A developer going by mrlokans just published a thorough walkthrough of their 2026 homelab, and buried in the media servers and file sync tools is something worth paying attention to: a fully self-hosted AI agent setup running through LibreChat. The stack runs on a GMKTec NUC with 32GB RAM alongside a Hetzner VM for always-on services, with everything containerized in Docker on Debian.
LibreChat handles multi-agent orchestration. You can define AI personas with custom system prompts, attach vector-based knowledge bases, and add tools like web search or code interpreters, which can lead to vibe coding. It connects to cloud APIs from OpenAI and Anthropic, plus local models through Ollama. All of this runs behind Traefik as a reverse proxy, Authentik for single sign-on, and Cloudflare Tunnels for secure external access without opening inbound ports.
The whole setup follows infrastructure-as-code practices using Ansible for automation and SOPS for secrets encryption. Each service, LibreChat included, is an Ansible role with tasks, defaults, handlers, and templates. The AI agent infrastructure can be reproduced or torn down in minutes.
You don't need Kubernetes or fancy orchestration to run multi-agent AI workflows at home. A NUC, some Docker containers, and open-source tools get you a private AI dashboard that rivals commercial offerings. Full control over your data. Zero per-token costs when running local models.