openeye pull

Download model weights from Hugging Face Hub to ~/.openeye/models/<model>/.

openeye pull yolov8
openeye pull depth-anything
openeye pull grounding-dino
openeye pull yolov8 --quantized
openeye pull --all

Arguments

Argument	Description
MODEL	Registry key to pull (see openeye list). Optional when --all is given

Options

Flag	Description
--all	Pull every model in the registry
--variant <name>	Pull a specific variant (yolov8 has a quantized variant, yolov8n_int8.onnx)
--quantized	Shortcut for --variant quantized
--force	Re-download even if already present

Registry

Key	Task	Weights
yolov8	detection	ultralytics/yolov8 (yolov8n.pt)
yolo26	detection	ultralytics/yolo11 (yolo11n.pt for now)
depth-anything	depth	depth-anything/Depth-Anything-V2-Small-hf
grounding-dino	detection (open-vocabulary)	IDEA-Research/grounding-dino-tiny
sam2	segmentation	facebook/sam2-hiera-small
rfdetr	detection	rfdetr/rfdetr-base
smolvla	vla (vision-language-action)	lerobot/smolvla_base

Running a model also needs its Python extra, for example pipx install "openeye-sh[depth]" --force; the CLI tells you which one if it's missing. To delete downloaded weights, run openeye remove <model> (see Model Management).