Built in the open
Why I'm building this
I was driving, dictating a text to my wife with Wispr Flow, one of the paid voice-to-text apps everyone recommends. What came back was funny, but it wasn't what I'd said. I'd already been through a free option that defaulted to English no matter what I did, and Willow before that: both genuinely capable apps, but neither one could hold Arabic, Malay, and Singlish together in one sentence, which is just how I actually talk, and how the people around me talk too.
I work in IT, and this is the era where building the tool yourself instead of waiting for someone else to ship it is genuinely possible. Not from scratch: Nasar Flow runs on whisper.cpp, entirely on-device, standing on real research and real open models, credited in full on the credits page. The idea was never to reinvent speech recognition. It was to point the pieces that already exist at a problem nobody else seemed to be pointing them at.
I don't have a Mac. Every build, every test, every TestFlight upload happens on GitHub Actions runners in the cloud, because that's what was actually available to me, not because it's the proper way to build an iOS app. If that sounds like a workaround, it is. It's also proof that not having the ideal setup isn't a reason to wait.
The source is public, MIT licensed. It'll have rough edges, because a solo build always does. That's exactly what this beta is for.
Aarif, Nasar Pte Ltd