sunmh207/AI-Codereview-Gitlab

What it solves

This project provides an automated code review system that integrates with GitLab to help development teams improve code quality and efficiency. It eliminates the manual effort of initial reviews by using large language models (LLMs) to automatically analyze code changes during merge requests or pushes.

How it works

The system operates via GitLab webhooks. When a user submits code (e.g., a Merge Request), GitLab triggers a webhook that calls the system's API. The system then sends the code diff to a configured LLM (such as DeepSeek, OpenAI, or Anthropic) to generate a review. The results are then posted back to the GitLab Merge Request or Commit as notes.

It offers two review strategies:

1. Diff-only: Analyzes only the changes provided in the diff.
2. Agentic: The LLM is granted tool-calling capabilities (like read_file and a sandboxed shell) to clone the repository locally and explore the codebase for a more comprehensive review.

Who it’s for

Development teams using GitLab, GitHub, or Gitea who want to automate their code review process and integrate AI-driven feedback directly into their version control workflow.

Highlights

• Multi-model Support: Compatible with DeepSeek, ZhipuAI, OpenAI, Anthropic, Tongyi Qianwen, and Ollama.
• Instant Notifications: Pushes review results to DingTalk, Enterprise WeChat, or Feishu.
• Customizable Review Styles: Choose from Professional, Sarcastic, Gentleman, or Humorous personas.
• Agentic Review Mode: Allows the AI to autonomously explore the local codebase using a sandboxed shell for deeper analysis.
• Visual Dashboard: A dedicated UI to track all code review records and developer statistics.
• Automated Daily Reports: Generates summaries of development progress based on commit records.