This demo takes source code and asks both Jev and Codex to sort every comment into the same seven kinds, from useful context to stale or unclear. Each result shows the original code, the label, the suggested action, and how long the call took.
It does not change or run the code. The actions are recommendations. If one model fails, the other can still show a result.
It was built at the San Francisco Jev hackathon.
What you get
- TypeScript source under the MIT license
- The same comments classified by Jev and by Codex
- Timings and labels on the original code, with no edits applied






