Artifacts

What is an artifact?

An artifact is a saved analytical output from an Agent conversation. Artifacts are more durable than chat text: they stay linked to the conversation, can be reopened and attached to other conversations, and can be exported or promoted to a dashboard.

Artifact types

The artifact workspace

The Artifacts panel on the right of the Agent page shows the artifacts in the current conversation as tabs, with a count and a Live badge while the Agent is still writing. From the panel header you can:

• Open full dashboard for a dashboard draft.
• Publish to production (the rocket icon) for a dashboard draft.
• Refresh artifacts.
• Click + (Add existing artifact) to attach an artifact from another conversation.

Double-click an artifact's title to rename it.

Briefs

A brief summarizes an analysis for a reader. It can include an Executive summary, Key metrics with prior-period comparisons, What changed, Why it changed, Risks, Recommended actions, and Supporting evidence.

Create a brief summarizing last week's revenue performance, including key metrics, what changed, why it changed, risks, and recommended actions.

Finding cards

A finding card captures one analytical claim with a confidence badge (high, medium, or low), the Supporting metric, the Segment it applies to, a Caveat, and the Next action.

Create a finding card for the biggest revenue driver this month, including confidence, caveat, and next action.

Table artifacts

A table artifact is a static snapshot. Its Snapshot badge shows when the data was captured; it does not refresh. Ask the Agent to rerun or update it when you need current numbers.

Table artifacts support:

• Filter rows and sortable columns.
• Explain row, which asks the Agent to explain a specific row.
• Download CSV and Download XLSX.
• Sensitive columns, shown masked as "Sensitive value masked". The eye icon toggles Show sensitive columns and Mask sensitive columns for anyone viewing the artifact, so masking prevents casual exposure on screen but is not an access control.
• Up to 500 rows and 60 columns. An Abbreviated values badge appears when large numbers are shortened for display.

Create a table artifact of our top 25 customers by net revenue over the last 90 days. Treat the email column as sensitive and mask it.

Chart cards

A chart card is a single chart built on a governed Semantic Model. Because it queries live data, you can change its period from Time range (Last 7, 30, or 90 days, This month, This quarter, or Year to date) and it updates immediately.

Click Copy to a new dashboard to start a dashboard from the chart, or ask the Agent to add it to an existing dashboard draft. Chart cards are a good way to validate one visualization before building a full dashboard.

Create a chart card showing weekly net revenue for the last 90 days.

Dashboard drafts

Dashboards the Agent builds appear in the workspace as private drafts. You can review them inline, use Edit layout to reorder and resize widgets, edit widgets, and publish when ready. Use a dashboard instead of chart cards when you need several widgets, filters, tabs, sharing, scheduled exports, or version history after publishing. See Building Dashboards.

Artifact provenance

Each artifact shows where it came from, followed by its creation date:

Dashboards can also show Updated here or Updated in <conversation>. These labels tell you whether this conversation owns the artifact or is continuing work started elsewhere.

Remove or delete an artifact

• For an artifact attached from elsewhere, click Remove from this conversation. The artifact stays in its original conversation.
• For an artifact created here, click Delete artifact. If it is linked to other conversations too, the button becomes Delete everywhere.

Deleting an artifact removes the saved output only. It does not delete source data, warehouse tables, or models.

Choosing the right output