Self-hosted AI and automation
OpenClaw is my experiment with a self-hosted AI gateway: a central service that can connect chat, tools, automation and notifications. It runs in my own homelab, so I learn about managing and securing it as well as exploring its possibilities.
At a glance
The gateway runs in an Ubuntu Linux container and starts as a system service. OpenClaw can use Docker as a sandbox for isolated tasks. A chat connection enables interaction and notifications without having to work directly on the server every time.
The setup is deliberately modular. The gateway, plugins, access rules, sandbox and communication channels each have their own responsibility. This makes the system more flexible, but also requires clear boundaries between those parts.
A gateway is interesting because it does more than produce text. It can use available tools, run scheduled tasks and check the state of other components. For example, I created an automatic health check that periodically verifies whether the gateway is reachable and reports the result to a chat channel.
This turns OpenClaw into a connecting layer between AI and the day-to-day management of the homelab.
A service that can use tools and is accessible over the network needs careful boundaries. During the project I therefore checked network access, tokens, plugin policy, user permissions and the isolation of commands in Docker.
Security checks revealed several warnings. Critical issues were resolved, while the remaining warnings guide further hardening. Security therefore becomes a recurring part of the experiment rather than a one-time final check.
Building further
Two older servers form a personal Proxmox homelab for containers, virtual machines, storage, backups and self-hosted experiments.
View project →