A     I
C H A   T
C H A R T
    A R T I F A C T
C H A R T I F A C T

Declarative, interactive data documents

Chartifact is a low-code document format for creating interactive, data-driven pages such as reports, dashboards, and presentations. It travels like a document and works like a mini app. Designed for use with your LLM to produce a shareable artifact of your analytic conversations.

• Examples • Try now with your LLM • Try with Copilot in VsCode

The Chartifact GitHub repo has source code for these interoperating modules:

• A document schema that defines plugins and components that communicate together with reactive variables:

  • Text – Markdown with dynamic placeholders
  • Inputs – Textboxes, checkboxes, sliders, dropdowns
  • Tables – Sortable, selectable, and editable data grids
  • Charts – Vega and Vega-Lite visualizations
  • Diagrams – Mermaid diagrams (flowcharts, networks, and more) via the mermaid plugin, including tabular data-driven diagram generation
  • Structured content – Custom layouts via the Treebark plugin for rendering safe, templated HTML structures (cards, headers, footers, lists, and more)
  • Images – Dynamic image URLs based on variables
  • Presets – Named sets of variable values for quick scenario switching

• A sandboxed runtime that securely renders documents.

• A set of examples that your LLM can understand and remix for your scenario.

• A VS Code extension for editing, previewing, and exporting documents.

• A web-based viewer and editor for viewing, quick edits and sharing.

• Tools to export standalone HTML documents you can share or embed anywhere.

Chartifact documents can be written in two formats:

• Markdown – Human-readable, easy to write and review. Interactive elements are embedded as fenced JSON blocks.
• JSON – Structured and precise. Ideal for programmatic generation or when working directly with the document schema.

Both formats are functionally equivalent and supported across all tooling.

The format is designed with AI assistance in mind:

• Structured syntax makes documents easy to edit and generate with LLMs
• In-editor tools like Ctrl+I and agent mode available in VS Code
• HTML exports retain semantic structure for downstream AI tools

This enables both authoring and remixing workflows with language models and agent-based tooling.

The document runtime is reactive. Components stay in sync through a shared set of variables:

• Reactive variables update elements and data sources automatically
• Dynamic bindings let variables appear in chart specs, text, URLs, and API calls
• REST integration supports fetching data from external sources
• Vega transforms provide built-in tools for reshaping data
• Signal bus coordinates state across all components

Styling is done using standard CSS. Examples provided to style documents as articles, dashboards, or slides.

Chartifact is designed to be safe by default:

• Rendered in sandboxed iframes to isolate execution
• No custom JavaScript execution
• No raw HTML in Markdown
• XSS-Defensive CSS parsing