Infrastructure Stack
Tools selected for sovereign AI labs and production-oriented infrastructure. We only list software and providers evaluated for technical fit.
Hetzner Cloud
vpsBest Use Case: Best price-to-performance ratio for CPU-heavy inference and orchestration control planes. Strong European data sovereignty compliance.
Fit: Primary recommendation for K3s control plane and CPU-bound inference tasks. Evaluated in lab environments running Llama 3 8B CPU-only.
Evaluate Hetzner Cloud →n8n
automationBest Use Case: Self-hostable workflow automation with native API chaining. Ideal for connecting local LLM endpoints to internal business tools without data leaving the network.
Fit: Preferred automation layer for sovereign AI pipelines. Evaluated for connecting Ollama API to internal Postgres and webhook triggers in lab environments.
Evaluate n8n →RunPod
gpu-hostingBest Use Case: Ephemeral training, fine-tuning, and high-VRAM inference at a fraction of the cost of hyperscalers.
Fit: Use RunPod for ephemeral training/quantization. Avoid for high-uptime critical APIs unless using secure endpoints.
Evaluate RunPod →Tailscale
securityBest Use Case: Zero-config WireGuard mesh VPN for secure access to private AI endpoints across networks without exposing ports publicly.
Fit: Recommended for secure remote access to Ollama and Open WebUI without port-forwarding. Evaluated in lab environments with OPNsense VLAN routing.
Evaluate Tailscale →Vultr
vpsBest Use Case: Globally distributed compute with bare metal GPU options. Good for European latency requirements and GDPR-sensitive AI workloads.
Fit: Suitable for small-to-mid inference nodes when Hetzner is unavailable. Evaluated for deployment stability in lab environments.
Evaluate Vultr →