A self-hosted homelab running open-source language models, image generation, photo archiving, self-built web apps, and automated infrastructure — entirely on local hardware.
Open-source LLMs running locally on AMD Radeon GPUs via Ollama. Per-user chat instances, local document context, and a translation Chrome extension.
ComfyUI on a dedicated NVIDIA GPU. Text-to-image workflows running locally, no external service involved.
Turn any PDF into a German-language podcast or audiobook using local LLMs and voice cloning. Runs entirely on local hardware.
A conversational agent with a defined personality and long-term memory, running on local models. Reachable through WhatsApp, isolated from the main network.
A self-hosted Swiss transit app that knows your routes and writes a short AI status briefing instead of a wall of departure times.
A shared shopping list in the spirit of Bring!, extended with supermarket price comparison and a monthly household budget.
Immich with AI-powered face recognition, smart search, and automatic album generation across 82 000 photos going back to 2005.
Ansible and Semaphore orchestrate all servers. Scheduled backups, updates, and verified restores — all automated.
The local AI reads, writes, and reviews Ansible playbooks directly on the management server. Dual-model review on every write.
Three AI servers, two virtualisation hosts, two NAS units, and a fleet of VMs. All managed as code, no manual configuration.
Every query to a language model, every photo, every document stays on local hardware. No usage data leaves the house. The tradeoff is maintaining the infrastructure — which is, itself, a large part of the point.
Ollama · Open WebUI · ComfyUI · Immich · Nextcloud · Bitwarden · Mailcow · Pi-hole · Grafana · Prometheus · Ansible · Semaphore · Restic · Nginx · WireGuard · PostgreSQL · Ubuntu 24.04