Realized how underrated prompt versioning actually is

I recently have some LLM projects iterte and one thing that really hit me is how much time I wasted if I didn't do the right quick version. It is easy to chop the tasks and optimize them in an ad hoc, but if you circle back weeks later, do not remember what has worked, what broke or why a change has made things worse. I found coping prompts to the term and random documents, and it just doesn't scaled. Versioning entry requests feel almost like a version code: -Sie want to compare iterations side by side -you need a context why a change has been made -you have to roll back quickly if a little breaks down downstream -and ideally you want to examine tools, treat the input requests with the same discipline as code. Maxim AI, for example, seems to have a solid setup for versioning, chaining and even execution of comparisons about input requests, which honestly has the feeling where this room has to go. Would it like to know how you can all be able to handle all? Just record them somewhere, use GIT or rely on a dedicated tool?