Why Diffwright exists
Reviewing a pull request means gathering a lot of things. Diffwright tries to put them in one place, without deciding anything for you.
The problem
Reviewing a pull request on Azure DevOps means juggling: the diff, the discussion threads, the state of CI and branch policies, sometimes the code you need to run to check it does what it claims. Each piece is reachable, but not in the same place, and attention wears down in the back and forth.
Choosing a local cockpit
Diffwright is a desktop app, not an online service. The heavy work (calls to Azure DevOps, computing the diff, calling the AI) runs on your machine; your settings and history stay in a folder in your profile. There is no Diffwright server between you and your organization.
AI as an assistant
The AI proposes, you decide. It goes through the claude CLI that you installed, or a local model with Ollama: no built-in API key, no per-token billing. Each suggestion is approved, edited or ignored one by one, and nothing is written to Azure DevOps without an explicit click on that very item.
Where the project comes from
Diffwright follows PRSentinel, a native macOS app by the same author. The need became cross-platform (macOS and Windows): the business logic was ported to Rust, the interface rebuilt in React with the Monaco editor, and PRSentinel’s tests served as the behavioral reference.
Where the project stands
It’s a personal project, in beta. Some interactions with Azure DevOps haven’t been verified against a real organization yet, and the Windows build hasn’t been tested on a physical Windows machine. The site is upfront about that rather than promising what isn’t demonstrated.