quaedra.com

The Quaedra Research site, Nodd included: a React + Vite multi-page app served by one Cloudflare Worker (quaedra-www).

• pages/: one HTML entry per page, at its URL's path (pages/jet/docs/index.html → /jet/docs/), with the page's meta tags
• src/pages/: the Quaedra pages; src/routes.tsx maps each entry's <body data-page> to one. The build prerenders them to HTML and React attaches in the browser
• src/nodd/: Nodd's pages (demo, train, docs, models), rendered in the browser
• src/docs/: docs pages built from the Jet and megacode repositories (npm run docs)
• public/: static files; public/nodd/{models,ort,training} are copied from the Nodd repository and not committed
• worker/index.js: redirects old Nodd URLs and reassembles files split for the 25 MiB asset limit

Nodd's runtime (@nodd/browser, @nodd/core) builds from source in ../nodd. Nodd's models, onnxruntime files and training base model aren't in git: their published copy is the Hugging Face dataset quaedra/nodd-site.

Every push to main deploys: .github/workflows/deploy.yml checks out quaedra/nodd beside the site, pulls Nodd's files from Hugging Face, type-checks, builds and runs wrangler deploy.

npm install
npm run nodd:pull   # download Nodd's files from quaedra/nodd-site into public/nodd/
npm run sync-nodd   # or copy them from ../nodd (run npm run sync-model there first)…
npm run nodd:push   # …then publish them to quaedra/nodd-site for the GitHub deploy
npm run dev         # http://localhost:5173, clean URLs as in production
npm run typecheck
npm run build       # → dist/
npm run preview     # build, then serve dist/ with the worker (wrangler dev)
npm run deploy      # build and deploy from this machine (pushing to main does this too)
npm run docs        # rebuild the Jet and megacode docs from ../jet and ../megacode
npm run og          # re-render the Open Graph cards in public/og/