Claude Projects: How to Set Up and Use Them (2026)

Claude Projects are self-contained workspaces in claude.ai, each with its own chats, uploaded knowledge and custom instructions. Every chat you start inside a project already knows that material, so you stop re-uploading the same files. Projects work on all plans; Free accounts get up to five, and paid plans add retrieval for large knowledge bases.

This guide covers how to set up a project, what to put in it, how the knowledge limit and RAG mode work, and how sharing works on Team and Enterprise. Facts come from Anthropic’s Help Center, checked October 24, 2026. Claudekit.ai is an independent community site, not affiliated with Anthropic.

What a Claude Project gives you

A project bundles three things:

Part	What it does	Shared across chats?
Project knowledge	Documents, text, code and GitHub files Claude uses as background	Yes, every chat
Project instructions	Standing guidance such as tone, role or output format	Yes, every chat
Chats	Normal conversations, scoped to the project	No, each chat is separate
Project memory	Things Claude remembers from work in the project (when memory is on)	Yes, within the project

The one that trips people up: chats don’t see each other. The Help Center is explicit that context isn’t shared across chats in a project unless you add it to the project knowledge. If a chat produces something you’ll need again, such as a decision, a summary or a final draft, save it into the knowledge base.

Projects vs a normal chat vs Artifacts

	Normal chat	Project	Artifact
Purpose	One-off question or task	Repeated work on the same material	A thing you produce and share
Persistent files	No, attach per chat	Yes, in project knowledge	Lives beside the chat
Custom instructions	Account-level only	Per project	No
Sharing	Share a chat	Share the whole workspace (Team/Enterprise)	Share by link
Best for	Quick help	Ongoing clients, courses, codebases, reports	Docs, decks, dashboards, tools

Projects and Claude Artifacts work well together: build the artifact inside a project so it can draw on the project’s files.

Plan availability and limits

Plan	Projects	RAG for large knowledge	Sharing
Free	Up to 5	No	No
Pro	Yes	Yes	No
Max	Yes	Yes	No
Team	Yes	Yes	Yes
Enterprise	Yes	Yes	Yes, including groups (beta)

File limits from the Help Center’s upload article:

• Project files: up to 30MB per file, with an unlimited number of files as long as the content fits within the context window (RAG raises that ceiling on paid plans).
• Project files use text extraction, except multimodal PDFs.
• Supported document types include PDF, DOCX, CSV, TXT, HTML, ODT, RTF, EPUB, JSON and XLSX (XLSX needs code execution and file creation turned on).

For plan prices and usage limits, see our Claude pricing guide.

How to create a Claude Project

1. Go to claude.ai/projects (or click “Projects” in the left sidebar) and click + New Project.
2. Give it a name and description. Note that Claude can’t see these; they’re just for you.
3. On Team or Enterprise, choose visibility: private (only you and people you invite) or shared with your organisation.
4. Click Set project instructions, write your standing guidance, and save.
5. Click + in the project knowledge panel to upload files, paste text or add content from GitHub.
6. Start chatting inside the project.

You can also move an existing chat into a project from the dropdown next to the chat name (“Add to project”), and star projects you use often so they appear in the sidebar.

Writing good project instructions

Project instructions apply to every chat in the project, so treat them like a short brief rather than a single prompt. A pattern that works:

You are helping me write and maintain the documentation for Acme's billing API.

Audience: backend developers integrating for the first time.
Style: US English, short paragraphs, code samples in Python and curl.
Always: check answers against the files in project knowledge; say so if
something isn't covered there instead of guessing.
Format: Markdown with H2 headings; tables for parameters.

Keep instructions focused on what stays true across chats. Put task-specific detail in each chat. For more on writing instructions Claude follows, read our Claude prompting best practices.

Project knowledge and RAG

On paid plans, Claude handles large projects with retrieval augmented generation (RAG). According to Anthropic’s RAG article:

• When project knowledge approaches the context window limit, Claude automatically switches to RAG mode, which expands capacity by up to 10x.
• Instead of loading everything at once, Claude uses a project knowledge search tool to pull the most relevant passages for each question. You’ll see it call that tool in the chat.
• There’s no setting to toggle; activation is automatic, and a project can switch back to in-context mode if its knowledge shrinks below the threshold.
• RAG works with web search, extended thinking and Research.

To get the most out of a large project, Anthropic recommends clear, descriptive file names, keeping related material in the same project, and naming specific documents in your questions so retrieval looks in the right place.

On the Free plan there’s no RAG, so once knowledge fills the context window you can’t add more. Trim or split the project instead.

Adding a codebase from GitHub

Projects can pull files straight from GitHub. In the knowledge panel, click +, choose GitHub, search for or paste a repository URL, and pick the files and folders to include. Use Sync to refresh to the latest version and Configure files to change the selection. Anthropic’s advice is to start with a small, relevant subset rather than a whole repository.

If you want Claude to actually edit and run code, though, a project is the wrong tool. Use Claude Code, which works directly in your repository.

Memory in projects

Memory is on by default for Free, Pro and Max on web, desktop and mobile; on Team and Enterprise an owner has to enable it. Each project has its own memory. Inside a project, Claude uses that project’s memory plus your general memory from chats outside projects, but not memory from other projects. To keep a project fully separate, turn off “Use account memory” on the project’s page.

Sharing projects on Team and Enterprise

On Team and Enterprise plans you can share a project’s knowledge and instructions with colleagues:

Permission	Can do
Can view	See contents, knowledge and instructions, and chat in the project
Can edit	Also change instructions and knowledge, manage members and contribute

You can invite people by email, paste a list for bulk sharing, or make the project visible to everyone in the organisation, where it shows up under the Organization tab. On Enterprise, sharing with a group is in beta, and access then follows group membership.

Two things to know. First, your chats stay private even in a shared project unless you share them individually. Second, owners can turn off project sharing for the organisation (or for specific roles on Enterprise); existing shares stay in place when they do.

Archiving and deleting

Archive finished projects to declutter your list. Archived projects keep their members, permissions and knowledge, and you can still open their conversations. You can’t delete an archived project directly; unarchive it first, then choose Delete from the project menu.

The new version of Projects (beta)

Anthropic is rolling out a redesigned version of Projects, starting with Claude Code. In it, a project is one conversation: you describe what you need and Claude splits the work into parallel threads that run in the cloud, each starting with the project’s files, repositories, instructions and memory. The Help Center says it’s in beta for select Pro and Max users who use Claude Code, appearing under “Projects” at claude.ai/code and in the Code tab of the Claude desktop app. Chat, Cowork, Team and Enterprise are slated to follow, and moved projects keep their instructions, files, chat history and scheduled tasks. Running several threads at once uses your plan faster.

Ideas for using Projects

• Client or account workspace: contracts, briefs and past deliverables, with instructions on tone and format.
• Course or research topic: papers and notes, with instructions to cite the file each answer comes from.
• Product documentation: specs plus a style guide, so every draft follows the same rules.
• Codebase Q&A: key folders synced from GitHub for explaining code to new teammates.

If you find yourself repeating the same procedure across projects, package it as a skill instead; see our Claude Code skills guide. And to give a project live access to tools like Drive or Slack, add connectors, which are built on MCP.

Related guides

• What is Claude? for an overview of every Claude product.
• Claude models explained to pick the right model for project work.
• Claude vs ChatGPT if you’re weighing Claude against ChatGPT.
• What is a Claude kit? for building your wider toolkit around Claude.