Claude Projects: what they are and how to use them

A Claude Project is a persistent workspace: a knowledge base of files, custom instructions, and its own chat history, all loaded into every conversation you start inside it. It matters because you stop re-explaining the same background, your style guide, a client's brief, your class notes, every time you open a new chat.

What's actually inside a Project

Three things live inside a Project, and knowing what each one does clears up most of the early confusion.

A knowledge base holds whatever files or text you upload: past reports, a style guide, a syllabus, code, anything Claude should treat as background truth for this workspace. Project instructions are custom directions you set once (who you are, what the project's for, how you want responses formatted) that apply to every chat inside it, per Claude's help center. And each Project keeps its own chat history, kept separate from your other chats and other Projects.

Here's the part beginners usually get wrong: being in the same Project doesn't mean one chat can see another chat's conversation. Anthropic is explicit about this: "context is not shared across chats within a project unless the information is added into the project knowledge base." The knowledge base is what carries over, not the transcript.

On top of that, each Project keeps an isolated memory: a running summary Anthropic generates from your conversations in that workspace, refreshing roughly every 24 hours, fully editable from Settings, or just tell Claude in-chat to remember or forget something. It doesn't leak into other Projects or your regular chats. You can pause it, or wipe it entirely, though a full reset can't be undone.

Projects vs. Skills vs. Artifacts

This is the mix-up almost everyone hits first.

A Project is a container of background knowledge, always loaded whether or not it's relevant to what you just asked. A Skill is the opposite: a packaged procedure that only activates when your request matches what it's built for, and it works in any chat, not just one tied to a Project.

An Artifact isn't context or process at all, it's the thing Claude produces: a document, a page, a chunk of code, sitting in its own panel so you can keep iterating on it.

None of these need the others. You can use Skills and get Artifacts in a completely plain chat with no Project involved. A Project is just the workspace some of those chats happen to live in.

Setting one up

Real ways people use one

A recurring report. Upload past examples plus the house style guide once, and every new chat already knows the format and tone, so you just paste this period's raw numbers.

A client workspace. One Project per client, with that client's brief, brand guidelines, and past deliverables in the knowledge base, keeps each client's context separate and stops advice from bleeding between accounts.

A course companion. Upload the syllabus, lecture notes, and readings for one class so questions get answered against your actual course materials instead of generic web knowledge.

A codebase companion. Drop in architecture notes or a style guide for a specific project, then use that workspace for debugging and questions tied to it. Instructions can pin conventions, like always using parameterized queries or matching existing naming.

A job search. Résumé, past cover letters, and target-role notes in one Project so every new draft starts from consistent background instead of you re-explaining your history each time.

What it won't do

The knowledge base has a base window of 200,000 tokens. Once a project's knowledge approaches or crosses that, Claude automatically switches into retrieval mode: it searches for and pulls in the most relevant chunks per question rather than holding every file in memory at once, which can expand effective capacity by up to 10 times.

That's useful for large libraries of documents, but it also means a question that needs synthesis across many scattered files can miss something a smaller, fully-loaded project wouldn't, per Claude's help center. There's no official guarantee of full recall once retrieval mode is active.

Which plans get that retrieval mode is genuinely unclear right now: two of Anthropic's own support pages disagree with each other on whether the free tier gets it. Don't assume either way if it matters for your use case, check the current plan comparison yourself.

A few other limits worth knowing before you build a workflow around this:

• Free accounts cap out at 5 Projects total. Paid plans (Pro, Max, Team, Enterprise) have no stated limit.
• Memory reset is permanent. Pausing memory (stop new entries, keep what's there) is the reversible option if you just want it to stop learning for a while.
• No published ceiling on file size or file count for the knowledge base itself. Neither official article states one, so don't assume you can dump an unlimited archive in and expect consistent recall.
• Sharing a Project with teammates is Team/Enterprise only, with "can use" (chat, no edits) and "can edit" (change instructions and files) permission levels. Archiving a shared Project resets everyone's access back to private.

Tested on Claude Sonnet 5, July 2026: starting a fresh chat inside an existing Project pulled in the uploaded files and instructions immediately, no re-attaching or re-explaining needed, matching what the documentation describes.

This site isn't run by Anthropic. We're an independent guide to using Claude well, and when a feature has a genuine gray area, like the free-tier retrieval question above, we say so instead of guessing.

What is a Claude Project?

A persistent workspace inside Claude that bundles a knowledge base of your files, custom instructions, and its own separate chat history. Everything in it loads automatically into every new conversation you start there, so you're not re-pasting background each time.

Are Claude Projects free?

Yes. As of early 2026, Projects are available on every plan, including free accounts, though free users are capped at 5 Projects total. Paid plans have no stated cap.

How are Projects different from Skills?

A Project is always-on background knowledge, loaded into every chat in that workspace whether it's relevant or not. A Skill only activates when a specific request matches what it's built to do, and it works everywhere, not just inside one Project.

Can I share a Claude Project with my team?

Only on Team or Enterprise plans. You can grant "can use" access (chat, no edits) or "can edit" access (change files and instructions) to individuals or an entire org. Archiving a shared Project resets all of that back to private.

If you're earlier in the process, how to use Claude covers your first session end to end. For the wider set of write-ups, browse more reviews.