Skip to content

ubrandon/design-core

Repository files navigation

Design Core

Build interactive prototypes by chatting with AI. No coding needed.

Design Core is a design tool for teams. You describe what you want, AI builds it, and you share it with a link. It works alongside Figma, not instead of it: keep Figma for what it's great at, use this for fast AI-powered prototyping.


Get Started

Setup takes about 10 minutes and the AI does most of it:

Get Started


Pick your editor

Design Core is a normal folder of files plus an AI that edits them, so it works in any of these. Same repo, same rules, same result:

Tool Best for What it costs
Cursor The original Design Core experience Free Hobby tier to try it; Pro (~$20/mo) for real use. Auto mode gives near-unlimited fast usage
VSCode + Claude Code extension or GitHub Copilot People who already live in VSCode A Claude subscription or Copilot plan
Claude desktop app (Mac/Windows) Non-developers: no editor, no terminal Claude Pro or Max subscription

Each person needs their own account for whichever tool they pick. There is no separate Design Core license: the tool itself is free and open source.

Who is it for?

  • Design teams who want to prototype ideas quickly without waiting for developers
  • Product managers who want to explore UI concepts and share them with stakeholders
  • Anyone who can describe what they want in words -- the AI handles the code

What can you build?

  • Screens -- static UI mockups for exploring layouts, styles, and visual direction
  • Prototypes -- fully interactive mini-apps with working forms, animations, navigation, and real behavior
  • Design systems -- shared component libraries that keep everything consistent

Prototypes are fully functional, not link-navigation between mocked screens. When you build a prototype, state is real: selecting people updates chips/counts/CTAs, toggling members updates "X of Y" labels, clearing resets state, etc. Static screens belong on the canvas — prototypes simulate the actual feature behavior. If you only need clickthroughs between fixed mocks, that's a canvas job, not a prototype.

Example prompts

"Build a signup flow with email, password, and confirm. Show inline validation and a success animation."

"Create a settings page with toggles for notifications, dark mode, and location sharing."

"Make a multi-step onboarding: welcome screen, pick interests, set profile photo, done."


How a team uses it

  1. One person sets it up (Get Started): a private company repo on GitHub, created from this public template.
  2. Designers join by pasting one prompt; the AI clones the repo and walks each person through a 2-minute setup.
  3. Everyone works locally at http://localhost:3000, and nobody needs a terminal: designers ask the AI to start the tool, save and share work, or grab the latest from teammates.
  4. Stakeholders just get links: pushes deploy to GitHub Pages automatically.
  5. Tool updates flow from this repo: any teammate asks the AI to update the tool, then pushes. Designs and the tool never fight, because design work lives entirely under public/data/.
  6. Something looks broken? Ask the AI to run the health check; it explains the problem and the fix in plain language.

Why not other AI design tools?

Figma Make, Google Stitch, and similar tools are expensive, slow, or produce mediocre results. Design Core is free and open source; the only cost is the AI subscription you probably already have. You also keep everything: it's your repo, your files, plain HTML/CSS that any developer can read.

Features

  • AI-powered -- describe what you want, get working HTML/CSS/JS
  • Interactive prototypes -- not just mockups, real working UI with state, validation, transitions
  • Infinite canvas -- arrange and explore static screen concepts spatially
  • Design system -- shared tokens, components, and styles across all projects
  • Shareable links -- push to Git and prototypes deploy to GitHub Pages automatically
  • Team-friendly -- each designer gets their own identity, projects track who made what

Workspaces

Workspace What it does
Home Project list -- see all your team's work
Project Hub Jump to canvas or prototypes for a project
Canvas Infinite canvas for arranging static screen ideation
Prototypes Interactive HTML/CSS/JS mini-apps built with AI
Design System Global component reference with tokens and styles

Technical Details

Built with vanilla HTML/CSS/JS + Vite. No framework, no database, no backend. The dev server runs on port 3000 (fixed, so links always match).

File structure

Vite serves public/ at the site root. On disk, design data lives under public/data/; in the browser, paths start with data/ (no public/ prefix).

public/
  styles/                 # Tool + design tokens (shared.css, ds.css, app.css, …)
  data/
    site.json             Public site URL for prototype "Copy link" (written by finish-setup)
    projects/
      index.json
      <project-id>/
        project.json
        canvas.json
        screens/            Static HTML (no JS) for canvas ideation
        prototypes/
          index.json
          <prototype-id>/
            meta.json
            index.html      Interactive prototype (HTML/CSS/JS)
    design-system/
      registry.json         Ships empty; company fills with their groups/categories
      company.css           (Company repos) Brand overrides -- auto-loaded after shared.css/ds.css
      components/           Component HTML snippets
        <company-slug>/     Company repos put components here (e.g. acme/)
    captures/             Capture config + manifest (see docs/captures.md)

Root HTML pages (index.html, canvas.html, …) sit at the repo root; browser JS lives in public/scripts/ (so it ships with vite build / GitHub Pages). Node CLIs (doctor.js, update-tool.js, …) stay in repo-root scripts/. See docs/DESIGN_TOOL_PLAN.md for the full tree.

Commands

Designers never need these directly: they ask the AI ("start the tool", "update the tool", "run the doctor") and the AI runs them. For reference:

npm run dev                  # start the tool at http://localhost:3000
npm run doctor               # health check with plain-language fixes
npm run update-tool          # (company repos) pull the latest tool improvements
npm run finish-setup         # (company repos) one-time setup finalizer
npm run sync-public-url      # re-sync the public share-link URL
npm run build                # production build (CI runs this for GitHub Pages)

Sharing

Push to main and prototypes deploy to GitHub Pages automatically (enable once: Settings → Pages → Source → GitHub Actions).

Designers run the tool on localhost; stakeholders open Copy link URLs in a normal browser. Those links point at the deployed GitHub Pages site. npm run finish-setup configures this during company setup; run npm run sync-public-url again if the repo ever moves, or set the URL manually in public/data/site.json:

{ "publicBaseUrl": "https://your-org.github.io/your-repo/" }

More docs


License

MIT

About

Build interactive prototypes by chatting with AI. No coding needed. Free and open source. Works in Cursor, VSCode, or the Claude desktop app.

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors