← All InsightsClaude Agentic Features

Guide: Live Artifacts (Stretch)

Rob Floyd11 min read
Editorial illustration for "Guide: Live Artifacts (Stretch)"
Hero illustration generated with OpenAI (gpt-image-1)
Stretch guide

A live artifact turns a one-time answer into a page you re-open and it refreshes itself.

Instead of re-asking for the same dashboard, you build it once as a persistent page that pulls fresh data from your connectors every time it loads.

If you'd ask for the same view again tomorrow, build it as a live artifact instead of re-running the question.

Stretch Guide: Live Artifacts

Claude Cowork · Advanced feature


⚠️ Not a beginner’s starting point. Live Artifacts require at least one connected data source (a connector) and a clear sense of what you want to display. If you haven’t run a few standard Cowork recipes and you haven’t connected any external tools, this guide is ahead of where you are. Get a workflow running first, then come back here when you want to turn it into something you can reopen every day.


What a Live Artifact Is

A Live Artifact is a self-contained HTML page that Claude builds and saves for you. Unlike a regular Cowork output — a document, a spreadsheet, a summary — a Live Artifact persists across sessions and pulls fresh data from your connected tools every time you open it.

Think of it as turning a one-off answer into a page you can keep coming back to.

A pipeline status report you run every Monday becomes a page that’s always current when you open it. A cash position summary becomes a dashboard that refreshes on demand. A project tracker becomes a live board rather than a static file.

What a Live Artifact can do:

  • Call your connected tools to fetch live data (CRM, accounting software, calendars, Drive)
  • Run a quick Claude analysis over that data inline — summarize, classify, flag
  • Display charts (Chart.js), tables (Grid.js), and diagrams (Mermaid)
  • Remember your filter and sort preferences across sessions
  • Trigger a scheduled task you’ve already set up

What it can’t do:

  • Pull data from tools you haven’t connected
  • Run without an internet connection
  • Replace a proper dashboard tool for complex multi-source analytics

Prerequisites

  • [ ] You’ve run standard Cowork workflows and understand how Projects work
  • [ ] You have at least one connector authenticated — Google Drive, your CRM, calendar, or accounting software
  • [ ] You know what data you want the artifact to show, and you know which connector has it
  • [ ] Subscription: Pro or above (same requirement as Cowork)

Step 1 — Decide What You Want to See

Before asking Claude to build anything, be specific about the artifact’s purpose.

Ask yourself:

  1. What question does this page answer every time I open it?
  2. What data does it pull to answer that question, and from which connector?
  3. Will I want to filter, sort, or interact with it — or just read it?
  4. Will I share this with anyone else, or just use it myself?

The clearer your answer, the more useful the artifact. Vague requests produce vague pages.

Good candidates for Live Artifacts:

  • “Show me my open pipeline deals ranked by close date” (needs CRM connector)
  • “Show my cash position and outstanding invoices” (needs accounting connector)
  • “Show this week’s calendar with prep status for each meeting” (needs calendar connector)
  • “Show the status of all active projects and any items overdue” (needs project tool connector)

Poor candidates (better as a one-off Cowork task):

  • Something you’ll only look at once
  • Something that doesn’t involve live data
  • Something so complex it needs multiple unconnected sources

Step 2 — Probe the Data First

Before asking Claude to build the artifact, ask it to fetch the raw data and show you what it looks like.

  1. Open a Cowork session.

  2. Ask Claude to call the connector and return the data you want:

    “Pull my open deals from [CRM] and show me what fields are available.”

    “Fetch this week’s calendar events and summarize them.”

  3. Look at what comes back. Does it have the fields you need? Is anything missing or named differently than you expected?

This step saves time. Building an artifact on a misunderstood data shape means rebuilding it. Two minutes of probing first prevents that.


Step 3 — Ask Claude to Build the Artifact

Once you know your data shape, describe the artifact you want:

“Build a Live Artifact that shows my open pipeline. Use [CRM connector]. Display a table with deal name, company, amount, close date, and stage. Sort by close date by default. Add a summary card at the top showing total pipeline value and count of deals closing this month.”

Be specific about:

  • Which connector to use
  • What data to display
  • How to organize it (table, cards, chart)
  • Any summary metrics you want at the top
  • Any filters or sorting you want to control

Claude will build the page and save it. You’ll get a link or card you can reopen.


Step 4 — Review and Refine

  1. Open the artifact. Does it load? Does the data look correct?

  2. Click Reload (built into every artifact header) to confirm it fetches fresh data.

  3. If something is wrong — a field missing, a column mislabeled, the layout off — describe the fix:

    “The close date column is showing as a timestamp — format it as a readable date. And move the summary cards above the table.”

  4. Test any filters or interactive elements you asked for. Do they persist when you close and reopen the page?

Before relying on it: compare the data in the artifact against the source system. Does the pipeline count match what you see in your CRM? If not, the connector may be returning partial data or there’s a filter applied you didn’t intend.


What Can Go Wrong

Connector not authenticated. The artifact opens but shows no data, or shows an error. Check that the connector is still authenticated in your Cowork settings. Re-authenticating usually fixes it.

Data shape changed. If the connected tool changed a field name or structure, the artifact may break or show blank columns. Ask Claude to rebuild it against the current data shape.

Stale data. Artifacts cache reads to avoid hammering connectors on every interaction. If you need truly real-time data, hit Reload. If data still looks old after reloading, the connector may have a sync delay on their end.

Too many data sources. Artifacts that try to pull from three or four connectors simultaneously are slow and fragile. Start with one connector. Add a second only after the first is working cleanly.

You asked for something Claude can’t display well in HTML. Complex multi-tab financial models, large document sets, or anything requiring pixel-perfect layout don’t translate well to an artifact. For those, a Google Sheet or a Word doc is the better output.


When It’s Worth Building

A Live Artifact earns its setup time when:

  • You’ll open it more than once a week
  • The underlying data changes regularly and you need to stay current
  • The answer it gives you drives a recurring decision
  • You want to share a live view with someone else without re-running a report

If you’d only check it once, just run the Cowork task directly and read the output. The artifact is for things you want to bookmark.


Further Reading

About the Author

Written under the direction of Rob Floyd, founder of Eikon Digital. Rob runs the BOSNet and BOSGov product lines and writes about AI as production infrastructure — the operational, governance, and content workflows that turn an LLM from a clever toy into a business asset that earns its keep.

Drafting and structural editing performed by Claude (Anthropic) under Rob's orchestration, with every word vetted against the Eikon voice guide before publication.

Authored by Rob Floyd · Drafted with Claude (Anthropic) · Eikon Digital, 2026

Related Articles

Guide: Dispatch (Stretch)
Claude Agentic Features

Guide: Dispatch (Stretch)

Assign tasks to Claude from your phone and have it work on your desktop remotely. A power-user feature — with the warnings you should read first.

Guide: Scheduled Tasks
Claude Agentic Features

Guide: Scheduled Tasks

How to make Claude run work on a schedule — every morning, each Monday, or once at a future time — and when a recurring task beats doing it by hand. The automation behind the hands-off recipes.

Guide: Agents
Claude Agentic Features

Guide: Agents

What an agent is, how it differs from a Project, when Claude hands work to a sub-agent, and how to know whether you need one. The worker model behind the more involved recipes.

Stay Informed

Get insights on AI and digital transformation delivered to your inbox.