Kodus

Wanted to share a bit about what we’re building at Kodus — an open source, AI-powered code review platform designed for dev teams that care about quality but don’t want to slow down delivery.

Most of us have seen it by now: using LLMs alone for code review doesn’t really work. You end up with out-of-context comments, incorrect suggestions, or feedback that sounds smart but has nothing to do with your actual code. It creates more noise than value.

So we took a different approach. We open-sourced Kodus, a code review platform that combines AI with a deterministic, AST-based engine. Instead of letting GPT “guess,” we feed it precise, structured context extracted directly from the code itself. The result?
Way less noise, fewer hallucinations, and suggestions you can actually trust (and safely merge).

Here’s a quick rundown of what we’ve built:

• Hybrid engine (AST + GPT): We use a deterministic rule engine to analyze code and send only the relevant context to the LLM. This drastically cuts down on false positives and irrelevant feedback.

• Open source & self-hostable: Run it on your own infra. No code leaks, no data privacy headaches.

• Customizable rule engine: Define the rules that make sense for your team, and share them with the community. Fully adaptable to your context.

All of this runs directly in your Git workflow (GitHub, GitLab, Bitbucket, Azure DevOps), without disrupting how your team already works. The goal is to help you keep review quality high — without burning out your team or relying on processes that don’t scale.