The £8/Month Server Running My Entire Business

A single Hetzner VPS runs Spellcast, Postiz, Postgres, Redis, Temporal, and Caddy. Here's what's on it, how it's structured, and why it costs less than a lunch.

4 min readdevopsindie-hackinginfrastructuredocker

Everything I run lives on a single server. One Hetzner VPS in Helsinki, £8/month.

It hosts Spellcast (my social scheduling platform), Postiz (the underlying publishing engine), PostgreSQL, Redis, Temporal (workflow engine), and Caddy for HTTPS. Six services. One box. Handles hundreds of social posts a week.

Here's exactly what's running and why.

The server

Hetzner CX22: 2 vCPUs, 4 GB RAM, 40 GB NVMe, located in Helsinki. £8.29/month at current prices.

The choice of Hetzner over AWS or DigitalOcean is mostly cost. An equivalent AWS EC2 instance (t3.medium, 2 vCPU, 4 GB) is around £25-30/month plus storage and bandwidth. Hetzner is straightforwardly cheaper for the same hardware.

The trade-off: no managed services, no auto-scaling, no load balancing. For a solo product in the early growth stage, that's fine. The complexity of managed cloud doesn't pay for itself until you actually need it.

What's running

Everything runs in Docker. The full docker-compose.yml manages:

services:
  postgres:
    image: postgres:16
    volumes:
      - postgres-data:/var/lib/postgresql/data
 
  redis:
    image: redis:7-alpine
 
  postiz:
    image: spellcast-postiz:latest
    ports:
      - "5000:5000"
      - "4200:4200"
 
  temporal:
    image: temporalio/auto-setup:latest
    ports:
      - "7233:7233"
 
  caddy:
    image: caddy:2-alpine
    ports:
      - "80:80"
      - "443:443"
    volumes:
      - caddy-data:/data
      - ./Caddyfile:/etc/caddy/Caddyfile

The Spellcast API container runs alongside these but is managed separately (built from source, redeployed on changes).

What each service does

Postiz handles the actual social media publishing: OAuth integrations, the publishing queue, and direct API calls to Twitter, LinkedIn, Instagram, and the rest. It's an open-source scheduling engine that I've extended with custom features.

Temporal is the workflow engine. It handles long-running processes reliably: if a post fails to publish, Temporal retries it with backoff. If the server restarts mid-workflow, it picks up where it left off. This is overkill for small volume but it means I've never had a post silently fail to go out.

Caddy handles HTTPS automatically. Zero config: point a domain at the server, write a two-line Caddyfile, done. No Let's Encrypt certificate management, no nginx config.

Postgres stores everything: users, posts, schedules, account sets, brand voices, analytics. One database, two schemas (spellcast and postiz).

Redis handles the publishing queue, session caching, and pub/sub between Postiz services.

Networking

All services are on the same Docker network (spellcast_default). They communicate by service name over the internal network. Nothing except Caddy and the publishing engine ports are exposed to the internet.

The Spellcast API reaches Postiz at http://postiz:5000 internally. External traffic hits Caddy, which reverse-proxies to the right service. This means only ports 80 and 443 are open to the world.

Deployments

Deploying a change to the Spellcast API:

cd /opt/spellcast
git pull
docker build -f docker/api/Dockerfile -t spellcast-api:latest --platform linux/amd64 .
docker stop spellcast-api && docker rm spellcast-api
docker run -d --name spellcast-api --restart unless-stopped \
  --expose 3000 \
  --env-file docker/.env \
  --network spellcast_default \
  spellcast-api:latest

About 3-4 minutes from git pull to live. Not zero-downtime, but acceptable for a product at this stage. The Next.js frontend deploys to Vercel separately (Vercel handles that side; the VPS handles backend only).

What it costs to run

| Service | Cost | |---------|------| | Hetzner CX22 | £8.29/month | | Cloudflare R2 (media/podcast storage) | ~£0.50/month | | Vercel (frontend + API BFF) | £0 (hobby tier) | | Domain (sammii.dev, lunary.app, etc.) | ~£2/month amortised | | Total | ~£11/month |

The full infrastructure for five active projects costs less than a Spotify subscription.

When this stops working

Single-server setups have real limits. The current one breaks if:

  • Traffic exceeds what 4 GB RAM can handle (Postgres, Redis, and Postiz are the hungry ones)
  • The server goes down (no redundancy)
  • A single noisy process starves the others

None of these are currently problems. When they are, the answer is probably a second server (one for Postgres/Redis, one for app containers) rather than a jump to managed cloud. That keeps costs under £25/month and solves the resource contention issue.

The principle: run on the smallest infrastructure that actually works, and move up only when you have a real reason. Premature scaling is one of the more expensive ways to spend money before you have revenue.