Anthropic's publishing surface for Claude: private pages and typed documents, made from chat, the Artifacts tab, Claude Code or mobile. They can hold data, show it to the people you choose, and call those people's own connected apps. They have no server of their own. The limits below are the ones Anthropic documents, as of November 2026.
An artifact is a page or a typed document that lives on claude.ai, opens for the people you share it with, and runs on their own Claude plan and connections. It is not a hosted app with a backend.
Artifacts started as a side panel in chat: Claude writes a page, you look at it, you iterate. That is still the core use. What has changed is that artifacts now come in typed templates (Docs, Slides, Design, Data, Sheets, Whiteboards, Motion), can be shared with specific people or an organisation, and can read and write through the connected apps of whoever is looking at them.
Two things are easy to get wrong when you read older guides. Several third-party guides still say artifacts cannot store data, or cap them at 20 MB. Those statements describe the legacy kind of artifact only. The second is that "artifact" now covers more than a web page: a Docs document is an artifact too.
Artifacts made in a chat before 16 September 2026 are legacy artifacts. They keep working and can still be published and shared. They cannot be newly created.
Availability and export formats as the Help Center lists them. The Help Center marks only Sheets, Whiteboards and Motion as beta. Check the status of any template you plan to depend on, because beta status can change.
| Template | Plans | Exports to | Status |
|---|---|---|---|
| Docs | All plans, including Free | Word, PDF, Markdown, Google Docs | Not marked beta |
| Slides | All plans, including Free | PowerPoint, PDF, Google Slides | Not marked beta |
| Design | All plans, including Free | .zip, PDF, PowerPoint, standalone HTML | Standalone Claude Design closes 14 Dec 2026 |
| Data | Pro, Max, Ultra, Team, Enterprise | Not stated | Not marked beta |
| Sheets | Pro, Max, Ultra, Team, Enterprise | Not stated | Beta |
| Whiteboards | Pro, Max, Ultra, Team, Enterprise | Not stated | Beta |
| Motion | Team and Enterprise | MP4 | Beta |
Docs is the template most people will hand to colleagues. Its sharing rules are stricter than a page's.
Owner, editor, commenter and viewer. Anyone can comment with @Claude in a comment.
Sign-in is required. A link viewer gets view-only access.
Deleting a doc is permanent. There is no trash to restore from.
Doc edits and comments are not yet recorded in the Compliance API. Don't rely on the Compliance API for doc activity until Anthropic says otherwise.
An artifact runs in the viewer's browser and calls out only where it is allowed to.
Reads and writes go through each person's own connected apps, custom connectors included. Pro and above, on web and desktop. First use needs approval. Tools that need approval for each action are not available to artifacts.
A call from an artifact counts against the viewer's own plan limits.
Personal or shared. Shared storage shows the same data to everyone who opens the artifact.
Because connectors run as the viewer, two people can open the same page and see different data.
Sharing depends on your plan and your admin's settings. The Help Center has the full plan-by-access table; the rules below are the ones that most often catch people out.
Each artifact can be shared with up to 50 outside invitees. Pending invitations expire after 30 days.
An artifact that connects to other websites cannot use link sharing or email invitations.
Claude and connectors do not work for outside viewers. An outside viewer sees the page and its data, not the assistant.
Admins can switch each template on or off for the organisation.
A separate switch controls whether artifacts can be shared outside the organisation.
One organisation-wide switch. There is no per-app limit, so an admin cannot allow artifacts to reach some apps and not others.
Artifact use draws on seat limits or, for usage-based Enterprise, API rates.
Claude Design does not support data residency requirements.
Claude Code can publish a session's output as an artifact. It is a narrower thing than a claude.ai artifact.
One self-contained page with no backend. It cannot serve multiple routes or authenticate viewers itself. Sources can be .html, .htm or .md. The rendered page is capped at 16 MiB.
Its content security policy allows Google Fonts and five public CDN hosts. Network calls reach only the page's own origin, so any other outside call has to go through a connector.
Available on Pro, Max, Team and Enterprise. It is not available in sessions on Amazon Bedrock, Google Cloud or Microsoft Foundry, or in API-key sessions. Content is stored on Anthropic-operated infrastructure under the retention policy. Admins see claude_artifact_* audit events, and three Compliance API endpoints cover artifacts.
Connector calls from a Claude Code artifact need Claude Code v2.1.209 or later. The /slides and /design commands need v2.1.265 or later.
Honest tradeoffs. For the full argument on whether an artifact should be your long-term home, see the decision leaf.
Residency. Claude Code artifact content is stored on Anthropic-operated infrastructure. The Team and Enterprise admin guide gives no residency commitment for artifacts. Treat personal information in artifact storage as a POPIA Section 72 cross-border transfer, the same position this tree takes on Claude inference. See the residency validation page.
Cost. There is no hosting line item, which makes an internal tool look free in rand terms. The cost moves to seat limits, and a heavy shared dashboard shows up there.
Checked 21 November 2026. Artifacts change often; verify against these pages before you depend on a limit.