Voice-to-text anywhere on Windows

I type a lot of text that starts as thought and dies a little on the way to the keyboard. The tools that fix this exist and are good, and I wanted to know what it takes to build one, so I built Quill in a weekend: hold a key, talk, release, and clean text appears in whatever application has focus.

The interaction

The entire interface is one gesture. Hold Right Ctrl and a small pill appears with a live waveform. Speak. Release. The pill disappears and the text pastes itself where your cursor already was: an email, a code editor, a chat box. No app to switch to, no window to manage. Push-to-talk turns out to be the correct model, because the hotkey doubles as the boundary of the recording. Nothing listens unless a finger is down.

The pipeline

Release triggers a three-step pipeline. The audio goes to a speech-capable model for transcription. The raw transcript then makes a second pass through a small cleanup model with one job: remove the filler, fix the punctuation, keep the intent and the voice. The difference between raw dictation and the cleaned pass is the difference between something you must edit and something you can send. Then the result is pasted into the focused window as if typed.

What took the time

Not the AI. The desktop parts: a global hotkey that never steals focus, an overlay that floats above full-screen apps without becoming a window you have to manage, and paste behavior that works across ordinary and stubborn applications. Desktop software is a hundred small physics problems. The models were the easy third of the weekend.

Why it is on this site

Because it is the smallest honest example of what I sell. A human was doing repetitive translation work, thought to text, and now a system does it, invisibly, inside the tools they already use. The pattern scales from one keyboard to an entire back office.