Projects and Artifacts — organising real work

Working with Claude — CC BY 4.0

Most people use Claude the way they use a search box: open a chat, ask a thing, close it. That’s fine for one-off questions. It falls apart the moment you’re doing ongoing work — the same body of material, across days, where you need Claude to remember the ground rules and where you need a real document you can edit rather than a wall of chat text.

Two features cover that gap. Projects hold your context. Artifacts hold your working document. Neither is complicated, and both keep the important thing where it belongs — with you, in charge of what goes in and what comes out.

Projects — a workspace that remembers the ground rules

A Project is a self-contained workspace with its own chat history and its own knowledge base. You create one per body of work — a client, a course you’re writing, a legal matter you’re tracking, a book — and everything you do inside it stays in that context.

Two parts do the heavy lifting:

• A knowledge base. You upload the relevant documents, text, code or files once. Every chat you start inside that Project can draw on them, so you’re not re-pasting the same background each time. There’s a caveat below about what that means for accuracy.
• Project instructions. Standing directions that apply to every chat in the Project — tone, the role you want Claude to take, house rules, things it must never do. This is the same idea as the standing-instruction habit you’ll pick up in lesson 1.4, scoped to one body of work instead of everything you do.

Why this matters for reliable work: the instructions and the source material carry across conversations, so Claude behaves consistently instead of drifting from chat to chat. You set the rules once and they hold.

A worked example. Say you’re helping a community group respond to a council consultation. Make a Project called “Council submission”. Upload the consultation document, your group’s past submissions, and a one-page note on your position. In the Project instructions, write: “You are helping draft a formal submission to a local council. Plain English. Cite the specific clause of the consultation document you’re responding to. If you can’t find support for a claim in the uploaded files, say so — don’t fill the gap.” Now every chat in that Project starts from the same footing, and you can spin up separate chats for separate sections without losing the thread.

Availability and limits — confirm in-app. At the time of writing, Projects are available to everyone, including free accounts, with free accounts capped at a small number of Projects and paid plans not stated to have that cap. Paid plans add retrieval over larger knowledge bases. Plan availability and limits shift as the product changes, so check the current position in-app or in Anthropic’s help centre rather than taking a figure here as gospel.

The custody point. Uploading a document to a Project’s knowledge base means putting a copy on Anthropic’s systems. For anything sensitive — client files, personal information, anything covered by a confidentiality obligation or by data-protection duties — that’s a decision to make deliberately, not by habit. Check who you’re allowed to share the material with before it goes in. This is general education, not legal advice; for a specific obligation, take proper counsel.

Artifacts — the working document beside the chat

An Artifact is a proper working document that opens in its own window to the right of the chat, instead of scrolling past in the conversation. When you ask Claude to write something substantial, it puts the result in an Artifact you can read, edit and keep building on.

Artifacts cover the things you actually produce — documents in Markdown or plain text, code, single-page websites, diagrams and flowcharts, SVG images, and small interactive components. The point is that the output has a home: a canvas you return to and refine, rather than a message you have to hunt back through.

Editing works two ways. You can ask for changes in the chat — “tighten the second paragraph”, “make the tone more formal” — and Claude revises the Artifact. For text documents you can also edit in place: highlight the passage you want changed and use the in-place edit control (look for an “Edit with Claude” option on the highlighted text — confirm the exact wording in-app, as it changes). And there’s a version selector, so you can step back through earlier drafts if a change made things worse. That last part matters: nothing is a dead end, and you can always see what changed.

Turning it on. Artifacts are available across the plans, but the underlying capability may need switching on. In current builds it lives under the app’s settings, in a capabilities section, as a toggle for code execution and file creation — confirm the exact menu path in-app, since settings menus get renamed. If you’re on a Team or Enterprise plan, an admin may control this for the whole organisation.

How they work together

Projects and Artifacts are two halves of the same job. The Project holds the context and the standing rules; inside it, Claude produces Artifacts you shape into the finished thing. Context in, working document out — and the review, the corrections and the final say stay with you.

That’s the whole point of organising the work this way. You keep a tidy place where you can see what Claude was told, check what it produced, correct it, and own the result. The tools hold the material. You hold the judgement.