Claude Live Artifacts, Explained: How They Work and What They Can't Do (2026)

By Marcin, founder of Markloop · Updated July 2026 · Facts verified against Anthropic's Cowork documentation

What is a live artifact?

A regular Claude artifact is a page frozen at the moment it was made. A live artifact is the same idea with the freshness problem solved: per Anthropic's docs, live artifacts "pull from your connected apps and local files so the view reflects today, not the day it was built."

In practice that means you ask Claude in Cowork for something like a pipeline dashboard, a project tracker, or a competitor-pricing reference - and instead of a one-off report, you get a page you can reopen next Tuesday and trust. Anthropic's own framing: "dashboards and trackers connected to your apps and files."

Live artifacts shipped in Claude Cowork in April 2026 and quickly became the loudest artifact news of the year. They're also the third meaning of the word "artifact" in the Claude ecosystem - which is genuinely confusing, so there's a comparison table below.

How live artifacts work

Three mechanics do the heavy lifting:

• Connectors, approved once. A live artifact uses the data sources you authorized when it was created or updated - and from then on, "live artifacts use your connectors without asking." It won't silently reach into a source you never approved.
• Refresh on open. Most artifacts refresh themselves through caching and re-querying when you open them. If you want to force it, there's a Reload button in the artifact header.
• Versions on every iteration. "Each time you iterate on a live artifact with Claude, the previous version is saved" - you can review changes, compare versions, and restore an earlier one.

Where they live: the Artifacts view

Since July 2026, Chat and Cowork share one home in the Claude app - one sidebar, one search, and you pick the mode per conversation with a toggle in the message box. Artifacts you make in chat and in Cowork land together in the Artifacts view in that sidebar; live ones carry a Cowork label, and a "Filter by Cowork" toggle shows them alone (the docs call this the Live artifacts tab). This view is the collection: reopen any artifact, check its version history, restore an earlier state. You can also start one from there directly - "New artifact" → "Create Cowork artifact".

Live artifacts vs claude.ai artifacts vs Claude Code artifacts

"Artifact" now means three different things in the Claude ecosystem, and their sharing rules differ more than their looks:

For the full story of all three kinds - including what the claude.ai and Claude Code flavors are good at - see our explainer: What are Claude artifacts?

What live artifacts can't do (yet)

The docs are refreshingly direct about the boundaries. As of July 2026:

1. They don't leave your computer. "Live artifacts live on your computer. If you switch devices, they don't come with you." No sync, no web version, no mobile.
2. Desktop only, paid plans only. Live artifacts require a paid plan (Pro, Max, Team, Enterprise) and the Claude Desktop app - macOS, Windows, Linux in beta. Cowork itself started rolling out to web and mobile in July 2026, but live artifacts stay desktop-side for now.
3. No external or public links. Sharing (the Share button in the artifact header) exists only on Team and Enterprise, and "sharing stays within your organization. There are no external or public links."
4. A shared artifact isn't your view of it. This one surprises people: "Shared artifacts use the viewer's access, not yours." When a teammate opens your dashboard, it re-runs on their connectors - if they can't see the data source, they don't see your numbers.
5. No comments. Like every other Claude artifact, a live artifact has no feedback mechanism. If a teammate spots a wrong number, that conversation happens in Slack, not on the page.

Five things worth building as a live artifact

Prompts that play to the format's strength - data that changes under a stable layout:

• A pipeline or sales dashboard. Build a live artifact showing open deals by stage from my CRM, with a stale-deals list. Refresh on open.
• A project tracker. Make a live artifact tracking open issues and PRs for project X, grouped by owner, with a "blocked" section on top.
• A meeting-prep brief. Build a live artifact that shows today's meetings with the related email threads and files for each.
• A competitor watchlist. Create a live artifact comparing our plans and pricing against competitors, refreshed from the research folder.
• A team status page. Make a live artifact summarizing what shipped this week from the changelog and merged PRs.

The pattern: describe the layout once, let the connectors keep it current. Iterate in Cowork - each iteration versions automatically.

When the artifact has to leave your machine

Sooner or later a live artifact produces something a person outside your Claude org needs to respond to: a client should approve the plan behind the dashboard, a domain expert should correct the assumptions, a contractor should answer the open questions. That's exactly where the format ends - no external links, no comments.

The way out is the same as with every Claude artifact: get the content out as a self-contained HTML file (ask Claude in the session - Save this as a standalone HTML file). One honest caveat: outside Cowork the page is a snapshot - it won't re-query your connectors. For a review round, that's usually what you want anyway: a fixed version people can mark up.

From there, Markloop turns that file into something reviewable: upload it (or push it from Claude Code over MCP), send one link - and reviewers comment on the exact element, answer your embedded questions, no account needed. Your agent pulls the feedback back as structured context and publishes the next version. The full walkthrough is in how to share a Claude artifact outside your organization.

FAQ

Related: What are Claude artifacts? claude.ai vs Claude Code vs Cowork · How to share a Claude Code artifact outside your organization · Artifact sharing options compared