Repository navigation
Releases
v0.3.1
labwright 0.3.0
Guarded lw.py free (unload Ollama, then ComfyUI /free, only when nothing else runs), global --json envelope {ok, data, error} with exit 0/1, pick --model reads measured peak VRAM from a registry runs_file, doctor warns on a stale ComfyUI-Manager snapshot, status shows GPU temperature and power. Ideas from Comfy-Org comfy-mcp, comfy-cli and comfyui-benchmark (GPL: ideas only, no code).
labwright 0.2.0
First public release.
labwright lets a coding agent read your local AI servers (ComfyUI, Ollama) as one pool: which host is up, idle or busy, where a model lives, which host should take a job or which one to queue behind, and which drive can take a download while keeping a free-space floor. It only reads; it never submits, cancels or downloads anything.
lw.pycommands: hosts, status, where, pick, place, doctor.- Idle hosts report VRAM that is still held by a loaded model.
pick --vramnotes when a job will not fit on a host.- Your host registry stays in a private folder named by
LABWRIGHT_HOME.
Install: /plugin marketplace add m4bwav/labwright, then /plugin install labwright@labwright.