As a developer, writing code has never been easier. I'm shipping more and more with the tools I have.
But honestly, my trust in them is wearing thin. A day later, I can't even remember which prompt caused which change.
That's why I started running everything in sandboxes. Then the agent would tell me it "read x file," and I'd find other files had changed too.
I'd stare at my screen wondering why it reached that conclusion, or why it decided to do x, y and z.
So I started recording it
I wanted a record of what the agent said it did through its tool calls, side by side with what actually happened in the sandbox:
- every command and its real exit code
- every file change
- every outbound network call
If it touches my .env, I want to know. I don't want my API keys ending up in front of an LLM.
That's RayTrace. It records every prompt and every tool call, plus what was actually observed underneath.
It's open source at github.com/Godcreatebugs/raytrace-cli. It runs locally for free, and the evidence stays on your machine.
If you'd rather look before installing anything, there's an interactive demo, no account needed.
You can try it out with take a tour button on website, or run locally and contribute to the project on Github :)
