Magazine
Running your own AI: notes from the lab
Practical guides from our lab on running your own AI: which model fits which GPU, vLLM on a Hetzner server, uncensored models, GDPR and AI agents that do real work. Plus why, after ten years on WordPress, I rebuilt this site with Astro.
All articles
-
What is Astro, and how does it compare with WordPress on speed, SEO, editing and cost?
Astro turns a website into finished HTML files before anyone visits. WordPress usually builds each page at the moment someone asks for it. That one difference explains most of this comparison, from speed and security to editing and cost. This site moved from WordPress to Astro in September 2026.
-
Local AI agents for business: what they are and what they can do on your own server
An AI agent is a language model with tools that works in a loop until the task is done. The model, the loop and your internal tools fit on a server of your own, so your documents and tickets never reach a model provider. This guide shows what an agent can do for your business, which controls it needs and when a simple workflow does the job better.
-
Ollama vs vLLM: which one to run an LLM on your own server, and when llama.cpp is the better pick
Ollama gets a model talking on a laptop with one command. vLLM is built to serve many requests from one GPU server. This guide shows which one fits your case, with numbers from the vLLM server we run ourselves.
-
How a model loses its “no”: abliteration, Heretic and uncensored LLMs
Abliteration removes a model's refusals straight from its weights. No retraining, just a few hundred example prompts and a little linear algebra. Done carefully, the model answers what it is asked and keeps nearly everything it knows.
-
Moving away from WordPress after more than ten years inside it
I built on WordPress for more than ten years, ran a company around it and spoke at WordCamp Europe. AI made leaving cheap. Years of building on someone else's platform made me want to.
-
Which open and uncensored LLMs fit 20, 24 or 96 GB of VRAM
What really fits on a GPU with 20, 24 or 96 GB once you count the model file, the context and everyone using it at the same time.
-
Self-hosted LLM and GDPR: what your own AI server solves
With your own AI server in Germany, the biggest GDPR worries go away: no model provider reads along, the model needs no transfer to the US, and you decide what is stored. What that means for contracts, logs, erasure and the DPIA, and what stays on your list.
-
Running an LLM on a Hetzner GPU server with vLLM: the setup we use
We serve an open-weight model, Qwen3-14B, from a rented 20 GB GPU at Hetzner in Germany. Here is the command we run, where the memory goes, how fast it answers and what a Hetzner GPU server costs.