PR-Agent
Open-source tool that uses AI to describe and review pull requests and suggest improvements – self-hosted with the model of your choice.
What it does
PR-Agent reviews pull requests with the help of a language model. You trigger commands as comments in the pull request: /describe writes a title and description, /review gives an assessment with possible problems, /improve suggests concrete code changes and /ask answers questions about the changes. According to the project, /review and /ask need only a single model call. It supports GitHub, GitLab, Bitbucket, Azure DevOps and Gitea, and partly also Bitbucket Server, Gerrit and AWS CodeCommit. You can run it as a GitHub Action, via webhook, as a Docker container or on the command line. Models include OpenAI, Claude, Gemini, DeepSeek and Mistral, and via LiteLLM also local models through Ollama.
Who it suits
Development teams that want faster code reviews and can handle the setup themselves. It is especially interesting if no additional SaaS service should get access to your source code.
Data protection for businesses
PR-Agent is an open-source project you run yourself. The project does not offer a hosted service, business plans or a data processing agreement. Where your code goes depends on the model provider you choose: with your own OpenAI key, its API terms apply; with a local model, the code stays within your own network. Whether inputs are used for training and whether processing in the EU is possible is therefore settled in your contract with the model provider.
Limits
Setup, updates and operation are up to you. Qodo handed the project over to the community in April 2026 and remains a sponsor; volunteers maintain it, with no support commitments. It is not the same as the commercial Qodo product. The /help_docs command is temporarily disabled due to a security issue. AI comments do not replace review by humans.