The space between the letters.
Building open, self-hosted AI infrastructure that does not phone home. Code you can audit, models you can run, agents that answer to you.
I was hands-off with tech for about 15 years. You do not forget how to ride a bike. I still know what I learned, I just did not bother expanding it for a long time, for a lot of reasons.
In the time I built this home AI inference lab, I built a PC. Now I am tinkering, learning new things, having fun with it, and rediscovering what I loved about technology. This profile is that workbench. Some things I built, a lot of things I am studying.
If you are here to learn too, you are welcome here.
It started with AI art. I got into Midjourney, then I wanted more control than that offered. That led me to SDXL and ComfyUI.
But my hardware was limited. I had an old gaming PC that was not going to cut it, so I went to cloud services for a while. Then on a whim I purchased a Jetson Orin Nano to tinker with. Still needed a PC though. My MacBook and the old gaming PC were not going to cut it for what I needed. I had gone cheap on that gaming PC and it is missing some vital input ports. I had also become a console only guy by then.
All roads led to the same place. I was going to have to build my own. And once I start, I cannot stop. It is a habit. Here I am today, with a home lab and a build I did myself, still learning.
One house, four divisions, one rule: the work ships open, and nobody gets credit stolen.
The full INTRON brand system, logo marks, design tokens, booklets, lives in
the /svg and booklets artifacts of this and the division repos.
Most of what's under this account is a laboratory and a set of curated forks (which point to their true upstreams). The genuinely owned projects are the ones I built:
- Atlas, the agent harness. Token efficiency as a design principle, not a bolt-on. Provider-agnostic, no telemetry, bounded orchestrator loop. Built for environments where every line must be auditable.
- IntronTerminal, a local LLM pipeline: SFT, alignment work, eval, serving. The model on your disk answers to you.
- IntronStudio, the creative side: local image and model workflows.
- Armor, the Atlas interface layer (itself inspired by
pewdiepie-archdaemon/odysseus, credited in that repo). - Codex-Intel-Builder, tooling that re-architects an ARM Codex macOS DMG to run on Intel Macs.
The rest of the profile, OBLITERATUS, llama.cpp, T3MP3ST, heretic, jailbreak-autoresearch, and the long tail of forks, is work I study, fork, and build on, not work I claim. Each links to its real author. The fork relationship is the credit; I don't relabel it as mine.
A note on the account name: the login is
ColdSlither(legacy). The house name is INTRON. Same person, same work, the display name says Intron, the URL says ColdSlither. Both point here.
If you're browsing the repo list and seeing a wall of forks (llama.cpp, ComfyUI, VeraCrypt, LibreChat, and many more), those are forks, not claims. I read source before I trust it. Each fork links to its true upstream, and the fork relationship is the credit. None of that work is presented as mine.
If you see something of yours here that isn't credited the way you'd want, open an issue or DM. Credit is not optional, if I missed a line, it gets fixed.
- Self-hosted by default. Your data, your hardware, your inference. Cloud is a choice, never a requirement.
- Auditable over magical. If a system can't be inspected, it can't be trusted. Regulated environments don't accept black boxes, and neither do I.
- Local models are a right. Uncensored, honest, controllable, the weight on your disk answers to you.
- Credit is the floor. Inspiration is a gift; stealing the credit is theft. Every repo carries its sources.
- X / Twitter: @ItsColdSlither
- Location: New Jersey, USA
- Company: Intron
INTRON, respect first. Freedom always. The space between the letters.