Install
Install geo-mlops-sdk with pip and pick the extras you need.
geo-mlops-sdk is on the public PyPI. It needs Python 3.10 or later (3.10 to 3.14 are supported).
One-line install
To run the agent on a device, report NVIDIA GPU utilization, read PLCs and run YOLO models on the device:
python3 -m venv /opt/geo-mlops/venv
/opt/geo-mlops/venv/bin/pip install 'geo-mlops-sdk[edge,gpu,modbus,yolo]'
/opt/geo-mlops/venv/bin/geo-mlops-edge --version # 0.3.0/opt/geo-mlops/venv is the path the systemd unit in the Deploy page expects. If you put it somewhere else, change the unit's ExecStart too.
Extras
The base install (pip install geo-mlops-sdk) installs only two packages, httpx and pydantic, because an application that uses only CentralClient needs no web server or GPU libraries. Pick the rest as extras.
| Extra | Adds | When you need it |
|---|---|---|
edge | Agent runtime, local API (FastAPI), the geo-mlops-edge daemon | To run the agent on a device — in practice, always |
gpu | NVIDIA GPU utilization in heartbeats (nvidia-ml-py) | Devices with an NVIDIA GPU. Without it, GPU values are reported empty |
modbus | Modbus TCP collector (pymodbus) | To read PLCs |
yolo | Run YOLO models on the device (ultralytics, numpy, pillow) | To run YOLO inference on the device |
rfdetr | Run RF-DETR models on the device (rfdetr) | To run RF-DETR inference on the device |
all | All of the above | To try everything at once on a development PC |
If you call an agent feature such as geo_mlops_sdk.EdgeRuntime without its extra, the error tells you which extra to install.
ImportError: EdgeRuntime needs the edge runtime; install 'geo-mlops-sdk[edge]'The geo-mlops-edge command is installed even by the base install, but without the edge extra it exits with ModuleNotFoundError: No module named 'pydantic_settings'.
Do not use 0.0.2 from PyPI
Check the install
geo-mlops-edge --version
python -c "import geo_mlops_sdk; print(geo_mlops_sdk.__version__)"Next: in the 5-minute start, connect a real device.