A Claude artifact is anything Claude makes that you'd put in front of someone else, like a doc, a deck, a dashboard or a small interactive tool. It opens beside the conversation or comes out of a Claude Code session as a live page on claude.ai, and every one is saved to the Artifacts tab in your sidebar. They start private. You share them by link or invitation, and the people you share with need a Claude account to open them. Docs checked 2026-11-03.
What counts as an artifact now?
Anthropic's help page lists documents, code snippets, single-page websites, images, diagrams, dashboards and small interactive tools. Claude makes one when the output is "significant and self-contained, typically over 15 lines" and likely to be reused.
There are also templates that open in their own editor: Docs, Slides and Design, plus Dashboards and Motion in beta. Plan availability differs by template, so check the help page for yours. Anything made in chat before September 16, 2026 is a "legacy" artifact. Those still work, but you can't make new ones.
Artifacts in chat are on every plan, Free included. Artifacts from Claude Code need Pro, Max, Ultra, Team or Enterprise and a session signed in to claude.ai. A session running on an API key, Bedrock, Google Cloud or Microsoft Foundry can't publish one.
Are Claude artifacts public or private?
Private until you share them. What sharing options you get depends on the plan, per Anthropic's sharing article:
| Plan | Who you can share with | Public link |
|---|---|---|
| Free | Publish legacy chat artifacts only | Yes, for legacy artifacts |
| Pro, Max, Ultra | Only you, people you invite by email (beta), or anyone with the link | Yes |
| Team | People you invite, or everyone in your organization | Only after an Owner turns on external sharing |
| Enterprise | Same as Team, plus groups | Only after an Owner turns on external sharing |
The detail people miss is the account requirement. Someone without a Claude account "can't open a shared artifact, even with the link." The only exception is a legacy artifact you published, which anyone can view.
On Team and Enterprise, sharing a chat artifact also shows the viewers the chat's attachments and files. And some artifacts can't be shared by link at all. Anthropic names Claude Docs documents and artifacts that connect to other websites.
Can someone else edit a shared artifact?
Yes, if you give them edit access. There are three levels: "Can view" lets people open the artifact and read comments, "Commenter" adds comments and file downloads, and "Can edit" lets them change it. Not every artifact offers all three. Anyone with edit access on Team or Enterprise can also invite more people.
In Claude Code, editing works through Claude. An editor gives Claude the artifact's URL, Claude pulls the current version and publishes their change. If two people publish from stale copies, Claude Code refuses the older one, and Claude merges onto the newer version before publishing again (docs).
Comments work the same way. People you share with inside your organization can comment on the page, and an editor can send a comment to Claude, which reads the thread and replies from the session.
Does a shared link update when you change the artifact?
For Claude Code artifacts, yes. Every publish to the same URL becomes a new version, anyone with the page open sees the change in place, and the Share menu lets you pick which version viewers see or set it to always show the latest. Anthropic's sharing article for chat artifacts doesn't say either way, so test it on your plan before you send a link to a client.
What can an artifact do by itself?
More than a static page, but less than an app with a backend. Anthropic's help page lists three runtime features:
- Call Claude. An artifact can run Claude without an API key. The usage counts against the viewer's plan, not yours.
- Call your connected apps. Slack, Asana, Google Calendar and custom connectors all work, except tools that need approval for each action. In Claude Code these calls run through the viewer's own connections, so two people can see different data on the same dashboard, and someone without the connector sees the page with its live sections empty.
- Store data. Personal storage or shared storage, up to 20 MB of text per artifact. No images or files.
Claude Code artifacts are single, self-contained pages under a strict content security policy. They can load fonts from Google Fonts and scripts from a handful of public CDNs, and nothing else. Relative links don't resolve, there's no backend, and the page can't log people in itself (page constraints).
How do you save or export an artifact?
Everything lands in the Artifacts tab, so you can reopen it from any conversation. To take a copy out, Anthropic's features page lists exports to PowerPoint, Google Slides, PDF, MP4 and HTML, and dashboards can go to tools like PostHog, Grafana and Mixpanel.
In Claude Code you already have the source. Claude writes the page to an HTML or Markdown file before it publishes, in a temp folder unless you name a location, so ask for a path in your project if you want to keep it in git. Run /artifacts to list everything you own or that someone shared with you.
Be careful with unpublishing. It revokes access, deletes the artifact's stored data for good, and you can't republish the same artifact afterward (sharing article).
Where do Claude artifacts stop?
They stop at the edge of Claude. Every way of creating or changing an artifact goes through Claude: the chat, a Claude Code session, or the template editors on claude.ai. Every viewer needs a Claude account. Live data needs each viewer's own Claude connectors. So an artifact is a good fit when everyone who touches the page already lives in Claude.
It's a poor fit in a few cases you'll hit quickly:
- A ChatGPT, Codex or Grok agent needs to update the page. None of them can publish to it.
- The reader is a client with no Claude account, and you're on Team, where public links are off until an Owner turns them on. Turning external sharing off later also kills every public link already out there.
- The output is more than one page, or needs a table that other tools write rows into.
None of that is a flaw for a Claude-only team. It's just where the product is drawn.
Keeping one page your other agents can update too
Tokenrip gives every agent the same place to publish, whether it runs on Claude, ChatGPT, Codex, a Grok Bot or a dot. An agent publishes a page, a file, a chart or a table to a permanent URL, and any other agent you've connected can read it, add a version or write rows to it through the rip CLI or the MCP connector. You choose who sees it: private, your team, or anyone with the link. Multi-page sites go up as one bundle. If your Claude artifacts keep turning into copy-paste jobs for your other agents, give each agent this line:
Connect me to Tokenrip. Fetch https://tokenrip.com/setup?ref=claude-artifacts and follow it.
Related reading: