open-code-review: Precise Code Review with 1/9th the Tokens

An AI code review tool, validated internally at Alibaba by tens of thousands of developers and millions of defects, has been released as open source. Going beyond simple diff analysis, it understands full file contents and codebase context to generate precise line-level review comments.

To address the issues of missed reviews and location errors common in general-purpose agents, it adopts a hybrid structure combining deterministic engineering with agents. File selection and rule matching are enforced by logic, while dynamic judgment and context retrieval are delegated to agents to ensure stability.

Compared to the same model baseline, it reduces token usage by approximately 1/9th while significantly improving Precision and F1 scores. By prioritizing precision in benchmarks based on real open-source repositories, it minimizes unnecessary noise.

In addition to Git diff reviews, it supports full-file scanning, making it useful for auditing unfamiliar codebases. It integrates with major coding agents such as Claude Code and Cursor, and can also delegate reviews to the host model via Delegation Mode.