Local Assistant and Compute
Kubernetes server for OpenClaw, compute, and website hosting
Overview
Cluster started as an assortment of devices I already had at home, all connected through an unmanaged router. It now runs k3s and OpenClaw across the full stack, with robust pod redeployment across nodes, PersistentVolumeClaim storage, and an automatic backup helper for stateful workloads.
Networking
Flannel runs fully over ethernet, with cross-device public-key access between nodes. Critical machines are reachable through Tailscale, and game-server traffic is exposed through playit.gg. Each device can either use its own Wi-Fi uplink or be bridged through another node when needed.
Inference And Compute
The PC hosts vLLM and ollama endpoints for Qwen3.6 35B, while the Orins run a backup ollama endpoint for Qwen3.5 2B. For heavier jobs, the PC and Lenovo E580 take on the higher-compute workloads.
Deployments
- Local OpenClaw
- Headlamp dashboard
- Game servers and associated websites
- SearXNG
- Obsidian integration
