowledge file ("decisions-log.md").
- Start a new chat for each distinct task. Long chats get slower and drift.
- Turn off account memory for client projects so details from one client never surface in another.
How do you share a Claude Project with your team?
On Team and Enterprise plans you can share a project with colleagues, unless an admin has turned sharing off. Click Share project next to the project name, add people by name or email, and choose a role for each:
| Role | Can do |
|---|---|
| Can view | See the knowledge and instructions, and chat in the project |
| Can edit | Change instructions and knowledge, manage members, and contribute |
Shared projects appear under the Organization or Shared with you tabs on the Projects page. A shared project is a good way to give a team one consistent assistant: the same style guide, the same product facts, the same output formats.
Before sharing, remove anything from the knowledge that not every member should see.
Five Claude Project setups you can copy
These are starting points. Adjust the knowledge and instructions to your work.
| Project | Knowledge to add | Key instruction |
|---|---|---|
| Weekly reporting | Report template, three past reports, metrics definitions | "Match the structure of the past reports; flag any metric that moved more than 10 percent" |
| Sales proposals | Product sheet, pricing, case studies, proposal template | "Only quote prices from pricing.pdf; list assumptions at the end" |
| Research reading | The papers or reports for one topic | "Cite the document and page for every claim; say when sources disagree" |
| Codebase Q&A | Selected folders from the GitHub repo, architecture notes | "Answer with file paths; say when the answer is not in the synced files" |
| Personal writing voice | Five pieces you wrote, your banned-phrases list | "Edit for clarity without changing my voice; suggest, don't rewrite" |
The research project pairs well with the techniques in chat with PDF AI, which covers accuracy limits when working from long documents. For more everyday Claude features (artifacts, styles, research), see how to use Claude AI.
What are the new Projects in Claude Code?
Anthropic is rolling out a newer version of Projects, starting with Claude Code. In this version, described in the Claude Code projects docs, a project is one ongoing conversation where Claude coordinates a stream of related tasks and starts a thread for each, usually as a cloud session that runs in parallel and keeps going after you close your laptop. Threads share the project's repositories, instructions (up to 16,000 characters) and memory files.
As of October 2026 this version is in public beta on Pro and Max, rolling out gradually, and not yet available on Team or Enterprise. Repository rules still belong in each repository's CLAUDE.md, which every thread reads; project memory holds notes about the project as a whole.
If you run coding agents on a persistent machine instead, Orkio gives Claude Code an always-on cloud box with your repositories, where you start tasks from your phone and review pull requests later.
How to set up your first Claude Project today
- Pick one recurring task you do at least weekly.
- Create a project for it and gather three to five reference files: a template, good past examples, and the facts Claude must stick to.
- Write instructions using the template above: role, which file is for what, default output, what to do when information is missing.
- Run the task in a new chat and note what you had to correct.
- Fold each correction into the instructions or knowledge, not into individual chats.
- After two weeks, share it with a colleague (on Team or Enterprise) or start a second project for the next recurring task.
Frequently asked questions
Are Claude Projects free?
Yes, projects are available on the Free plan with a limit of five projects. Paid plans (Pro, Max, Team and Enterprise) add retrieval mode for large knowledge bases, which expands how much material a project can hold, and Team and Enterprise add sharing.
Do chats in a Claude Project share context with each other?
Not directly. Each chat starts with the project's knowledge and instructions, but the content of one chat is not visible in another unless you add it to the project knowledge. Project memory can carry some details across chats when memory is turned on.
How much can I upload to a Claude Project?
Project knowledge is limited by Claude's context window, and on paid plans Claude switches to retrieval mode as you approach that limit, which Anthropic says expands capacity by up to 10 times. Very large collections work better split into focused projects.
What is the difference between Claude Projects and ChatGPT Projects?
Both group chats with shared files and instructions. Claude Projects add a GitHub integration for project knowledge, project-scoped memory and team sharing with view or edit roles on Team and Enterprise plans. Check each vendor's current help pages, since both change often.
Can I use Claude Projects with code?
Yes. You can add files from a GitHub repository to project knowledge and sync them later. For agentic coding work, Claude Code reads a CLAUDE.md file in your repository, and a newer beta version of Projects in Claude Code runs tasks as parallel cloud sessions.
