AI Git Tools: Seamlessly Integrate Large Language Model Capabilities into Local Git Workflows

This article introduces the open-source project ai-git-tools, which seamlessly integrates Large Language Model (LLM) capabilities into local Git workflows via Git hooks. It supports multiple AI engines like Gemini, Ollama, and Codex, providing AI code review and intelligent commit message generation features to help developers identify potential issues before submission and improve code quality.

ai-git-tools was created by developer lihom. Its core positioning is to bring LLM capabilities into local Git workflows, emphasizing local execution and developer control. Unlike AI assistants that rely on cloud services, it supports multiple AI engines and automatically triggers auxiliary functions at key code submission nodes via Git hooks—without disrupting the development rhythm while resolving issues in advance.

AI Code Review (pre-commit hook)

Automatically analyzes staged code changes during submission, reviewing from three dimensions: bug detection, security risk scanning, and code smell identification. Results are graded by severity; Critical or High-level issues will automatically block the submission (can be bypassed with git commit --no-verify).

Intelligent Commit Message Generation (prepare-commit-msg hook)

Analyzes git diff changes, generates commit messages compliant with the Conventional Commit specification, and pre-fills them into the editor. Developers can modify and confirm, promoting consistency in team submission standards.

The project supports three mainstream AI engines:

• Gemini: Install via npm with @google/gemini-cli, uses the gemini-3-flash-preview model by default;
• Ollama: Localized solution that can run open-source models like gemma3; sensitive code never leaves the local environment;
• Codex: Via the @openai/codex package, uses the gpt-4o model. The multi-engine design respects developers' choice; a unified abstraction layer shields underlying differences, making engine switching simple.

Installation: Clone the repository and run the setup.sh script to complete Git hook installation; Configuration: Set AI_ENGINE (options: gemini/ollama/codex) and corresponding models via the .env file; supports custom prompt templates (REVIEW_PROMPT_PATH, COMMIT_PROMPT_PATH); Privacy Security: The Ollama engine processes code locally; cloud engines only send code snippets during submission. The project is open-source under the MIT license, with transparent code and preserved human decision-making control.

Suitable scenarios:

• Individual developers: Improve code quality and make up for the lack of review partners;
• Small teams: Replace manual reviews and save resources;
• Educational scenarios: Help learners master good coding practices;
• Enterprise intranets: Support use in isolated environments via Ollama. Tool value: Lightweight integration without adding cognitive load; helps develop good submission habits and enables collaboration between AI suggestions and human decisions.

ai-git-tools demonstrates the possibility of combining LLMs with the development toolchain. It does not replace developers but provides intelligent assistance at key stages. Through lightweight integration via Git hooks, it helps developers improve work efficiency and code quality. As LLM capabilities continue to advance, such tools will play a more important role in software development workflows.