daniel-carreon/sflow ? reverse-engineered prompt

Build me a macOS app that does voice to text system wide. I want to hold a shortcut, talk, let go, and have the words typed into whatever app I was using. It should live in the menu bar, start with the Mac, and show a small floating pill while recording with simple audio bars. When it is processing, show a loading state, then paste the transcription right back into the current text field automatically.

Please make it support two ways to record, one where I hold a hotkey to talk, and another where I can double tap Control to start and stop. Save every transcription locally with a timestamp and duration, and add a small web page on localhost where I can browse, search, and copy my past dictations. Ask for the API key on first launch instead of making me edit config files. It should feel native, quiet, and not steal focus. If you need current docs for the API or macOS permissions, look them up online.