Cloudflare Workers
& Pages.

Each Worker runs in a V8 isolate — the same primitive a Chrome tab uses. An isolate spins up in roughly a millisecond, shares a single V8 process with thousands of peers, and has no filesystem, no Node.js, no native modules. That's how you get sub-millisecond cold starts and per-request pricing. It's also why you can't `require('sharp')` or shell out to ImageMagick.

// Every request gets a fresh isolate context.
// No container. No VM. No cold start.
export default {
  async fetch(req, env, ctx) {
    const { colo } = req.cf;
    return new Response(`hello from ${colo}`);
  }
};

Pages compiles your site from a Git repo and serves the output from every edge location. Drop a file under functions/ and that route becomes a Worker on the same deploy, with automatic preview URLs per pull request. For the common case — static site with a handful of dynamic endpoints — you skip most of the serverless config ceremony entirely.

// functions/api/hello.ts — becomes /api/hello
export const onRequest: PagesFunction = async (ctx) => {
  const body = { ok: true, city: ctx.request.cf.city };
  return Response.json(body);
};

Workers doesn't give your code an SDK for KV or D1 — it gives it a binding. You declare the resources you need in wrangler.toml and they arrive as properties on the env argument. No credentials to rotate, no connection strings, no client libraries to keep in sync. The primitives you get are KV (eventual key-value), D1 (SQLite), R2 (S3-compatible object storage with zero egress), Durable Objects (stateful, single-writer), and Queues.

# wrangler.toml
name = "edge-api"
main = "src/index.ts"

[[kv_namespaces]]
binding = "CACHE"
id = "a1b2..."

[[d1_databases]]
binding = "DB"
database_name = "prod"

[[r2_buckets]]
binding = "ASSETS"
bucket_name = "uploads"