Blog
Notes from practice – automation with n8n, language models on my own hardware, and whatever the data centre throws at me. Also available as an RSS feed.
Two embedding models, one vector space
Index with one model, search with another, and you get either an error – or quietly wrong results.
When the server asks permission
Human-in-the-loop with n8n and Telegram – maintenance that runs itself but asks at the right moment.
What has to run locally – and what does not
Not every step belongs in house, and not every step may leave it. A decision guide from practice.
Monitoring without a monitoring stack
A scheduled job, one command, a Telegram message. When that is enough – and when it is not.
One server, seven services, 7.7 GB of RAM
Nextcloud, GitLab, n8n, YouTrack, a vector database and a Minecraft server on one machine. What runs out first.
A product advisor for your own shop
Shopify catalogue into Qdrant, vector search as a tool for a language-model agent, Telegram as the interface.
My own server, my own site
Why this site lives on a machine I run myself – and what will show up here.