DEV Community

Luna
Luna

Posted on • Originally published at builderlog.net

I Built an AI Publishing Stack for $31/Month — Real Numbers After 15 Posts

I've been building a solo content operation in public for the past month. The goal: run a real publishing workflow — blog posts, analytics, affiliate tracking — entirely on AI automation, and document whether it can pay for itself.

The full stack costs $31/month

  • Claude API (~$20): content drafting, topic generation, quality checks
  • Hetzner CX21 VPS (~$6): self-hosted n8n for workflow automation
  • .dev domain (~$1): annualized
  • GitHub Pro (~$4): private repos + Actions minutes

What's actually running on this

  • Auto-publishes 1 post/day (EN + KO) from a topic queue
  • Collects daily view counts via Cloudflare KV
  • Tracks affiliate link clicks with a /go/ redirect layer
  • Sends Telegram alerts when something breaks

15 posts live so far. Every post has real numbers — actual traffic, actual revenue, actual mistakes. No "lessons learned" without the receipts.

The parts people ask about most

n8n self-hosted: runs on the Hetzner VPS for free. Handles all scheduling, webhook routing, and API calls. Add a 2GB swap file early — n8n can spike on heavy loads.

Prompt caching: this one change cut Claude API costs by ~60%. Caching the system prompt and batching overnight runs brought spend from ~$40 to ~$20/month.

Cloudflare KV for analytics: Workers increment a view counter on each page load, a nightly cron rolls up daily totals. Free tier covers everything under ~100k reads/day — no separate analytics database needed.

Current status

The stack works. Whether it earns back the $31 is the question I'm documenting in real time.

Full breakdown with every line item: builderlog.net

Happy to go deep on n8n setup, Claude API cost optimization, or the Cloudflare Workers + KV pattern.

Top comments (0)