New AI Model

I’ve been working on my open source tool that can engineer a Postman collection out of a recorded session’s HAR file.

I originally started working on this project as a way to automate the automation of tests. What I’ve seen basically everywhere is that automated testing scripts tend to be very expensive, and rather fragile things. The ability to build good tests is pretty rare. The frameworks that exist in the wild tend to be bloated disasters as well. It’s just a mess.

My idea was to be able to augment manual testers’ work by recording successful tests, then be able to play them back.

The difficulty is that in the APIs that I tend to deal with it’s not as easy as simply playing back the recording. The result of one call tends to be used as an input in a subsequent call.

My first pass got to generally working, but all of the variables used in my generated Postman collection were “var1,” “var2,” and the like. It works, but it’s not something that helps me.

I decided to throw this problem to OpenAI. And it’s amazing how good it’s gotten. I was trying the latest 4.1-nano and -mini models from OpenAI and I have to say that I’m impressed. The cost to get variable names, the JavaScript to extract the values, the generated names of the call all gets done in about a penny — which is far more cost effective than doing it manually.