Sage

Sage

Sage is like a senior engineer that watches your coding agent's every move.

Unlike existing review tools, Sage doesn't just look at code — it critiques your agent's responses, reasoning, and plans so you can catch issues before they're implemented.

Table of Contents

- Demo
- Why current code review tools fall short
- Automating model cross-checking
- How It Works
- Setup
- Empirical Benefits
- Roadmap
- Troubleshooting
- Architecture
- Known Limitations
- Who are we
- Contributing
- License

Demo

Why current code review tools fall short

AI review tools like Greptile and CodeRabbit are excellent at catching syntax errors, edge cases, and security issues. However, they only review code diffs and have no context on the AI conversation that produced them.

When you're coding with AI you might spend ~80% of your time planning and discussing approaches with the agent, and ~20% actually writing code. The agent suggests an implementation path, you refine it together, and only then does code get written.

If there's a fundamental flaw in the approach, reviewing the final code won't catch it. You’ll end up optimizing a suboptimal solution instead of reconsidering the entire direction.

Most issues that diff reviews catch could actually be identified earlier by reviewing the plan before any code is written.

Sage lets you review and critique that plan layer, so you can identify these issues *before* they enter the codebase.

Automating model cross-checking

Each AI model has different strengths and weaknesses.

Many developers already take advantage of this manually by running two agents side-by-side and copying one’s response into the other for a second opinion.

It works, but it's tedious and completely breaks flow.

Sage automates this process.

- Currently, Sage…