Personal site · Local AI · Self-hosted

Miroslav Kubík

AI that runs on hardware I actually own.

I build, tune, and maintain private AI systems that run entirely on my own machines — and I help other people do the same. This page is my quiet corner of the internet where all of that lives.

Email me →
Setup  owned hardware, local models Favours  benchmarks over press releases Rule  your data stays yours
01 — About me

A bit about me

I like machines I understand. That preference led me to local AI: running models privately, on hardware I own and configure myself, instead of renting someone else's intelligence by the token.

Most of my time goes into the parts nobody demos — sizing GPUs, tuning drivers, quantizing models until they are fast without being dumb, watching thermals and token throughput. Unglamorous work, but it is the difference between a local AI you actually use and one that quietly died in a forgotten notebook.

People still ask whether local AI is a compromise. It isn't anymore — the models are good, the tooling has matured, and privacy turns out to be a feature nobody can take away. Friends, colleagues, and the occasional stranger who lands here get the same treatment: I build the thing, then I write it down so someone else can too.

02 — Interests

Where the time goes

Less a service menu, more a list of things I genuinely enjoy figuring out — and usually end up helping someone else with anyway.

/ 01

My own setup

The homelab is the hobby itself: hardware chosen for the workload, inference tuned until it feels fast, and everything documented well enough to rebuild from scratch.

/ 02

Models that earn their size

Quantization, context strategy, honest benchmarks. A model that answers in a second beats one with a better score that keeps you waiting — and I like finding where that line is.

/ 03

Private workflows

Notes, documents, and automation running against local models. Search that never leaves the house. This is the part that convinced me the whole thing isn't a fad.

/ 04

Helping others go local

Friends, colleagues, and whoever asks get the same thing: hardware advice, a clean install, sensible defaults, and notes written for whoever inherits it next.

/ 05

Keeping things alive

Systems drift — drivers break, models update, disks fill up. I maintain what I build, quietly and on purpose. Uptime is underrated and rarely interesting, which is exactly the point.

03 — Contact

Say hello

If you want to talk about local AI — a project, a stubborn benchmark, or which GPU makes sense this year — write to me. I read everything, and I reply to most.