Moving 8plus to Cloudflare Workers

In February, Cloudflare announced vinext: the Next.js API reimplemented on Vite, deployable to Workers with one command. It started as an "AI rebuilt Next.js in a week" experiment and has since reached beta. Frameworks decoupling from their hosting platforms is one of the front-end trends I'm watching most closely this year.

Rather than just read about it, I tried it on my own site.

Principle: run in parallel, don't replace

• Keep the Next.js source and build as they are, and add a separate vite build for Workers.
• Vercel stays as the rollback origin; DNS is not touched first.
• Preview on *.workers.dev, pass the smoke and SEO tests, and only then attach www.8plus.app/* as a Worker Route.

What the inventory showed

This site is an App Router + Velite content site: no Server Actions, ISR or API Routes, and no Vercel-only services. That is exactly the kind of site where vinext is easiest to adopt. The one visible difference is that React's ViewTransition is currently a passthrough, so page transitions don't animate.

Three gotchas

1. Date.now() is 0 during module initialisation. On Workers the clock doesn't advance while modules load. I filtered published posts at module top level, so every post looked future-dated and returned 404. Computing it per request fixed it, with no change to Next.js behaviour.
2. Rewrites to static files run at a different phase. vinext checks public/ before afterFiles rewrites, which broke /tool/qrcode → qrcode.html. Moving it to beforeFiles solved it.
3. Toolchain types collide. The Next.js type check picked up the Vite 8 config files and failed, so vite.config.ts and cloudflare.config.ts are excluded in tsconfig.

All three were small changes, but without page-by-page testing the first one would have made the whole blog quietly disappear.

An existing bug found along the way

One of the best side effects of a migration is being forced to look at the whole site again. The production robots.txt and sitemap.xml were pointing at localhost, because the deployment was missing the site-URL variable. Unrelated to the move, but an SEO gap that had gone unnoticed for a long time.

Status and next steps

• www.8plus.app is now served by the Worker, and all 29 smoke checks plus the SEO tests pass.
• The apex domain still redirects via Vercel; once things are stable I'll switch DNS and rehearse the rollback.

If you're thinking about leaving a platform or cutting hosting costs, my advice is to build a feature inventory first and pick the framework second. To talk through your situation, book a 30-minute intro.