This repository contains the IaC (Infrastructure as Code) configuration for my homelab.
My homelab runs two infrastructure stacks: Kubernetes and Proxmox VMs running Docker.
Kubernetes is the primary platform. The cluster runs on bare-metal Talos Linux nodes defined as code with talstomize, and everything in it is managed with GitOps practices.
Legacy VMs are Ubuntu cloned from templates I created with Packer. I have been migrating my Ubuntu VMs over to NixOS (see Nix config here), and going forward all VMs will be NixOS.
To automate infrastructure updates, I use Github Actions, which trigger workflows upon changes to this repo. This ensures seamless deployment and maintenance across my homelab:
- Flux manages Continuous Deployment (CD) for Kubernetes, deployed via Flux Operator.
- Docker CD Workflow handles Continuous Deployment for Docker services.
- Renovate keeps services updated by opening PRs for new versions.
- Ansible is used to execute playbooks on all of my VMs, automating management and configurations
- Comin keeps NixOS flakes in sync with this repo
- tofu-controller GitOpsifies my Terraform deployments, letting Flux plan and apply from this repo
For Secret management I use Bitwarden Secrets and their various integrations into the tools used.
Kubernetes is using External Secrets implementation of BWS, not official. BWS Access Key is SOPS encrypted.
GitLeaks makes sure before every commit no secrets are exposed, GitGuardian makes sure to alert me if something slips through GitLeaks.
Each container image is automatically scanned by Trivy, with detected vulnerabilities published to Github Security
I use RackNerd for their very reasonably priced VPS and deploy Docker services that require uptime here. Tailscale is used to connect my home network to the various VPS's securely using Zero Trust architecture.
I use Cloudflare for my DNS provider with Cloudflare Tunnels to expose some of the services to the world. Cloudflare Access is used as Zero Trust for public websites, this is paired with Fail2Ban looking through all my reverse proxy logs for malicious actors who made it through Access and banning them via Cloudflare WAF.
I also utilize Unifi's IDS/IPS for intrusion detection on my home network.
I use a combination of Grafana, fluent-bit, VictoriaLogs and Prometheus with various exporters to collect and visualize system metrics, logs, and alerts. This helps maintain visibility into my infrastructure and detect issues proactively.
- Prometheus – Metrics collection and alerting
- Victoria Logs – Centralized logging
- Grafana – Dashboarding and visualization
- Exporters – Blackbox Exporter, Speedtest Exporter, etc.
I use Bifrost as a model gateway for local and cloud AI providers.
- Providers: Ollama, NVIDIA NIM, Opencode ($10/m), Claude ($20/m), Codex ($20/m), and OpenRouter
- Interfaces: Open WebUI for chat, T3Code & OAI Compatible Provider for Copilot
- Infrastructure:
- cli-proxy-api exposes subscriptions over API to Bifrost
- mcphub provides one MCP connection to all of my MCP servers
- SearXNG provides web search
- Langfuse provides LLM observability (traces, costs, latency)
- ai-proxy (custom) backfills
/v1/modelswith models.dev metadata so chat and IDE clients actually know each model's capabilities
Although I try to self-host everything I can, my infra still relies on the cloud for certain services.
| Service | Use | Cost |
|---|---|---|
| Proton | IMAP, SMTP, VPN (Pass once there is Autofill Hotkey) | ~$120/yr |
| Bitwarden | Secrets for all tools | Free |
| Backblaze B2 | Offsite backups | ~$5/mo |
| Cloudflare | Domain, DNS, WAF | Free |
| GitHub | Hosting this repo and continuous integration/deployments | Free |
| RackNerd | RackNerd VPS, runs services such as Forgejo | ~$60/yr |
| Total: ~$20/mo |
This repo is not structured like a project you can easily replicate. Although if you are new to any of the tools used I encourage you to read through the directories that make up each tool to see how I am using them.
Over time I will try to add more detailed instructions in each directories README.
Some good references for how I learned this stuff (other than RTFM)
- Kubernetes Cluster Setup
- Kubernetes + Flux
- Kubernetes Secrets with SOPS
- Finding Kubernetes HelmReleases
- Packer with Proxmox
- Terraform with Proxmox
- Docker
- Ansible
Special thank you to @chkpwd for helping me get this started. His repo was the inspiration for this.
Proof that you don't need expensive new equipment to run infra like mine. Mostly everything here is secondhand, bought over time, totaling less than ~$4k.
Servers
| Name | Device | CPU | RAM | Storage | GPU | Purpose |
|---|---|---|---|---|---|---|
| Talos-4 | Lenovo P330 Tiny | Intel i7-9700 | 64GB DDR4 | 2x 512GB NVME | Integrated | k8s control-plane |
| Talos-5 | Lenovo P330 Tiny | Intel i7-9700 | 64GB DDR4 | 2x 512GB NVME | Integrated | k8s control-plane |
| Talos-6 | Lenovo P330 Tiny | Intel i7-9700 | 64GB DDR4 | 2x 512GB NVME | Integrated | k8s control-plane |
| TrueNAS | Custom | AMD Ryzen 5 5500 | 32 GB DDR4 | 1TB NVMe, 4x4TB RAIDZ1 (Media), 2x4TB Mirrored (Backups) | Arc A310 | NAS + Jellyfin Server |
| PVE | Custom | AMD Ryzen 9 5950X | 64 GB DDR4 | NVMe for boot and VMs | Nvidia 1660 6GB | Main proxmox node |
| Pi | Raspberry Pi 4 | 8GB | 1TB m.2 SATA SSD w/ USB HAT | n/a | Home Assistant Server | |
| Proxmox Backup Server | Mini-PC | Intel N150 | 8GB | 2TB SATA | n/a | Backup Proxmox VM's |
Personal
| Name | Device | CPU | RAM | Storage | GPU | Purpose |
|---|---|---|---|---|---|---|
| Gaming PC | Custom | Intel i7-13700k | 64GB DDR5 | 10TB NVMe | Nvidia RTX 5070 | Main Machine |
| Laptop | HP 15-eh1097nr | AMD Ryzen 7 5700U | 32GB DDR4 | 1TB NVMe | Integrated | On the go/bed machine |
Networking
| Name | Device | Purpose |
|---|---|---|
| Switch | Pro XG 8 PoE | Switch with PoE |
| Router | Cloud Gateway Fiber | Router/Firewall |
| AP | U7 Pro XG | AP |
See Project Board
