Liner Notes

The Hat, the Band, and a Tool Nobody Asked For

1912. A storm. Three strangers under one absurdly large sombrero. Forty minutes later, a mariachi band.

We didn't need forty minutes. A friend knew a guy. Coffee at Starbucks. We walked out with a concept and no idea it was a band yet.

That's the whole origin story. We're not going to dress it up more than that.

Here's what we keep thinking about. A song either gets hummed on a porch and travels on its own, or it sits behind a hundred-dollar ticket for people who were already coming. Same chords. Different reach. The music doesn't decide that. The door does.

Most tech has the shape of the expensive show. Impressive to the five people already fluent in it. Everyone else stands outside the venue.

We'd rather build the porch song. Something a business can pick up and actually use — no research team required, no translator standing between the tool and the person who needs it.

That's the idea. We're still tuning it.

Somewhere in working on it, a small tool fell out. Nobody asked for it. It's not the record — just a chord we played to check the strings were in tune. Down the hall, in the green room.

A Riff, Played Once, Recorded Anyway

Documents don't talk to LLMs well. PDFs come out mangled. Word docs lose their shape. Someone always ends up copy-pasting into a chat window and hoping for the best.

So — a converter. PDFs, Word, PowerPoint, Excel, straight into clean Markdown. Runs on your machine, not ours. No upload, no account, no cloud round-trip. What goes in your folder stays in your folder.

It pulls figures out into their own asset folder instead of losing them. It fixes the broken math symbols PDFs love to mangle. It handles a locked file if you hand it the password. None of this took a company to build. It took a weekend and a reason to want it.

It's free. It's open source, AGPL, code's on GitHub if you want to see how it works or make it work better. Windows only, for now.

This isn't the record. It's a riff we worked out in the green room, in the time it took the real thing to tune up. If it's useful, take it.

Why a PDF Costs More Than It Looks Like

An LLM doesn't read a PDF. It reads a PDF's description of itself — fonts, positions, spacing, decoration, a whole costume of formatting — and has to dig the actual words out from underneath.

All that digging costs tokens. Tokens cost money. They also cost context — the room a model has left to actually think, once it's done unpacking your file.

PPTs are worse. A slide isn't text with a shape around it. It's shapes, and text is one of the things that happens to live inside them. The model has to guess what belonged next to what, in what order, before it can even start reading.

Markdown skips the costume. It's just the words, in the order they matter, with nothing to strip away first.

Nobody's measured an exact saving for you here — it depends on the document, how it was built, how much junk is hiding in it. But the direction holds: less to strip out, less wasted, more room left for the part you actually wanted the model to think about.

If you've ever pasted a PDF into a chat window and watched it lose the thread halfway through — that's usually why.