Desktop app · macOS and Windows
Review and test with the AI you choose.
The diff, threads, testing and AI review of your pull requests, brought together in one app. Built for Azure DevOps, with GitHub and GitLab in beta.
Public beta. Nothing is written to your platform without your click.
- version
- v0.17.0 ·
- platforms
- macOS · Windows
- forges
- Azure DevOps · GitHub · GitLab
- ai
- claude · Ollama · API (optional)
Animation: a fictional pull request from the Contoso organization reviewed in Diffwright, in four steps. 1. Read the diff from the file tree, with the PR status. 2. The AI review proposes a comment, added to your pending review on a click: nothing is sent. 3. Reply to a discussion thread, then resolve it. 4. Vote, tick the checklist, then “Finish my review”: you select what goes out and confirm the publication.
features
The whole review, in one window
A readable diff, in a real editor
Changed files open in a full-fledged code editor, in File or Flow view, with blame and an on-demand explanation of the file. Images, PDFs and Word documents are rendered, not shown as raw content.
Threads, comments and votes
Read threads, reply, resolve them and vote without going back to the browser. Every publication starts with an explicit click from you.
AI-assisted review
The AI of your choice (Claude Code, Codex, Claude API or OpenAI-compatible, Ollama) analyzes the diff and proposes suggestions that you dismiss, edit or publish one by one. A chat anchored on the pull request and on your selection answers your questions.
Impact beyond the diff
Impact analysis answers “who calls this?” by examining the code around the changed lines, not only the lines themselves.
Test a pull request
The Test tab runs the pull request’s project on your machine, so you can check what it actually changes. Test feedback can include an annotated screenshot.
Team activity
Commit pace, review turnaround, CI health and signals to check: questions to ask people, never a score.
test
Test before you approve
The Test tab runs the pull request on your machine. Annotate a screenshot, attach the errors it caught and publish precise test feedback. Nothing is sent without your click.
Animation: testing a fictional pull request from the Contoso organization in Diffwright, in four steps. 1. From the PR’s Test tab, open its preview environment in an isolated test window, whose Diffwright bar counts the page errors. 2. Click Capture: the page freezes and a writing panel opens beside it. 3. Annotate the capture: a rectangle around the empty card, an arrow, a text label, a blurred e-mail address, then a comment, with the page errors attached. 4. Click Publish: the feedback lands in the PR’s published feedback, pending, then a new push suggests a re-test. Nothing is published without that click.
appearance
A workspace that fits the way you work
These are the app’s Appearance and Background settings. Try them: every app example on this page updates live.
ai
The AI you choose, under your control
By default, the AI runs through the claude CLI that you installed and signed in yourself, or through a local model with Ollama. Codex CLI and pay-per-use APIs (Anthropic, OpenAI-compatible) are optional: you bring your own key and consent service by service.
With these paid engines, automatic review stays off. If no AI is available, or if you turn it off, its features disappear from the interface instead of staying grayed out. Nothing the AI proposes is published without your approval.
trust
Your code stays under your control
Diffwright works on your code and your work accounts. Four rules govern every feature, without exception.
No write without a click
Nothing is published, voted, merged or resolved on your platform without an explicit click on that very item: never from an automatic process, never from a pre-checked batch, never as a silent retry.
Repository content is treated as untrusted
Files from the repository (launch configuration, review guidelines, tested pages) are never executed without your confirmation or displayed as HTML, and the requests sent to the AI guard against instruction injection.
The work happens on your machine
Diffwright calls your platform directly from your computer and keeps its settings and history in a folder in your profile. There is no Diffwright server between you and your organization.
Optional telemetry, off by default
By default, the app sends nothing to Diffwright. Usage statistics and error reports are only sent if you turn them on; the privacy policy details exactly what they contain.
get started
Your next review starts here
Available for macOS and Windows. Connect your account with a token from your platform: the setup assistant guides you on first launch.
v0.17.0 · macOS (Apple Silicon, Intel) · Windows 64-bit
Should
plantsstay read-only now that it is a signal?