designmd.supply

A supply of design md, generated. Drop in any public domain — get back a Google-spec DESIGN.md ready to feed an AI agent.

designmd.supply is a small Next.js app that turns any website into a self-contained design system document. It pulls brand identity, design tokens, a hero screenshot, and the live page Markdown, then composes them into a DESIGN.md file with YAML frontmatter and the canonical sections (Overview, Colors, Typography, Layout, Elevation, Shapes, Components, Do's and Don'ts).

Powered by Context.dev

The whole pipeline runs on four endpoints from the Context.dev API:

Context.dev API	What we use it for
Styleguide API — web.extractStyleguide	Primary source of truth for design tokens: colors, typography, spacing, radii, components.
Brand API — brand.retrieve	Logos, backdrops, brand colors, slogan, industry classification — used to render the directory and the canvas hero.
Screenshot API — web.screenshot	High-quality homepage capture, fed to the model as visual context for tone and layout.
Markdown API — web.webScrapeMd	Clean Markdown of the live page so the model can ground component and copy guidance in real content.

The four payloads converge in lib/design-md.ts, where they're shaped into a single prompt and sent through the Vercel AI Gateway to produce the final DESIGN.md.

Get a free Context.dev key at context.dev.

Quickstart

npm install
cp .env.example .env
npm run dev

Then open http://localhost:3000 and search for any domain.

Environment variables

CONTEXT_DEV_API_KEY=             # https://context.dev
VERCEL_AI_GATEWAY_API_KEY=       # https://vercel.com/ai-gateway

# Turso (Vercel Marketplace integration — auto-injected on Vercel)
designmd_TURSO_DATABASE_URL=
designmd_TURSO_AUTH_TOKEN=

The designmd_* prefix matches the variables provisioned by the Turso integration on the Vercel Marketplace, so on Vercel they're injected automatically. Locally, copy them from the project's Vercel dashboard into .env. If they're absent, the app still works — it just falls through to the live Context.dev APIs on every request with no cache.

How it works

            ┌──────────────────────────────────────────┐
            │             Domain in (any URL)          │
            └──────────────────────────────────────────┘
                                  │
                                  ▼
                       Turso cache hit?  ──► serve cached payload
                                  │
                                  ▼ (miss)
       ┌──────────────┬───────────┴──────────┬──────────────┐
       ▼              ▼                      ▼              ▼
  Styleguide API   Brand API           Screenshot API   Markdown API
  (tokens)         (logos, colors)     (hero image)     (page text)
       │              │                      │              │
       └──────────────┴──────────┬───────────┴──────────────┘
                                 ▼
                  Prompt assembly  (lib/design-md.ts)
                                 ▼
                   LLM via Vercel AI Gateway
                                 ▼
                    DESIGN.md  (YAML + Markdown)
                                 ▼
                       write back to Turso

The four Context.dev calls run in parallel. The model receives the styleguide JSON as the primary token source and the screenshot + page Markdown as supporting evidence for tone, components, and layout — so values come from real data rather than the model's imagination.

Project layout

app/
  api/
    brand/         GET  — Context.dev Brand API proxy (Turso-cached)
    styleguide/    GET  — Context.dev Styleguide API proxy (Turso-cached)
    screenshot/    GET  — Context.dev Screenshot API proxy (Turso-cached)
    design-md/     GET  — Turso cache lookup
                   POST — compose DESIGN.md from the three payloads + Markdown
  guides/[domain]/ route for an individual generated guide
  page.tsx         home with search + directory of curated domains

components/        UI: directory, search, canvas, orchestration, copy block
lib/
  design-md.ts     prompt + spec summary
  domains.ts       curated list of domains shown on the home page
  turso.ts         libSQL client + cache helpers (get/set/batch/listDomains)

Caching with Turso

The four Context.dev payloads (styleguide, brand, screenshot, markdown) and the final composed DESIGN.md are all cached in a single Turso table, domain_cache, keyed by (domain, kind). The schema is created lazily on first use — there's no migration step.

• First request for a domain hits Context.dev + the LLM and writes every payload back to Turso.
• Subsequent requests for the same domain serve straight from Turso.
• generateStaticParams for /guides/[domain] reads the set of domains with a cached designmd row from Turso, so the build only pre-renders pages that actually have content.
• Homepage directory uses a single batched read (getCachedBatch) to pull brand + screenshot for every curated domain in one round trip.

Cache writes are best-effort — if Turso is unreachable, the request still succeeds, the user just doesn't get a cached copy. The same applies if the env vars are missing entirely (see above).

To add a domain to the curated directory, append it to TOP_DOMAINS in lib/domains.ts. The first visit will populate the cache.

Scripts

npm run dev         # Next dev server
npm run build       # production build
npm run start       # serve the production build
npm run lint        # eslint
npm run typecheck   # tsc --noEmit

Tech stack

• Next.js 16 (App Router, RSC)
• React 19
• Tailwind CSS 4
• context.dev SDK
• Vercel AI SDK + AI Gateway
• Turso (libSQL) via the Vercel Marketplace integration

Contributing

Issues and PRs welcome. To add a domain to the curated directory, append it to TOP_DOMAINS in lib/domains.ts — the cache populates on first visit.

License

MIT