Building a Self-Hosted Search Engine for Under €30/Month
You don't need Google's infrastructure to build a useful search engine. You need smart architecture and €30 per month.
The Stack
| Component | Tool | Cost |
|---|---|---|
| VPS | Hetzner CPX21 (4 vCPU / 8GB) | €15/month |
| Meta-search | SearXNG + custom aggregators | Free |
| API queries | Bing Search API | ~€7/month |
| Caching | Redis | Free |
| Database | PostgreSQL | Free |
| Frontend | Next.js on same VPS | Free |
| Total | €22/month |
Architecture Decisions
Why Meta-Search Over Crawling
Crawling the entire web requires petabytes of storage and massive compute. Meta-search queries existing indexes (Bing, Brave, Mojeek, Wikipedia, GitHub) and ranks results locally. Coverage is 95%+ of what users actually need.
Why Hetzner
German data centers. EU jurisdiction. DSGVO-compliant by default. €15/month gets you more compute than AWS €50/month. No egress charges, no surprise bills.
Scaling Path
At 1,000 daily users: €22/month handles it.
At 10,000 daily users: Add a second VPS (€15) + load balancer (€5) = €42/month.
At 100,000 daily users: Then you think about Kubernetes. Not before.
The Technical Build
Backend: Python FastAPI handles query dispatch, result ranking, and caching. Each query hits 8-12 sources in parallel. Response time: 300-800ms.
Frontend: Next.js with server-side rendering. The server makes API calls, not the client. User IP never touches external search sources.
Privacy layer: Query text discarded after 30 seconds. No persistent logs. Aggregate analytics only.
Graham Miranda builds privacy-first search and automation tools at Graham Miranda UG (Berlin, HRB 36794).
Top comments (0)