Origin Story — UnscriptedByKraMy¶
Before Everything¶
Two or three months before I wrote any code for this, I was looking for a username. I settled on UnscriptedByKraMy. I didn't want a name that represented what I knew at the time, because I was just starting out. I wanted a reminder to be honest about where I was starting from—to learn publicly, grow publicly, and share the actual journey instead of pretending to already be an expert. I didn't know it yet, but that mindset ended up defining how this whole project came together.
Trying to Learn AI¶
I wanted to learn AI properly and get my hands dirty, but the ecosystem felt massive. I didn't want to get stuck in an endless loop of just consuming content. My first idea was to use ChatGPT Projects to build a personal knowledge base, kind of like a second brain. But to do that right, I realized I needed to learn proper software engineering practices first. That meant learning Git.
Refusing to Quit¶
I quickly hit a wall. Git didn't feel like the right tool for building a personal knowledge base. Usually, this is the exact point where I would drop the project and move on—a bad habit of giving up at the very beginning. This time, I made a conscious decision not to quit. If Git was the wrong tool for a knowledge base, I needed a real project that actually required it so I could keep going. That led to the idea for an AI-powered iPhone storage cleanup app. The goal was simple: build something useful, learn Git, use AI to help me code, and spend as little money as possible while I figured things out.
The Subscription Roadblock¶
To build the app, I started using Codex as my first AI coding assistant. I hit the basic subscription limits almost immediately. It wasn't necessarily that I couldn't afford paid tools, but I hadn't explored the ecosystem enough to know what was actually worth paying for. I wanted to understand the space first—how different models actually worked, what routing was, context windows, and how real engineering teams used these tools in production.
Getting Sidetracked by Infrastructure¶
At my office, I noticed engineering teams were starting to adopt Claude and were actually shipping real projects with it. Around the same time, I stumbled across a social media reel showing how to use Claude Code alongside OmniRoute and free providers. I saw it as an opportunity. It was a practical way to keep building the iPhone app and explore the ecosystem without immediately committing to multiple expensive subscriptions.
But the moment I set it up, the iPhone app took a back seat. Instead of just learning how to use an AI assistant, I accidentally started learning Docker, gateway architecture, and how to read session transcripts.
The Lie in the Statusline¶
While digging through those gateway logs to see how things were routing, one specific detail kept bothering me. Claude Code's built-in statusline kept displaying placeholder labels like auto/cheap.
The statusline was lying by omission. It showed what I requested, not what actually happened. If a request was routed to a specific model by the gateway, I had no way of knowing which one actually answered unless I manually dug through the local session transcripts or the gateway logs.
auto/cheap is what you asked for; opencode/big-pickle is what answered.
I just wanted to know which model was actually doing the work.
Finding the Truth¶
I originally expected this to be a small fix, but the investigation quickly became much larger than the feature itself. Where is the truth actually stored? Can it be displayed reliably? How should the statusline fail gracefully if the data isn't there?
I realized that the reliable answer lived only in the local session transcripts and the gateway's call logs. To actually get that information, I had to trace the data myself, verify it locally, and figure out how to pull it safely. I didn't want to guess. I just wanted the tool to tell me the truth about what happened, rather than what I originally requested.
AI Engineering Lab¶
Once I finally got the statusline working for myself, I realized I probably wasn't the only person entering the AI space this way. Other people likely had the same constraints, the same questions, and the same desire for transparency.
That realization is why this repository exists. The AI Engineering Lab isn't just a place to host finished projects. It exists to preserve the thinking, the investigation, and the messy chain of events that produced them. The path went from wanting to learn AI → to Git → to an iPhone app → to Codex → to Claude → to OmniRoute → and finally to a statusline.
None of those later steps were part of the original plan, and that is exactly the point. It's okay that the project completely changed direction, because the goal was always to learn honestly and share the reality of how things get built. That's exactly what UnscriptedByKraMy was meant to be in the first place.