Installing Gorget

Prerequisites

Supported Python versions: 3.10 to 3.14.

Using pip

Note

Consider installing Gorget in a virtual environment like venv or conda.

pip install gorget

This installs Gorget with ONNX Runtime and without PyTorch: every scanner runs on CPU, and the install is several times smaller than one that pulls PyTorch.

To run models with PyTorch instead, for example on a GPU, add the torch extra:

pip install "gorget[torch]"

With PyTorch installed, scanners use it unless you pass use_onnx=True. The GORGET_BACKEND environment variable (onnx or torch) overrides the choice for every scanner.

Moving from LLM Guard

Gorget continues LLM Guard, so existing code keeps working:

pip uninstall llm-guard
pip install gorget

from llm_guard.input_scanners import PromptInjection  # still works
from gorget.input_scanners import PromptInjection  # the new name

Both imports return the same classes. LLMGuardValidationError is still available as an alias of GorgetValidationError.

No downloads at runtime

Gorget does not install anything while it runs. Models come from the Hugging Face Hub on first use and are cached; spaCy models and NLTK data are no longer downloaded. To run offline, fill the cache once and set HF_HUB_OFFLINE=1.

Install from source

git clone https://github.com/Perruer/gorget.git
cd gorget
python -m venv venv
source venv/bin/activate
python -m pip install -e ".[dev,torch]"