OpenCode AI Reviewer puts AI review on every pull request, iterates fixes until builds go green, and audits entire repositories. It ships as a reusable GitHub Action and installs as a Probot GitHub App, so it fits existing CI without new infrastructure.
The four modes
- Review — flags bugs, security issues and style problems on every pull request.
- Auto-fix — iterates its own patches against compiler and test errors until they are actually resolved.
- Audit — scans an entire repository and files findings as GitHub issues.
- Analyze — turns an issue into a step-by-step implementation plan before any fix is attempted.
Model support
Free OpenCode-hosted models run out of the box, and the reviewer also speaks OpenAI, Anthropic and Gemini APIs. Written in TypeScript, managed with pnpm.
Open engineering
The agentic loops and prompts are documented openly: the repository’s architecture guide (agent.md) and ADR directory explain how each review and auto-fix cycle works before you run it on your own repos. MIT licensed — see the source at github.com/nilesh32236/opencode-ai-reviewer.