About

neo ~3-4 min read

← Back to posts

hi, I'm neo

I build systems, break systems, and then write down what I learned so I do not repeat the same mistake twice. I still get a small dopamine hit when a shell script prints the right thing on the first run.

My background is the usual arc: beige-box PCs, late-night reinstalls, and eventually getting paid to do the same thing in a more controlled, less dramatic way. These days I run a homelab I call the grid, mostly because calling it "a few servers" would be a lie. 

          ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄
        ▄█ ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄█ ▄
       ██   ▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀   ██
       ██                                         ██
       ██   ┌───────────────────────────────┐     ██
       ██   │  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓  │     ██
       ██   │  ▓▓▓  GRID  MODE  ACTIVE  ▓▓▓  │     ██
       ██   │  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓  │     ██
       ██   └───────────────────────────────┘     ██
       ██              neo@80x24.ch               ██
       ██                                         ██
       ██   ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄   ██
        ██ ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄█
          ▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀

what is 80x24.ch

This site is my public notebook. It is not a product, not a course, and not a benchmark blog. It is a collection of practical notes that come out of running local LLMs and doing boring ops work. In my lab, "boring" is a compliment. Boring means repeatable. Boring means you can sleep.

I try to write the kind of posts I wish existed when I am staring at a terminal at 02:00. Not perfect blueprints, just clear narratives: what I tried, what I expected, what actually happened, and what I would change next time.

the vibe: local-first, but realistic

I like local-first setups because they remove a bunch of invisible friction. When the meter is not running, I explore more freely. When the data stays inside my rack, I am less careful in a good way: I can paste messy notes, half-finished scripts, and the weird questions you only ask when nobody is watching.

That said, I am not allergic to the cloud. I use it when it makes sense. The theme here is: build a default workflow that works offline and keeps working when a vendor changes pricing, policies, or feature flags.

what you'll find here

  • Local LLM ops: inference runners, quantization, prompt hygiene, and tool calling.
  • Homelab notes: Proxmox, ZFS, backups, and the stuff you only respect after your first outage.
  • Automation and scripting: defensive Bash, Python tooling, and small glue scripts.
  • Mistakes: because I keep making new ones, and they are expensive to forget.

rough "specs" (today, not forever)

I avoid publishing a full parts list because hardware changes constantly and I do not want these pages to age into a shopping guide. The broad strokes are stable though. In my lab right now, the grid is roughly:

  • Linux everywhere (Ubuntu 24.04 LTS shows up a lot).
  • Proxmox 8.x for the virtualization layer.
  • ZFS for storage because snapshotting is how I avoid self-inflicted disasters.
  • Local inference mostly via llama.cpp, with other runners when convenience wins.

operating rules

The grid stays fun as long as it stays operable. A few rules keep me honest:

  • Prefer boring defaults. If it requires heroics to keep running, it will not keep running.
  • Measure something. "Feels faster" is not a metric. Even rough timing beats vibes.
  • Write it down. Future me is a stranger with a deadline.
  • Backups are real. Snapshots, replication, and test restores, not just good intentions.

contact

Email: neo@80x24.ch. If you disagree with something I wrote, even better, just bring receipts. If you spot a mistake, please tell me. I would rather fix it than be confidently wrong.