← Back to Discover

PR-Agent

Open-source AI code review agent that describes, reviews and suggests improvements on pull requests and answers questions about them, using the model provider you choose.

Code reviewCommand-line tools
Visit website ↗

What it does

PR-Agent is an open-source, AI-powered code review agent. It runs as commands on a pull request, such as /describe, /review, /improve and /ask, or from the command line. An issue-scoped /similar_issue command looks for similar issues in a repository. Each tool makes a single LLM call, which the project says takes about 30 seconds at low cost, and a PR compression strategy lets it handle both small and large pull requests. Review categories and behaviour can be customised through configuration files using JSON-based prompting. It works with GitHub, GitLab, Bitbucket, Azure DevOps and Gitea, and can run as a GitHub Action, from the CLI, in Docker, self-hosted or through webhooks. Supported models include OpenAI GPT, Anthropic Claude, Google Gemini, DeepSeek and Mistral, plus anything reachable through LiteLLM, such as Azure OpenAI, AWS Bedrock, Vertex AI, OpenRouter and Ollama. The project describes itself as community-maintained, a legacy project donated by Qodo, and separate from Qodo's own code review offering. It is released under the MIT license.

Capabilities

  • /describe, /review, /improve and /ask commands on pull requests
  • Runs as a GitHub Action, from the CLI, in Docker or self-hosted
  • Supports GitHub, GitLab, Bitbucket, Azure DevOps and Gitea
  • Works with OpenAI, Claude, Gemini, DeepSeek, Mistral and Ollama
  • Review behaviour customised through configuration files
  • Processes both small and large pull requests

When to use it

  • You want automated AI review comments on pull requests in your own CI
  • You want to self-host and choose your own model provider and keys
  • You want to ask free-text questions about what a pull request changes