Homelab
One NAS, 20+ services and 5 AI agents. All self-hosted, all in production.
Network topology
Internet
│
├── VPN ── Acceso remoto
│
Router
│
├── NAS
│ ├── Contenedores (+20)
│ │ ├── Domótica
│ │ ├── Servidor de medios
│ │ ├── DNS / Bloqueo de anuncios
│ │ └── Herramientas de desarrollo
│ └── Agentes de IA (×5)
│ ├── Orquestación
│ ├── Desarrollo
│ ├── Trading
│ └── Medios
│
└── Dispositivos IoT
├── Zigbee
├── WiFi
└── miioWhy
The cloud won't let me touch what I want to touch.
It started as curiosity and became a need. I wanted a place to experiment with AI, home automation and DevOps without depending on third parties or paying per container. A central NAS running 20+ containerized services with Docker Compose: development lab, multi-protocol smart home hub and AI agent platform. All under my control.
Architecture decisions
Every decision has a why.
- Docker Compose over Kubernetes: single node, simplicity wins
- SSD for databases, HDD for media: cost/performance tradeoff
- DNS-level ad blocking: network-wide, zero client config
- Mesh VPN for remote access: no exposed ports, no router changes
- Multi-agent AI system: specialized agents per domain
Multi-agent system
The differentiator: specialized AI running locally.
A multi-agent system runs directly on the NAS. Each agent has a clear domain: development, trading, media, orchestration, each with its own tools and context. They communicate in real time and coordinate tasks autonomously. The design decision was specialization over generalization: agents with bounded domains instead of a monolithic model. Result: contained errors, parallel coordination, and horizontal scaling by capability.
- Specialized agents with separate domains
- Real-time inter-agent communication
- Autonomous coordination without human intervention
- Persistent services with automatic startup