Docs

Artifacts

Artifacts are rich rendered outputs that agents can produce alongside chat messages. Instead of dumping raw HTML or a wall of markdown into the chat, the agent attaches the output as an artifact that opens in a slide-in side drawer — clean, interactive, and persistently accessible.

Supported types

htmlHTML
Full HTML/CSS/JS pages. Use for data visualizations, dashboards, mockups, calculators, and interactive tools. Rendered in a sandboxed iframe.
Examples: Charts, UI component previews, interactive reports
svgSVG
Vector graphics and diagrams. Use for architecture diagrams, flowcharts, and any structured visual output.
Examples: System diagrams, entity-relationship models, flowcharts
markdownMarkdown
Formatted documentation, reports, and summaries. Use when the agent is producing structured text that benefits from headers, tables, and code blocks.
Examples: Code review reports, sprint summaries, API docs

How agents create artifacts

Agents call the render_artifact MCP tool with the artifact type, a title, and the content:

MCP tool call
render_artifact({
  projectCwd: "/path/to/project",
  type: "html",
  title: "Sales Dashboard",
  content: "<html>...</html>"
})

The content is sent to the UI, which opens the artifact drawer and renders it. Agents can create multiple artifacts per response — each appears as a separate entry in the drawer.

What they're good for

ScoutPresenting research results as a formatted report with links and key findings.
BuilderGenerating a visual mockup of a UI component or page layout.
ReviewerProducing a structured code review with issue categories, severity, and line references.
OrchestratorCreating a project status dashboard with issue counts, agent activity, and progress charts.

In the UI

When an agent produces an artifact, a View artifact button appears in the chat message. Clicking it opens the artifact in the side drawer. The drawer slides in from the right and stays open while you continue chatting — you can reference the artifact while sending follow-up messages.

Multiple artifacts from the same session stay accessible throughout. A tab row at the top of the drawer lets you switch between them without re-opening.

Security

HTML artifacts render inside a sandboxed iframe. The sandbox configuration:

✓No access to your filesystem or credentials.
✓No ability to make network requests to external URLs.
✓No access to the parent window or other app state.
✓Scripts run in a restricted context — DOM manipulation and local calculations work, but external fetches are blocked.
Note:SVG and Markdown artifacts are rendered directly — SVG via a sanitised renderer, Markdown via a safe parser. Neither executes scripts.