Getting Started¶

Everything you need to install KITT, run your first benchmark, and deploy with Docker.

What's in This Section¶

Install KITT using Docker (recommended) or from source with Poetry. Covers prerequisites, optional extras, and GPU setup.

A step-by-step walkthrough from hardware fingerprinting through benchmark execution to storing results in KARR.

Run KITT entirely from a container -- build the image, mount models and results, use Docker Compose, and generate production deployment stacks.

Requirement	Why
Docker	All inference engines run as containers
NVIDIA GPU + drivers	Required for GPU-accelerated inference
NVIDIA Container Toolkit	Lets Docker containers access the GPU
A model on disk	GGUF, safetensors, or PyTorch format depending on engine

Tip

The fastest path to your first result is the Docker method described in the Installation guide -- no Python environment needed on the host.