Deploy a self-hosted Gitea git server and Jellyfin media server on spare PCs with a single command.
Use GitOps to manage all home services through code, commit a config change and ArgoCD applies it automatically.
Expose home services securely to the internet via Cloudflare Tunnel without opening router ports.
Run a local AI assistant using Ollama on your own hardware inside the cluster.
Requires 4+ bare-metal machines connected via a network switch, comfort with Kubernetes, GitOps, and Linux administration is essential.
This is a personal homelab configuration: a complete blueprint for turning a set of bare-metal machines into a fully automated self-hosted server environment. The creator runs it on four small form-factor PCs connected through a network switch. The project starts from empty hard drives and, with a single command, automatically installs an operating system on each machine over the network using PXE boot (a method that lets machines install software over the network without a USB drive), sets up a container orchestration platform called Kubernetes, and then deploys a collection of self-hosted services. Those services include a git server (Gitea), a media server (Jellyfin), monitoring dashboards (Grafana), a Matrix chat server, a document management system (Paperless), and a CI/CD pipeline for automated testing and deployment. Everything is managed through code stored in git repositories, a pattern called GitOps. When you want to make a change, you update a configuration file and commit it. Tools called ArgoCD and Woodpecker CI then automatically apply those changes to the running system without manual intervention. The feature list covers a wide range: automatic OS and Kubernetes updates, SSL certificate management, DNS record updates for exposed services, VPN access via Tailscale or WireGuard, the ability to expose services to the internet through Cloudflare Tunnel, single sign-on, distributed storage, monitoring and alerting, and automated backup and restore. There is also an experimental self-hosted AI section using Ollama, though the creator notes it runs slowly without a dedicated graphics card. The project is labeled alpha, meaning breaking changes can happen and a complete redeployment may be required after upgrades. It is intended as a framework and reference for people building their own homelabs rather than a production-ready product.
← khuedoan on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.