Cloudflare Workers

A serverless function normally means a container somewhere that boots when a request arrives. Cloudflare Workers doesn’t do that: your code runs in a V8 isolate inside Cloudflare’s edge network, in whichever city the request hit, with effectively no cold start.

The free tier is one of the last genuinely generous ones on the internet: 100,000 requests a day, unlimited bandwidth, and it hasn’t moved in years.

Write and deploy one link

  npm create cloudflare@latest my-worker
cd my-worker
npx wrangler dev            # local, with the real runtime
npx wrangler deploy
  

  export default {
  async fetch(request, env, ctx) {
    const url = new URL(request.url);
    if (url.pathname === "/api/time") {
      return Response.json({ now: Date.now(), city: request.cf?.city });
    }
    return new Response("Hello from the edge");
  },
};
  

Deploy takes seconds and the result is live in every Cloudflare location. request.cf carries geo information you’d otherwise need a lookup service for.

The pieces that come with it link

Bindings connect them without credentials — you declare them in wrangler.toml and they arrive as env:

  name = "my-worker"
main = "src/index.js"
compatibility_date = "2026-08-01"

[[kv_namespaces]]
binding = "CACHE"
id = "..."

[[d1_databases]]
binding = "DB"
database_name = "app"
database_id = "..."
  

  await env.CACHE.put("key", "value", { expirationTtl: 300 });
const { results } = await env.DB.prepare("SELECT * FROM items WHERE id = ?").bind(id).all();
  

No connection string in an environment variable, no secret to rotate. For a small app, that alone removes a category of work.

The constraint that decides everything link

Workers is not Node. It’s a V8 isolate with web APIs, and although the Node compatibility layer covers a lot now, anything needing raw TCP sockets, native modules, long CPU time, or the filesystem does not belong here. CPU time per request is limited — generous on paid plans, 10ms on free — while time spent waiting on I/O is not.

So: request handling, API routes, auth checks, redirects, image transforms, streaming proxies, and glue between services — excellent. Video encoding, a long report build, or anything holding a Postgres connection pool — put it on Railway or Fly.io and call it from here.

If you do talk to Postgres, use a driver that speaks HTTP — Neon ships one for exactly this reason — or Hyperdrive, which pools connections on Cloudflare’s side.

Pages, for the frontend half link

Connect a repository and Pages builds and serves it, with a preview URL per branch — the same shape as Vercel, with no commercial-use restriction and unlimited bandwidth. Next.js and other frameworks run through the Workers runtime, so the same constraint above applies to their server code.

Next link

Every app needs somewhere to keep data → PostgreSQL