The code shipped.
The thinking vanished.
I'm fixing that.
Prompture exists because I got tired of my own git history lying by omission. This page is the short version of why, and where to follow the long version.
Git remembers half the work
I write code every day, most of it with Claude Code, Cursor, and a rotation of CLI agents. One evening I scrolled back through a repo and realized the commits told me what changed, but nothing about how I got there. The prompts, the rejected approaches, the constraints I gave the AI, all of it was gone. The most valuable part of the work had evaporated the moment each commit landed.
Nobody can prove they're good at driving AI
Every resume now says "uses Copilot" or "AI-native developer". None of them can show it. Hiring managers ask to see how you actually drive an agent, and the honest answer is a shrug. Teams adopting AI have the same blindness at scale: they know AI wrote a chunk of the codebase, but not which chunk, driven by whom, or how well. Compliance teams are starting to ask questions nobody can answer.
Six signals, one link
Prompture captures prompts from every surface I work in, CLI agents, IDE tools, web chats, and matches each one to the git commit it produced. Matching sounds trivial and isn't. Prompts arrive late, commits land out of order, sessions span branches. The engine scores six independent signals before it claims a link, and the signal that ended up carrying the accuracy was not the one I expected. I write about these details as I build.
Built nights and weekends, in public
I have a day job. Prompture gets built in the hours around it, and I post the engineering as it happens: the matcher internals, the git-hook edge cases, the bugs that never threw errors. If you want the unfiltered version of how a solo founder ships a dev tool with AI, that's the feed.
Follow the build
Engineering stories as they happen, roughly daily. And if you try Prompture and tell me honestly where it falls short, you are doing me a favor.