This article continues the story that started with the Birth of the Infrastructure.
In the beginning, everything lived on a small VPS: reverse proxy, applications, experiments, monitoring. It worked – but it was never meant to scale. The next logical step was obvious:
Move the heavy work home, keep the VPS as a lean edge.
That is how the HOME server was born as an AI & compute hub.
1. Why I Moved Compute Home
Running everything on a VPS was convenient, but it had clear limits:
- Resources – RAM and CPU on a small VPS are always a bottleneck.
- Cost – bigger VPS = higher monthly bill.
- Control – I wanted full hardware control (disks, power, network, topology).
- Privacy – some workloads (AI, logs, automation flows) should stay fully local.
So I decided to split the architecture into two clear roles:
- EDGE (VPS) – public entry point, TLS, routing, VPN, relay.
- HOME – actual compute: containers, AI, databases, automations.
The HOME server became the place where everything real happens.
2. The Hardware and Base System
The HOME node is a small, quiet but powerful box:
- x86_64 machine with enough RAM and SSD storage
- Connected to the main LAN
192.B.B.0/24 - Running Debian 13 “Trixie”
Base setup:
- minimal Debian installation
- OpenSSH with key-based auth
- basic hardening (firewall, updates)
docker+docker composeas the core runtime
From that point on, almost everything is managed as containers.
3. The Service Stack on HOME
Over time, the HOME server grew into a full-stack platform with these key components:
3.1 Pi-hole – DNS and Filtering
Pi-hole is the DNS backbone of my network:
- runs on HOME as a container
- answers DNS for the entire LAN
- provides blocking, filtering and visibility
- is also used by WireGuard clients as their DNS
Almost every service, including the AI stack, implicitly depends on Pi-hole working correctly.
3.2 WordPress (public-website.example)
The educational project Edu-Cat was migrated from external hosting to the HOME server:
- WordPress runs in Docker with a dedicated MariaDB container
- reachable from the internet through the VPS + Traefik + WireGuard path
- backed up and monitored as part of the standard stack
This move gave me:
- full control over plugins, PHP version and performance tuning
- independence from shared hosting limitations
- better integration with the rest of the ecosystem
3.3 n8n – Automation & Orchestration
n8n is my automation engine:
- run as a container on HOME
- accessible internally and from the internet through
automation.example(via VPS and WireGuard) - connected to local services and external APIs
Typical use cases:
- health checks for services (HTTP, DNS, Docker)
- alerts and notifications
- scheduled maintenance tasks
- AI-assisted workflows that combine Sim AI / Ollama with remote services
n8n is the glue that binds infrastructure pieces together.
3.4 Sim AI (OpenWebUI)
Sim AI is my main UI for working with AI models:
- hosted on HOME, again via Docker
- used both through browser and via API (e.g. n8n workflows)
- acts as a unified front-end for local models (Ollama) and remote models (OpenAI, etc.)
It turns the HOME server into a personal AI lab, not just a static apps box.
3.5 Ollama – Local Models
Ollama runs on HOME to provide private, local inference:
- large language models stored and loaded directly on the box
- used by Sim AI and n8n through the Ollama API
- ideal for logs analysis, experiments and tasks where I want data to stay at home
The difference is simple:
- Cloud models → best quality, no hardware cost, less privacy
- Ollama → full control, offline-capable, limited by local hardware
The HOME server makes both options possible and easy to switch between.
4. How HOME Connects to the Edge (VPS)
The HOME server is not directly exposed to the internet.
Instead, everything goes through:
- VPS (EDGE)
- Traefik as reverse proxy
- WireGuard server (10.A.A.1/24)
- RustDesk relay
- MikroTik router
- WireGuard peer (10.A.A.2)
- LAN
192.B.B.0/24routing
- HOME server
- WireGuard peer (10.A.A.3)
- services bound only to LAN / local ports
Public requests (for example to automation.example or public-website.example) travel like this:
Internet → VPS (Traefik, HTTPS) → WireGuard tunnel → HOME:port → container
This design gives me:
- end-to-end encryption
- a single public IP at the edge
- a private, isolated compute environment at HOME
5. Docker Structure and Organization
All HOME workloads live under a clear directory structure, e.g.:
/srv/<project>/– root for all compose stacks- one or more
docker-compose.ymlfiles - dedicated volumes for:
- n8n data
- Sim AI config
- Ollama models
- WordPress data
- databases
- Pi-hole configuration
This separation allows:
- easy backup of critical data
- quick redeploy after OS reinstall
- clean upgrades of individual services
6. Monitoring, Recovery and Reality
The HOME server is not just “another machine”:
- it runs the main DNS resolver (Pi-hole)
- it hosts critical automation (n8n)
- it serves public content (public-website.example, SystemLog)
- it provides AI tools I use daily
Because of that, I treat it like a small production environment:
- I keep documented recovery steps for the server, Docker, and the network paths
- I test WireGuard access from external clients
- I track how services behave after power outages and reboots
- I continuously refine the layout based on what breaks in real life
It is not a lab. It is the core of my digital infrastructure.
7. Lessons Learned
Some of the key lessons from building the HOME server as an AI & compute hub:
- A small, well-configured home box can replace multiple VPS services.
- Splitting responsibilities (EDGE vs HOME) improves clarity and security.
- Docker is extremely powerful, but only if you document and backup.
- Local AI (Ollama) is great – but only as part of a hybrid setup with cloud models.
- Automation (n8n) is not “nice to have”. It is necessary if you want to sleep.
The HOME server is not a one-time project. It is an evolving platform that grows together with my needs and my understanding of how I want my systems to behave.
This article is just the snapshot of one point in time.