Llama
Gemma
Qwen
Ministral
GPT-OSS
DeepSeek
Phi
Falcon
// linux, self-hosting, and local AI, one bad decision at a time
Notes from the homelab.
A build log of self-hosting, home-grown AI infrastructure, and the occasional 2am fix that turned out to be a mount path all along.
asrock-desktop
RX 9060 XT · gemma4:12b + gpt-oss:20b
active (running)
gx531gm
GTX 1060 · batiai/gemma4-e2b:q6
active (running)
fullmetal975
RX 580 · nextcloud, jellyfin, pihole
active (running)
comfyui-box
GPU pending
planned
$ ls -la posts/
- 2026.07.07 › Llama 3.2 Vision Doesn't Run Here (And Here's Why)
- 2026.07.03 › Five Models, One GTX 1060: The GX531GM Local AI Bake-Off
- 2026.07.03 › Getting this blog and its own email address live: a full night of self-inflicted problems
- 2026.07.02 › Building out my gaming server on Linux Mint
- 2026.07.02 › Can a local model actually design something? gpt-oss:20b vs Gemma 4 12B
- 2026.07.02 › Finding the Best Local LLM for a 16GB AMD GPU
- 2026.07.02 › Open WebUI: why I stopped trusting Knowledge Bases and just typed the facts into Memory
- 2026.07.02 › Three tests, one winner: what Gemma 4 got right that gpt-oss didn't
- 2026.07.02 › Unresolved: intermittent response delays on the laptop, and everything I ruled out
- 2026.07.01 › Three Bugs Deep: Getting Open WebUI Running on My ASUS ROG Laptop
- 2026.06.30 › Finding the Right Local AI Model for a 6GB Laptop GPU