Cut your cloud bill 5×. Keep your uptime.
We move small SaaS teams off AWS to Hetzner bare-metal with Kubernetes, then run it for you. Fixed price for the migration, from $290 a month after.
typical result: $1,900/mo on AWS → $340/mo on Hetzner, same workload, same SLO
Built for teams without a DevOps hire
Solo and small SaaS.
Product built, infra duct-taped. You find out it's down from a customer.
Dev studios.
You ship frontend and backend; clients ask who runs it. We're your ops subcontractor.
Teams tired of the AWS bill.
You're paying for managed services you could run on three servers.
How a migration goes
- 01
Audit, 3 days.
Send your bill range and a list of services. You get a sized Hetzner plan with a monthly number. No call, no access needed.
- 02
Build, 1–2 weeks.
k3s cluster, GitOps with ArgoCD, Postgres with backups, monitoring, TLS. Runs next to your current setup.
- 03
Cutover, 1 day.
DNS switch with rollback ready. Your users don't notice.
- 04
Run, monthly.
We're on call, patch, upgrade, and answer in Slack within hours.
What you get
| Kubernetes (k3s) on Hetzner dedicated servers | Sized to your real load, not AWS defaults. |
|---|---|
| GitOps deploys | Push to main, ArgoCD rolls it out, rollbacks in one click. |
| Postgres with PITR backups | Daily full, WAL every 5 minutes, restore tested monthly. |
| Monitoring and alerts | Prometheus, Grafana, alerts to your Slack. |
| TLS, DNS, ingress | Let's Encrypt, zero manual renewals. |
| Secrets | Vault or sealed secrets, nothing in git. |
| Cost reports | Monthly cost report per service, right-sizing, idle resources found and removed. |
| Runbooks | Written for your team, so you're not locked to us. |
Pricing
Audit
Free
Send your bill range and a list of services. Get a sized Hetzner plan within three business days. No call, no access needed.
Migration
from $1,500fixed
Up to 3 services and one database. From $2,500 with queues, several databases, or AWS-only services to replace. $790 when bundled with 6 months of Cluster or Studio. Paid 50% at start, 50% after cutover.
Run
Single$290/mo
One server. Backups with test restores, monitoring and alerts, business-hours response (UTC+6, overlapping Europe), patches.
Cluster$790/mo
k3s cluster. Outage response 24/7, patches and upgrades, Slack in business hours, monthly report, up to 4 hours of changes included. Next cluster +$300/mo.
Cancel with 30 days notice.
Studio
$590/mo per client cluster
For agencies with two or more clients on Rackfox. White-label, 3-month minimum. A single client from an agency is priced as Cluster.
Hetzner servers are billed to you directly. We don't resell hardware or mark it up.
Add-on to a managed cluster
AI Inference
+$390/mo· one-time setup $490
Your LLM, on your Hetzner GPU. Nothing leaves the EU.
Run open models next to your own services instead of sending customer data to OpenAI. One OpenAI-compatible endpoint: your code and your coding agents don't change, only the base_url.
- Chat & agents: Qwen 3.6 (35B, 256k context, tool calling), or a lighter 14B model on the 24 GB tier
- Embeddings & reranking: bge-m3 and bge-reranker for RAG
- Speech-to-text: Whisper for calls and meetings
- OpenAI-compatible API (/v1/chat/completions, /v1/embeddings); works with LangChain, Cursor, Cline, Continue
- Dedicated GPU: no shared capacity, no per-token bill, no third-party rate limits
- Same DPA and 24/7 on-call as your cluster; model updates and GPU monitoring included
Hetzner GEX line, from ~€200/mo for the 24 GB tier, ~€1,000/mo for 96 GB. You pay Hetzner directly, as with everything else we run.
Not a managed ChatGPT. We don't resell OpenAI or Anthropic. If you need GPT or Claude, we wire the endpoint to them and keep the EU-only models for the sensitive paths.
Why Hetzner and not another cloud
A dedicated AX-series server with 64 GB RAM and NVMe costs about €50/month. The same RAM on AWS EC2 is $250–400 before storage and traffic.
Kubernetes doesn't care whether it runs on EC2 or on a box in Falkenstein. Your app doesn't either.
You lose managed RDS and a few conveniences. We replace them with Postgres operators and backups that we test. That's the trade.
Proof: we run our own stack
Every client of our sister studio Buildify runs on the same setup we'd build for you: self-hosted Kubernetes, GitOps, secrets in Vault. These sites are live. Open them.
- muras.buildify.info
School site in three languages
- cafe.buildify.info
Café CRM: menu, orders, kitchen screen, waiter app
k3sArgoCDGitLab CIVaultAnsiblePrometheus
Who's behind Rackfox
Dmitry Zubrikov, founder. I've been a DevOps engineer since 2021. I've run infrastructure for schools, cafés, and SaaS products on self-hosted Kubernetes. GitLab, Vault, ArgoCD, and Ansible are in production for me every day. You talk to me, not to an account manager.
- team
- A team, not one person. First response within an hour.
- location
- UTC+6 · overlapping Europe mornings
- fit
- If your task isn't a fit, we say so in the first call.
FAQ
What if a server dies?
Workloads reschedule to another node; databases fail over from a replica. We design for one node loss from day one.
Do we lose AWS-only services?
Lambda, SQS, DynamoDB need replacements. We tell you in the audit which ones and what they cost to replace.
What about compliance and GDPR?
Hetzner is in Germany and Finland, data stays in the EU. That's usually easier than AWS us-east-1.
Can we leave later?
Yes. Everything is in your git repo and your Hetzner account. Runbooks are written for your team.
Do you work with agencies?
Yes, white-label is fine. We're the ops line item on your invoice.
Why not just use the OpenAI API?
For most teams you should, for most things. The AI Inference add-on is for the paths where customer data, contracts or health records must not leave your infrastructure, and for teams whose token bill has become unpredictable.
What time zone are you in?
UTC+6. Overlap with Europe mornings and US late afternoons. On-call alerts are 24/7 on Cluster and Studio.
Do you sign a DPA?
Yes. A DPA with EU Standard Contractual Clauses, and Rackfox goes on your subprocessor list.
Who is on call?
A team, not one person. First response within an hour; 24/7 on Cluster and Studio, business hours on Single.
Find out your Hetzner number
Send your bill range and a list of services, get a sized Hetzner plan within three business days. No call, no access needed.