bulaev.ai

Cloudflare Workers, D1 and R2

CCCFreemium

Infra · Build

Edge compute, a SQLite database and S3-compatible storage under one CLI, with a generous free tier.

Open workers.cloudflare.com

What it is

Cloudflare Workers run JavaScript, TypeScript and Wasm on Cloudflare's edge, close to the visitor, with cold starts measured in milliseconds. Around the runtime sits a set of bindings you declare in wrangler.toml and use as plain objects in code: D1 (a serverless SQLite database with migrations and read replication), R2 (object storage with an S3-compatible API and no egress fees), KV, Queues, Durable Objects for stateful coordination, cron triggers, Workers AI, and Access for putting a login wall in front of anything. wrangler is the CLI for local dev, migrations, secrets and deploys.

The free plan covers a real amount of traffic and storage; paid usage is metered per request, per row read and per gigabyte stored. For a one-person or ten-person company this is the difference between running a product and running a Kubernetes cluster.

Where it fits in the AI OS

Build, the runtime for everything you ship: the website, the API worker, the Telegram bot, the cron that refreshes GitHub stars. Because every piece is a small deployable unit with typed bindings, it is also the environment agents are best at working in: one file, one config, one command to deploy.

Insider notes

members

How Serge's stack uses it

The membership site runs on it entirely: apps/web (Next.js via OpenNext) and apps/core (a Hono worker for membership logic) deploy as Workers, with D1 for the data model in the spec and R2 for media (an ideas-media bucket for clips, posters and screenshots, served through a CDN hostname with tier checks). Staging lives on stage.bulaev.ai through Worker custom domains. The live bulaev.ai site and Serge's earlier Telegram bot factory use the same Workers + D1 + R2 pattern, and Punctual, the open-source scheduler from his company, runs on it too.

Setup in 5 minutes

  1. npm create cloudflare@latest my-worker and pick the Hello World template; npx wrangler login.
  2. npx wrangler d1 create my-db, paste the binding into wrangler.toml, write migrations/0001_init.sql, run npx wrangler d1 migrations apply my-db --local.
  3. npx wrangler r2 bucket create my-media, add the R2 binding.
  4. In src/index.ts read from env.DB.prepare(...) and env.MEDIA.get(key); npx wrangler dev to test locally.
  5. npx wrangler deploy, then npx wrangler secret put API_KEY for anything sensitive.

Alternatives

  • Vercel + Neon/Turso + S3 — the other default for Next.js; more moving parts, more vendors.
  • Fly.io / Railway — containers with volumes when you need long processes or Postgres.
  • Deno Deploy / Supabase Edge Functions — edge functions with their own data stories.
  • AWS Lambda + DynamoDB + S3 — the enterprise version of the same idea, with the enterprise setup cost.