What is a Claude Artifact? Types, editing, and publishing

What is a Claude Artifact

A Claude Artifact is a substantial, self-contained output that opens in its own panel beside the conversation: a document, code file, diagram or interactive tool you inspect and revise while the chat continues. It stays inside your Claude account until you Publish it, which creates a public link, or Share it within a Team or Enterprise organization. One thing decides most questions about artifacts: Anthropic uses the word for four experiences with different creation, sharing and plan rules.

Four things called an artifact

What counts as an artifact

Anthropic's artifact guide describes content that is substantial, self-contained, useful outside the immediate answer and likely to be edited or reused, typically over 15 lines: Markdown and text documents, code, single-page HTML, SVG, diagrams, interactive React components.

Interactive is not the same as AI-powered. A calculator whose JavaScript runs in the browser is interactive; it becomes AI-powered only when it calls Claude for its user.

Artifacts require Code execution and file creation in Claude's capability settings. If Claude keeps answering in the chat instead of opening the panel, check that first, then ask directly and name the deliverable:

Create this as a Claude Artifact. Keep the result self-contained so I can
revise, inspect, and share it separately from our conversation.

Create, edit, and keep the right version

1. Step 1CreateAsk for a self-contained artifact in a conversation, or start one from the Artifacts area.
2. Step 2ReviseDescribe the change in chat; for Markdown, select the exact text and use Edit with Claude.
3. Step 3InspectUse the version selector, choose among several artifacts in the chat, read the underlying code.
4. Step 4Export or distributeCopy or download, or select the intended version before you Publish or Share.

Three things the steps do not show:

• Changing the artifact does not rewrite Claude's memory of the earlier content, so say when a revised fact should guide the rest of the conversation.
• The old Remix action on someone else's published artifact is gone. Anthropic's publishing guide now says to copy the code into a new chat, which creates an independent artifact.
• An artifact created in a conversation does not appear in the sidebar's Artifacts collection until it is published. Keep the source conversation while it is a draft, because Publish also changes who can access it.

Where the four meanings diverge

AI-powered artifact

Claude writes the tool, Anthropic runs it, and each user authenticates with their own account: no API key to paste, no shared limits. Published artifacts can also use storage and outside connections: Anthropic documents 20 MB per artifact, text only, separate personal and shared stores (a shared store shows a confirmation on first use), and connections over the Model Context Protocol that every user authorizes for themselves. Why the connection belongs to the user rather than being a password inside the artifact.

Claude Code artifact

Anthropic's artifact documentation for Claude Code describes a self-contained .html, .htm or .md file published from a session to a URL like claude.ai/code/artifact/5fbea6f3-…, served from a sandboxed *.claudeusercontent.com origin, updated at the same address as the session continues, each publication a version. Good for annotated pull-request walkthroughs, investigation timelines and dashboards from session data. Documented limits: one page, no backend, no stored form submissions, no routes, strict external-request limits, 16 MiB rendered. A later session needs the URL to update the page; without it, it creates another artifact.

Cowork live artifact

An interactive HTML view around ongoing work that can refresh from connected services. Anthropic's live-artifact guide says it is a paid Claude Desktop feature that lives on the computer that created it: Pro and Max cannot share it, Team and Enterprise share inside the organization, and a recipient opens it in Claude Desktop with their own access to the connected services. Live means a working view that refreshes, not a public website.

What it means to publish an artifact

• Publish, on Free, Pro and Max, creates a public link. Anyone with it uses basic functionality without an account; AI features need a Claude account and the viewer's limits.
• Share, on Team and Enterprise, keeps access inside the organization. Viewers sign in, and an artifact created in a Project also needs access to that Project. Sharing also gives them the attachments and files from the conversation that created it.

A published artifact can be embedded on a site whose domain the publisher lists. Before distributing, check the version, the source material the link exposes, and whether it uses personal or shared storage.

Artifact, Project, Canvas

An Artifact is the output. A Project is the context workspace: chats, knowledge, instructions; it can contain chats that create artifacts. ChatGPT Canvas is the editing surface of another product. Choose a Project when several chats should share background; an Artifact when one result deserves its own revision surface.

When a Birta page is the better artifact

The artifact is the finished deliverable while the work is between you and Claude: a reviewable document, a prototype, a session walkthrough, a small tool whose Claude access model suits everyone using it. The moment a third person enters, the same conversation can publish it as a page on Birta, and the difference is what happens after you share the link.

Why an artifact link is not a website owns the ownership argument; the same-chat hosting guide has the transfer steps for a static artifact.

Frequently asked questions