Running on your machine in about five minutes.

Ollama is the free, open-source engine that runs the model on your graphics card. Download it from ollama.com/download and install it as normal. It works on Windows, macOS and Linux.

Open a terminal and pull the recommended model. It is about 4.9 GB, so it will take a few minutes on a normal connection. You only ever do this once.

# Thornhaver 1.8 — the recommended version
ollama pull granite3.3:8b

Want something faster, or focused on code? Any of these work, and each becomes a version in the app:

# fastest, lightest — Thornhaver 1.2
ollama pull qwen3:4b

# for writing code — Thornhaver 1.5
ollama pull qwen2.5-coder:7b

Download the app files, put them in a folder, and run it. You need Python 3.9 or newer, which most systems already have. There is nothing to install with pip — the whole thing uses only the standard library.

python thornhaver.py

Then open http://127.0.0.1:3000 in your browser. The app finds whichever models you pulled, assigns them version numbers, and loads the recommended one into memory so your first answer is quick.

Thornhaver can measure every version installed on your own machine — speed, the delay before the first word, and whether it declines general questions — then set the best one as your default.

python thornhaver.py bench --set-default

Your hardware is not our hardware, so your numbers will differ from the published benchmarks. Trust yours over ours.