Inspiration
Many of the contributions on our open source projects are authored in collaboration with AI coding agents. When we review such PRs, we want to know what was written by AI and what prompts were used, so that we know what the developer intended the code to do. This gives us the context that we need in order to review these PRs.
What it does
- GitHub Action that runs git-ai to pull AI authorship data for PRs, submit that data to our central server, and then update the PR comment with the review URL.
- NextJS app that runs Gradient AI to create code and prompt summaries and review the pull request. The NextJS app also serves the PR Review UI.
- Developers can click the PR description link to use the Git AI PR Review bot.
How we built it
- We extended git-ai (an existing open source project) to get all the data on prompts that the agents used and the code that they wrote.
- We build a GitHub Action that runs the extended git-ai to capture and submit all this data to our NextJS server that stores it in a Postgres DB.
- The NextJS app serves the UI and uses Gradient AI to power all of the AI features that help developers review PRs better with more context
Challenges we ran into
- Parsing git diffs is hard!
- Connecting two scroll panes is somehow even harder.
- Found a tool calling bug in Claude Code
- Encountered some issues with the Gradient Typescript SDK
Built With
- next
- rust

Log in or sign up for Devpost to join the conversation.