terminal — ~/desk

80×24

rig

mac miniM4 · 24 GB unifiedmodel host · MLX
mac pro 5,12× Xeon X5690 · RX 6900 XT 16 GB · 128 GB DDR3heavy lifting · Blender · long jobs

current experiment

local agent

Codex CLI driving macOS through a Qwen model on MLX. No cloud.


best local model

Qwen3.8-27B 4-bit

MLX-VLM on the mini — text + screenshots. the one that actually finishes tasks.

projects

3 items

/robert/desk/projects

notes.txt

— edited