README.md

Two pre-commit hooks:

• llm-code-review: sends the git diff to a LLM, so that it performs a code review on it.
• llm-commit-message: sends the git diff to a LLM, so that it can suggest a git commit message

LLM Code Review

The hook can be executed at two different stages: pre-commit, or pre-merge-commit.

Usage with ollama

Add to your .pre-commit-config.yaml:

- repo: https://github.com/qdii/pre-commit-llm-code-review
  rev: v2.0.0
  hooks:
    - id: llm-code-review
      args:
        - "--llm=ollama"
        - "--ollama_base_url=https://llm.dodges.it"
        - "--model=mistral-small3.2:latest"

Usage with Gemini

Add to your .pre-commit-config.yaml:

- repo: https://github.com/qdii/pre-commit-llm-code-review
  rev: v2.0.0
  hooks:
    - id: llm-code-review
      args:
        - "--llm=gemini"
        - "--model=gemini-2.5-flash"
        - "--api_key=YOUR_API_KEY_HERE"

Usage with Mistral

Add to your .pre-commit-config.yaml:

- repo: https://github.com/qdii/pre-commit-llm-code-review
  rev: v2.0.0
  hooks:
    - id: llm-code-review
      args:
        - "--llm=mistral"
        - "--model=mistral-small"
        - "--api_key=YOUR_API_KEY_HERE"

Options

Option	Description
--llm	'ollama', 'gemini' or 'mistral'
--ollama_base_url	The URL of the Ollama server to query
--model	The name of the model to use
--api_key	The API key to use Gemini or Mistral. This overrides environment variables (GEMINI_API_KEY or MISTRAL_API_KEY).
--debug	Extra logging for debugging purposes
--temperature	Inference temperature, [0.0 - 2.0]
--retries	How many times LLM should be queried
--stage	Either 'pre-commit' (default) or 'pre-merge-commit'

Note about context

This hook sends the output of git diff --cached to a LLM and prompts it to perform a code review on it.

First, this diff can be large. If you are using Gemini, you probably have a rate-limit in place that prevents you from sending too many tokens per minute.

Second, only the diff is sent to a LLM, not the whole file. As a result, the LLM has limited context to work with. This can however be tweaked by changing GIT_DIFF_OPTS to include more context lines in the output of git diff.

LLM Commit Message

The hook must be executed at stage prepare-commit-msg

Usage with ollama

- repo: https://github.com/qdii/pre-commit-llm-code-review
  rev: v2.0.0
  stages: ["prepare-commit-msg"]
  hooks:
    - id: llm-commit-message
      args:
        - "--llm=ollama"
        - "--ollama_base_url=https://llm.dodges.it"
        - "--model=mistral-small3.2:latest"