Claude Projects & Artifacts

A normal chat starts blank every time. Projects fix that: a Project is a self-contained workspace with its own chat history and a knowledge base of files every conversation inside it can see.

Artifacts are the other half. When Claude produces substantial, standalone content — a document, code file, SVG, flowchart, or interactive component — it opens in a dedicated window beside the chat instead of getting buried in the message stream, where you can edit, version, download, and share it.

How the four building blocks fit together:

In the left sidebar, open the Projects panel and create a new Project. Name it after the job, not the tool — "Weekly Investor Update" beats "Claude Stuff". The most valuable field is custom instructions: persistent directions Claude applies to every chat in the Project, so you stop re-explaining yourself.

A strong instruction block covers role, audience, tone, and format:

> You are my editorial assistant for a weekly investor update.
> Audience: busy non-technical investors. Write in plain English,
> no jargon. Keep paragraphs to three sentences. Lead with the
> single most important number. End every draft with a one-line
> "What changed since last week".

The structure: who Claude is, who reads the output, the rules, the required shape. Vague instructions ("be helpful") waste the feature; specific ones compound across every future chat.

Inside the Project, add files to the knowledge base — the documents that define your context: a style guide, a glossary, last quarter's report. Every chat draws on these without re-attaching them, and you can add documents, text, code, or other files.

Once files are loaded, test that Claude uses them:

> Using only the style guide in this Project's knowledge base,
> rewrite this paragraph to match our house tone:
> "The thing went up a lot which is good for us we think."

If the rewrite reflects rules from your uploaded guide, the knowledge base is wired in correctly.

On paid plans, Claude searches a larger knowledge base using retrieval rather than reading everything at once, letting a Project hold far more material than a single chat; free accounts get a smaller capacity.

Artifacts appear automatically when Claude produces substantial content — generally over about fifteen lines you'd want to edit or reuse. Supported kinds include documents, code, HTML pages, SVG images, diagrams and flowcharts, and interactive components.

Ask for something concrete and the panel opens to the right:

> Create a one-page onboarding checklist for a new hire as a
> clean, printable document. Sections: Day 1, Week 1, Month 1.
> Use checkboxes and short action items.

The document opens beside the chat, and you refine it in place ("Add a Manager sign-off line to each section; make Day 1 bold"). Each change creates a new version; use the panel's version controls to step back and compare, and its buttons to copy (for an email), download (to save locally), or publish (to share a link) — copying and downloading work on every plan.

The reuse workflow: build a template Artifact once, then each week start a fresh chat and ask Claude to fill it with new data — the Artifact is your scaffold and the chat supplies this week's content.

Sometimes you don't want a whole Project — just one file to interrogate now. Use the attachment (paper-clip) control in the message box to upload straight into a chat. Claude reads PDF, DOCX, CSV, TXT, HTML, RTF, EPUB, JSON, and XLSX spreadsheets, plus images in JPEG, PNG, GIF, and WebP.

Different file types invite different prompts:

> Summarise this report in five bullet points an executive can
> act on. Then list any numbers that look inconsistent.

> This is our sales spreadsheet. Which three products grew
> fastest month over month, and which one is quietly declining?

For an image, try: "Read this screenshot of an error message and tell me, in plain language, what's failing and the likely cause." You can attach several files to one chat, and individual chat uploads allow a notably larger per-file size than files added to a Project's knowledge base, so big one-off documents are best dropped straight into a chat.

The context window is everything Claude can "hold in mind" at once: your instructions, the conversation so far, and every file in play. It's large but finite, so a very long chat or many big files can push the earliest parts out of focus and answers drift. The warning signs: Claude forgets a detail you set earlier, re-asks for a file you already uploaded, or replies get slower and vaguer.

Three habits keep you sharp:

The recap move is the most useful:

> Summarise everything we've decided so far as a tight bulleted
> brief I can paste into a new chat to continue.