open-code-review

open-code-review

Reviews Git changes with an AI agent that reads files and searches the codebase, then writes comments tied to exact lines. It suits developers who want automated review from the command line.

Made by alibaba on GitHub. These notes are Mindket's, written from its README; the project's own files and license are in its repository.

About

What it is: Open Code Review is a command line tool that checks code with an AI model. It reads your Git diff, passes the changed files on to a model you configure, and returns review comments tied to exact lines. It began as an internal review assistant at Alibaba before it was released as open source.

Who it suits: Developers and teams who want a first pass of review before a person looks at the code. It also suits anyone taking over an unfamiliar codebase.

Industry | Who uses it | How they use it | Example
Software teams | Developers and reviewers | Run a review on a Git diff before asking a colleague | A change comes back with comments that point to individual lines
Web development | Backend and frontend engineers | Check changes against the built-in rules for XSS and SQL injection | A query built from user input gets flagged on its line
Concurrent services | Engineers on multithreaded code | Look for thread-safety and null pointer problems in a change | A shared field touched by several threads draws a comment
Legacy code | Engineers inheriting old projects | Scan whole files or folders that have no recent diff | The scan returns comments on files nobody has changed lately

If you do not work with Git or a terminal, you can skip it.

Where to use it: The tool runs where your code and a model endpoint are within reach.

• A terminal on a machine with Git
• A Git repository with changes to review
• A folder of code with no diff, using the scan command
• Any model endpoint that speaks the OpenAI or Anthropic format

Use case | Who | What they get back
Review a change | Developer | Line-level comments, with context drawn from other files
Audit unfamiliar code | New maintainer | A review of whole files or directories
Check against built-in rules | Reviewer | Findings on null pointers, thread safety, XSS and SQL injection

How to set it up: Before you start, have these ready.

• Git and a repository to review
• A model endpoint compatible with OpenAI or Anthropic
• A terminal

1. Get the project's code from its GitHub repository, which the kit's notes point to.
2. Follow the setup steps in the project's README.
3. Configure your model endpoint.
4. Run a review on a repository that has changes.

How to use it: Run the review on a repository with changes and read the comments it returns. For a folder with no meaningful diff, the scan command reviews whole files.

ocr scan

Tool | What it does
Diff review | Reads Git diffs and writes comments tied to lines
ocr scan | Reviews entire files or directories, useful when there is no diff

Example: You inherit a folder of old code and nobody has changed it recently. You run ocr scan on it. The agent reads the files in full, searches the codebase for context, and sends back review comments that each point to a line.

For newcomers and experts: If you are new to this, start small. Pick a repository you know well, make one small change, set up a model endpoint, and run a review on that diff. Compare the comments with what you would have caught yourself.

Under the hood, fixed pipeline steps work together with an LLM agent. The agent can read whole files, search the codebase and look at other changed files, so its comments draw on more than the diff text. The README says it reaches higher precision and F1 than a general-purpose agent on the same model, using about a ninth of the tokens. Those figures are the project's own claim.

Read the Git diff
Run deterministic pipeline steps
Agent reads files and searches the code
Model writes review comments
Each comment points to a line

Good to know: The kit holds notes about the project, and the code lives in its own repository.

• License: Apache-2.0
• Cost to run: needs an OpenAI or Anthropic compatible model endpoint, and the README names no price or key
• Data it reaches: your Git diffs and the files of your repository, which go to the model you configure
• Kit contents: notes only, and nothing in this kit runs by itself

On GitHub, 14 November 2026

• Stars: 50,665 (215 that day)
• Forks: 3,771
• Watching: 145
• Open issues and pull requests: 341
• Language: Go
• License: Apache-2.0
• Topics: agent, agent-skills, code-review, code-review-assistant, harness, repository-level-context
• Trending: #1 repository of the week on GitHub trending (14 November 2026)

Links

• Code: https://github.com/alibaba/open-code-review
• Website: https://open-codereview.ai/
• Releases: https://github.com/alibaba/open-code-review/releases
• Issues: https://github.com/alibaba/open-code-review/issues
• Star history: https://www.star-history.com/#alibaba/open-code-review&Date

Get the code

git clone https://github.com/alibaba/open-code-review.git

Then follow the setup in its README. Nothing in this kit runs by itself.