Skip to content
JoseEstevez520Public

About

A compact UI language that lets AI agents turn intent into standalone HTML interfaces.

Topics

Resources

Stars

3 stars

Watchers

0 watching

Forks

Latest commit

 

History

24 Commits

Folders and files

Repository files navigation

Describe the interface, not the HTML

MIT license UIDL/1 Standalone HTML output Up to 76 percent token savings

AI agents describe intent. A renderer builds the interface.

A2TL-Web turns a compact UI specification into a complete standalone HTML page—charts, metrics, cards, tables, and theming included. Agents express what the page should communicate; the renderer owns the implementation details.

Relationship to SkillNet

A2TL-Web was built and evaluated as part of the SkillNet Generative UI research. It helped test compact, agent-authored interface descriptions and remains a useful reference for that line of work. SkillNet's current generated-interface path uses OpenUI, so A2TL-Web is related research rather than a current SkillNet dependency.

Compact by design Renderer-owned polish Ready for agents
A dashboard can use ~360 tokens instead of ~1,471 in raw HTML. No generated CSS or hand-assembled component boilerplate. Use it from the CLI, MCP server, or agent skill.

See the format

UIDL/1
theme dark

h1 "Q3 Revenue"
metrics 2
  "MRR" "$84.5k" green "+12%"
  "Churn" "2.1%" red "-0.4pp"
chart line "Growth"
  x "Apr" "May" "Jun" "Jul"
  series "Revenue" 62 71 78 84 blue

This ~360-token description renders to a standalone page with charts, metrics, and theming. The equivalent raw HTML typically needs ~1,471 tokens.

Quick start

npx a2tl-web render examples/analytics.uidl -o output/analytics.html

Or run the repository directly:

git clone https://github.com/JoseEstevez520/a2tl-web.git
cd a2tl-web
node tools/generate.js examples/analytics.uidl output/analytics.html

Use it your way

Interface Best for Start here
CLI Generating HTML from a .uidl file. a2tl-web render input.uidl -o output.html
MCP server Letting an AI agent call render_page. cd tools/mcp && npm install && npm run build
Agent skill Teaching the format directly to an agent. Copy skill/uidl.md into its prompt system.

Components

Component Syntax Use it for
Headings h1, h2, h3 Clear page hierarchy.
Text text "..." style Supporting copy and insights.
Metrics metrics N KPI cards and trends.
Charts chart line|bar|pie "Title" Series data and comparison.
Tables table "Title" Structured rows and columns.
Cards / lists cards N, list "Title" Grouped content.
Code / collapse code lang, collapse "Title" Technical detail and progressive disclosure.

Theming without generated CSS

Branding lives in a preset, so the agent does not need to produce CSS:

UIDL/1
theme dark
brand skillnet
a2tl-web render input.uidl --brand skillnet

Built-in presets include default and skillnet; custom themes live in tools/mcp/themes/. Read the extension guide for details.

Format rules

  1. Begin with UIDL/1.
  2. Put components at column 0 and sub-items two spaces in.
  3. Wrap strings containing spaces in double quotes.
  4. Use named or hexadecimal colors.
  5. Use blank lines to close blocks and // for comments.

Project map

  • skill/ — prompt material for agents learning UIDL.
  • tools/ — parser, renderer, CLI, MCP server, and themes.
  • examples/ — sample UIDL specifications.
  • tests/ — renderer and parser tests.
  • docs/extending.md — component and theming reference.

Development

npm test

License

MIT

About

A compact UI language that lets AI agents turn intent into standalone HTML interfaces.

Topics

Resources

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages