Six months ago I was paying DigitalOcean, Heroku, and a managed Postgres provider a combined $100/month for a side project that made exactly $0. I killed all of it, moved to a 100% free stack, and my uptime monitor has shown zero downtime in 6 months. Here's the exact setup.
The Free Stack (Production-Verified)
| Layer | Before (Paid) | After (Free) | Free Tier Limit |
|---|---|---|---|
| Frontend | Netlify Pro ($19/mo) | Cloudflare Pages | Unlimited bandwidth, 500 builds/mo |
| Backend API | Heroku Eco ($5/mo + sleeping) | Cloudflare Workers | 100,000 requests/day |
| Database | Managed Postgres ($15/mo) | Neon | 0.5 GB storage, autoscaling to zero |
| Auth | Auth0 ($23/mo) | Supabase Auth | 50,000 MAU |
| File storage | S3 + CloudFront (~$8/mo) | Cloudflare R2 | 10 GB, zero egress fees |
| Cron jobs | VPS cron ($5/mo) | Cloudflare Cron Triggers | Free with Workers |
Total before: ~$100/month. Total after: $0/month. That's $600 saved so far.
The Numbers That Matter
Real traffic on this project last month: 41,000 requests, 2.1 GB bandwidth, ~180 MB database.
- Cloudflare Workers: 41k/3M monthly requests = 1.4% of free quota
- Neon: 180MB/500MB = 36% of free storage, hibernates when idle
- Supabase Auth: 340 MAU / 50,000 = 0.7% of quota
- R2: 800MB/10GB = 8%
Cold start on Neon after hibernation: ~300ms on first query. Cloudflare Workers: no cold starts at all (V8 isolates, not containers). That's the dirty secret people don't tell you about Lambda — 1-3s cold starts vs Workers' effectively zero.
Deploy a Worker in 4 Commands
npm create cloudflare@latest my-api
cd my-api
npx wrangler secret put DATABASE_URL # Neon connection string
npx wrangler deploy
Query Neon from the Worker:
import { neon } from '@neondatabase/serverless';
export default {
async fetch(request, env) {
const sql = neon(env.DATABASE_URL);
const rows = await sql`SELECT * FROM posts LIMIT 10`;
return Response.json(rows);
}
}
What Actually Broke (Being Honest)
-
Neon's 0.5 GB fills faster than you think — my
logstable ate 300MB in two months. Fix: partition + auto-delete after 7 days. - Supabase free tier pauses after 7 days of inactivity — I only use it for auth (constant activity), DB is on Neon. If your project is low-traffic, this pause is a real gotcha.
- No long-running processes — Workers cap at 30s CPU on free tier. My image-resize job moved to a GitHub Actions cron instead (2,000 free minutes/month).
The Controversial Part
Railway, Render, and Fly.io free tiers are effectively dead or crippled — Railway killed theirs in 2023, Render's free web services sleep and have bandwidth caps, Fly removed the free allowance for new users. Yet I still see "deploy your side project on Railway/Render free tier" in every 2026 tutorial. The Cloudflare + Neon + Supabase triangle is the actual free stack now, and it's not close.
The one thing you genuinely can't get free: a custom domain email. Everything else? There's a real, production-grade free tier for it.
I write about free dev infrastructure weekly. More resources: https://ly.cyberserval.tech/iIETXiF
What's in your free stack — and what finally made you pay for something? For me it was a domain ($12/year). Curious where others draw the line.
Top comments (0)