Piyush Bhuva

As a software developer I am shipping more and more through my agents. Do I trust them doing a sane work? Absolutely not! Heck, I did not even remember the prompts and corresponding changes it caused. Sad part of all this is , I am staring at a screen asking my genie(claude code or codex) a question and genie grants me a wish with all the things getting done. And this genie sometimes deletes the production database. I don't trust these agents anymore. And that is the exact reason I have built rayTrace - rayTrace is a platform that connect what AI said it did vs what actually happened at kernel level. We store these actions for you so if you want to go in past and check what your agents did you can. You then can take this evidences and fine tune your own model with evidences , since now you know the short coming of any frontier models. We have open sourced the whole thing. If you want to check that one out its here - https://github.com/Godcreatebugs/raytrace-cli/pulls
you can share the prompts, exact state of the repo with your friends, so they can spin their own sandboxes and check out if it works the same way. Weather you wanna design a harness, store evidences, get deeper understanding of how the models think and their shortcomings , we are providing tools for that. If you want to try it out there is a designed interactive demo, please do it from - https://www.raytracer.si/get-started.html . Open to any constructive feedback, hope this is helpful to