Nikita-Filonov/ai-review

What it solves

AI Review is designed to reduce noise in pull requests and speed up the code review process. It automates the identification of bugs, enforcement of coding consistency, and the generation of high-level summaries, reducing the manual effort required by human reviewers.

How it works

The tool integrates directly into CI/CD pipelines (such as GitHub Actions or GitLab CI/CD) and uses Large Language Models (LLMs) to analyze code changes. It can operate in several modes:

• Inline Mode: Adds line-by-line comments directly in the diff.
• Context Mode: Analyzes multiple files to detect cross-file inconsistencies.
• Summary Mode: Provides a high-level overview of the changes, strengths, and issues.
• Reply Mode: Participates in existing review threads to provide follow-up replies.

Additionally, it features an Agent Mode using a ReAct-style loop, allowing the AI to autonomously explore the repository using shell commands (like ls, cat, rg, and git) to gather deeper context before producing a final review.

Who it’s for

Software development teams who want to automate parts of their code review workflow across various Version Control Systems (VCS) like GitHub, GitLab, Bitbucket, Azure DevOps, and Gitea.

Highlights

• Broad LLM Support: Compatible with OpenAI, Claude, Gemini, Ollama, Bedrock, OpenRouter, and Azure OpenAI.

• VCS Integration: Works out-of-the-box with major platforms including GitHub, GitLab, Bitbucket, and Azure DevOps.

• Agentic Exploration: Ability to explore the codebase via shell commands for more context-aware feedback.

• Privacy-focused: Runs fully client-side in the CI/CD environment, sending data only to the configured LLM provider without intermediary servers.

• Cusomizable: Supports customizable prompt templates and configuration via YAML, JSON, or environment variables.