inference online
Local AI

Famous models.
Your hardware.

What actually runs, what doesn't, and why. Notes from the hardware.

$ proxy-switch genesis
switching backend → Qwen3.6-27B · vLLM · port 8022
tg512: 83.4 tok/s · ctx 32K · MARLIN_USE_ATOMIC_ADD=1
genesis is live.

Latest
The Upgrade That Wasn't
The Local Inference Chronicles: Everything We Tried