PR-Agent

PR-Agent is an open-source AI code review agent maintained by the PR-Agent community organization on GitHub. It describes, reviews and improves pull requests, and can be run from a PR comment, a webhook, CI or the command line. The README says it began as a Qodo project that Qodo has since donated to the open-source community.

What It Is

PR-Agent is a pull request assistant that works through commands such as /describe, /review, /improve and /ask. Each command runs against a pull request and posts results back as a comment, for example a reviewer guide with an estimated review effort, a note on whether the PR contains tests, security observations and recommended focus areas. An issue-scoped command, /similar_issue, finds similar issues in a repository.

How It Runs

The README documents several ways to run it: a GitHub Action workflow, the pip-installed CLI (pr-agent with optional git provider extras), a Docker image published as pragent/pr-agent, self-hosting, and webhooks. Setup guides cover GitHub, GitLab, Bitbucket and Azure DevOps, and the site lists Gitea as supported. Users supply their own model API key and a git provider token.

Models and Customization

The README states that each tool uses a single LLM call and that a PR compression strategy handles both small and large PRs. Models include OpenAI GPT, Anthropic Claude, Google Gemini, DeepSeek and Mistral, plus any model reachable through LiteLLM such as Azure OpenAI, AWS Bedrock, Vertex AI, OpenRouter and Ollama. Review behavior can be customized through JSON-based prompting and configuration files.

Current Status

The README notes that /help_docs has been temporarily disabled since v0.36.1 pending a fix for a credential-exposure issue. It also distinguishes this community project from Qodo's own AI code review offering, and says Docker images from 0.34.2 onward are published under the pragent/pr-agent namespace.