deCDN — marketing site

A decentralized delivery layer for bytes at scale. Anyone can serve bytes; clients pay per megabyte in USDC.

This is the source for the deCDN marketing site and blog.

Stack

Next.js 16 (App Router) · React 19 · Tailwind CSS v4 · TypeScript · pnpm · Node ≥24 (.nvmrc). Static export, deployed to Cloudflare Pages.

Getting started

pnpm install      # required (not npm/yarn) — husky hooks shell out to `pnpm exec`
pnpm dev          # dev server on :3000
pnpm build        # static export → ./out, then the check:out + check:og-image guards
pnpm test         # vitest run
pnpm typecheck    # tsc --noEmit
pnpm lint         # eslint (flat config)
pnpm format       # prettier --write .
pnpm format:check # prettier --check . (CI)

Project layout

• app/ — App Router entry (layout.tsx, page.tsx, globals.css) plus the blog routes (blog/, blog/[slug]/), the legal/[doc] route, the sitemap/robots handlers, the llms.txt / llms-full.txt handlers, and file-convention metadata assets.
• components/layout/ — the site chrome mounted in app/layout.tsx on every route (Chrome → MobileMenu, Footer, ScrollReveal) plus chrome-active, the nav's section and tone tables.
• components/home/ — the homepage sections composed by app/page.tsx, one folder per section (hero/, compare/, method/, faq/, try/, contact/) with its sub-parts, plus brand.tsx, which they share.
• components/blog/, components/legal/ — components used only by those routes (PostRow, LegalDoc).
• components/ui/ — generic primitives with no site content (Frame, SectionHeader, Prose, Figure, PanelFigure, Pill, CodeBlock). A component that takes site data or domain types goes in its feature's folder; a primitive with no site content stays in ui/ even if only one feature uses it today.
• lib/ — shared helpers (links.ts, copy.ts, stats.ts, blog.ts, faq.ts, legal.ts, jsonld.tsx, schema.ts, …). The site's prose and its displayed values live in lib/copy.ts as plain data — the comparison table maps over it and the llms-full.txt mirror serialises it, so anything inlined in a component drifts from the mirror. Prose that names the product renders through highlightBrand (components/home/brand.tsx).
• test-utils/ — test-only helpers for walking the element tree a server component returns.
• scripts/ — postbuild guards run against ./out by pnpm build: check-out.mjs (advertised URLs, dotted routes, panel captions, the prerendered mobile menu toggle) and check-og-image.mjs (social cards).
• content/blog/ — MDX posts loaded by lib/blog.ts and rendered by app/blog/[slug]/page.tsx.
• content/legal/ — MDX for the legal pages (privacy, terms, disclaimer), loaded by lib/legal.ts and rendered by app/legal/[doc]/page.tsx.
• docs/ — Mintlify source for docs.decdn.org. Built and deployed independently of pnpm build; not part of the static export and not imported from the website code.
• Path alias: @/* maps to the project root (e.g. @/lib/links, not @/src/...).

Gotchas

• Static export only. next.config.ts sets output: "export"; the build emits ./out. No SSR, ISR, middleware, or Image Optimization API. Route handlers (app/robots.txt/route.ts, app/sitemap.xml/route.ts, app/sitemap-pages.xml/route.ts, app/llms.txt/route.ts, app/llms-full.txt/route.ts) must be dynamic = "force-static" and GET-only; the dotted segment is what makes them emit real files rather than <path>/index.html. robots.txt is a hand-written route handler rather than the Next app/robots.ts metadata file, so it can emit the non-standard Content-Signal: directive (per contentsignals.org).
• trailingSlash: true. Every route emits <path>/index.html. Internal links and hand-built URLs (sitemap entries, JSON-LD @ids) should expect a trailing slash — see SITE_URL and BLOG_URL in lib/links.ts.
• Tailwind v4. Theme tokens live in app/globals.css under @theme inline { … } — there is no tailwind.config.*.
• Conventional commits enforced. commitlint runs in the commit-msg husky hook; non-conforming messages are rejected.
• metadataBase is live. lib/links.ts site is the real origin and INDEXABLE is true. Anything anchored on this origin — OG and canonical (via metadataBase); JSON-LD and the sitemap/robots emitters (via SITE_URL) — ships to production. Adding a non-blog page means appending an entry to app/sitemap-pages.xml/route.ts; blog posts are auto-derived from content/blog/, while legal pages are driven by the closed LEGAL_SLUGS list in lib/legal.ts (add both the MDX file under content/legal/ and a slug there). Flipping INDEXABLE flips <meta name="robots"> only; robots.txt and the sitemap stay unconditional by design (rationale in lib/links.ts).
• Machine-readable surfaces. /llms.txt is a curated index of this origin and /llms-full.txt is every page as one plain-text document; both derive their page entries from listPosts(), , and , are listed in , and are advertised through in and a header in . resolves every advertised same-origin URL against after a build. The protocol documentation has its own pair on , served by Mintlify.

Testing

This project is tested with BrowserStack. Unit tests run under vitest (pnpm test); cross-browser checks use BrowserStack through its Open Source Sponsorship Program.

Deploy

Cloudflare Pages serves the out/ directory produced by pnpm build.

Agent contributors

See AGENTS.md for the full ruleset (CLAUDE.md is a one-line include of it).