Skip to main content

Installation

This guide walks you through installing Ollama, pulling the right model, and wiring git-sage into your git workflow. The whole process takes about 5 minutes, most of which is the model download.

Requirements​

  • Python 3.9 or higher
  • Ollama installed and running
  • macOS, Linux, or Windows (WSL2)
  • ~5 GB disk space for the model

No GPU required. The default model runs on CPU.

Install Ollama​

macOS

brew install ollama

Or download the desktop app from ollama.com.

Linux

curl -fsSL https://ollama.com/install.sh | sh

Windows

Download the installer from ollama.com. Run it inside WSL2 for best compatibility.

Start Ollama and pull the model​

Start the Ollama server in one terminal:

ollama serve

In a second terminal, pull the model (~4.5 GB):

ollama pull qwen2.5-coder:7b
Why qwen2.5-coder:7b?

It's a code-specialized model trained on source code and diffs. It understands unified diff format natively, flags real issues, and runs comfortably on 8 GB RAM at Q4 quantization — the best code-focused model available at the 7B size tier.

Verify Ollama is running and the model is available:

ollama list

You should see qwen2.5-coder:7b in the output.

Using a different model​

qwen2.5-coder:7b is the recommended default, but git-sage works with any model available in Ollama. You might want a different model if:

  • Your machine has less than 8 GB RAM. Try qwen2.5-coder:3b (~2 GB) for a lighter footprint
  • You have more RAM or a GPU. Try qwen2.5-coder:14b or qwen2.5-coder:32b for sharper reviews
  • You prefer a general-purpose model. llama3.2:3b or mistral work fine for mixed codebases
  • You want maximum accuracy and have the hardware. deepseek-coder-v2:16b is excellent for complex review tasks

Pull any model the same way:

ollama pull llama3.2:3b
ollama pull deepseek-coder-v2:16b

Then pass it to git-sage with the --model flag:

git-sage review --model llama3.2:3b

To browse the full library of available models, visit ollama.com/search. Filter by Code tag to find models optimised for programming tasks. Any model listed there can be pulled and used with git-sage immediately.

Persisting your model choice

If you always want to use a specific model, set the GIT_SAGE_MODEL environment variable in your shell profile:

# ~/.zshrc or ~/.bashrc
export GIT_SAGE_MODEL=deepseek-coder-v2:16b

Then in your hook script (.git/hooks/pre-push), update the call to read it:

git-sage review --hook --model "${GIT_SAGE_MODEL:-qwen2.5-coder:7b}"

Install git-sage​

From PyPI

pip install git-sage

From source

git clone https://github.com/wolz-CODElife/git-sage
cd git-sage
pip install .
Virtual environments

It's good practice to install inside a virtual environment:

python3 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install git-sage

Install the pre-push hook​

Navigate to any git repository and run:

git-sage install

This writes a small shell script to .git/hooks/pre-push that calls git-sage review --hook on every push.

Verify everything is set up:

git-sage status

Expected output:

  git-sage  v0.1.0
  [✓]  Ollama running at http://localhost:11434
       Models: qwen2.5-coder:7b
  [✓]  pre-push hook installed

Uninstalling​

To remove the hook from a repo:

git-sage uninstall

To uninstall the tool entirely:

pip uninstall git-sage