AI Code Reviewer
github.com
- Free plan
- Open source
- Linux
- Self-hosted
What AI Code Reviewer is
AI Code Reviewer is a free, MIT-licensed tool that automatically reviews GitHub pull requests. It runs as a GitHub Action added through a workflow file, without requiring a server or checkout step. Reviews produce a sticky pull-request summary and inline comments anchored to changed lines, with severity, explanations, and optional one-click suggestion blocks. Teams can choose Anthropic, OpenAI, local Ollama, or OpenAI-compatible endpoints. Settings include severity thresholds, file inclusion and exclusion patterns, incremental reviews, and a default maximum of 20 inline comments. A Markdown rules file can add house style or framework conventions. The action reads diffs through the GitHub API and does not check out, build, or execute pull-request code. Provider keys use GitHub secrets and are masked in logs. With Ollama on a self-hosted runner, code and diffs stay within the user’s network. Reviews skip public-repository fork pull requests and, by default, changes affecting more than 50 files.
Who it suits
It suits teams wanting a self-hosted GitHub review action with model-provider choice and configurable rules. Teams should understand GitHub workflow permission and fork limitations.
Good to know, and what to watch
- Runs automatically as a GitHub Action
- Supports hosted and local model providers
- Adds inline findings and suggestion blocks
- Allows custom Markdown review guidelines
- Keeps Ollama processing within your network
- Public fork pull requests are skipped
- Reviews skip changes over 50 files
- Inline comments default to a 20-comment cap
- Model-provider usage can add cost
Plans
Every plan and priceFree
Free plan
The verdict
AI Code Reviewer offers configurable, self-hosted pull-request analysis with several model choices and privacy controls. Its file-count, comment, and public-fork restrictions matter when designing a review workflow.
What the maker publishes
- Chrome extension details
- The unrelated Chrome listing reports eight users, version 0.0.5, 346 KiB size and an update date of February 21, 2026. source
- Configuration
- Settings can be supplied in .github/ai-code-reviewer.yml or as action inputs, with action inputs taking precedence. source
- Cost example
- The FAQ estimates roughly $0.05–$0.20 per typical pull request with Claude Sonnet, a few cents with Haiku and zero model charge with local Ollama apart from compute. source
- Custom guidelines
- A Markdown rules file can provide house style or framework conventions for reviews. source
- Custom rules
- A Markdown rules_file can add repository-specific coding guidelines, and it is read from the base branch. source
- Data handling
- The project says it does not proxy requests through a service it runs; with Ollama on a self-hosted runner, code and diffs stay within the user’s network. source
- Development quality bar
- Contributor documentation specifies type checking, linting, Vitest tests, action builds and at least 80% coverage for all metrics. source
- Execution safety
- The action reads diffs through the GitHub API and does not check out, build or execute pull-request code. source
- Incremental behavior
- When incremental mode is enabled, a new push reviews only newly added commits and rerunning the same head SHA is skipped. source
- Installation
- Installation consists of adding a GitHub Actions workflow using xiaofuqing13/ai-code-reviewer@v1 and a provider secret; checkout and a separate server are not required. source
- Integrations
- Documented OpenAI-compatible examples include DeepSeek, GLM, Kimi, Together, Groq, vLLM and LM Studio through a configurable base URL. source
- Large-change handling
- For very large pull requests the reviewer caps files or chunks and reports the truncation in its summary. source
- Licence
- The repository is released under the MIT License, permitting reuse, modification, distribution and sale subject to retaining the notice and disclaiming warranty. source
- License and support
- The project is MIT-licensed and welcomes contributions; the README does not state a paid support offering. source
- Limit: G2 CodeAnt listing
- Reviews on that separate CodeAnt product mention stylistic nitpicks, cautious suggestions requiring manual adjustment and integration taking time. source
- Limit: Product Hunt name collision
- Product Hunt discussion for the unrelated product reports single-file or text-box-only review, missed async/await errors, client-side exceptions and long-code formatting problems. source
- Maker identity
- The GitHub account displays the name xiaofuqing and lists 48 repositories, six followers and one following. source
- Model providers
- It supports Anthropic, OpenAI, local Ollama, and OpenAI-compatible endpoints. source
- Model providers
- The action supports Anthropic Claude, OpenAI, local Ollama and OpenAI-compatible endpoints. source
- Notable limits
- The default configuration skips reviews when more than 50 files change and caps inline comments at 20 per pull request. source
- Package status
- The package manifest describes version 0.1.0, marks the workspace private and identifies the project as an MIT-licensed self-hosted GitHub pull-request reviewer. source
- Permissions
- The example workflow requests contents: read and pull-requests: write permissions. source
- Permissions
- The example workflow requests contents read and pull-requests write permissions. source
- Praise: G2 CodeAnt listing
- Reviews on the separately identified CodeAnt product praise accurate actionable findings, faster reviews and catching issues missed by linters. source
Checked against the maker's pages on 21 Oct 2026.