Why I Built UnType

I Was Already Talking To My Mac

For months I dictated almost everything. Emails, client messages, briefs, even my prompts for Claude.

I used Wispr Flow first, then Willow. Between them I did thousands of dictations.

They were good. But three things kept bugging me.

Three Things Bugged Me

• The waitBoth send your voice to servers a long way from Brisbane. You let go of the key, then wait for your words to come back.
• The priceEach one wanted about A$21 a month, charged in US dollars.
• My voiceEverything I said, client work included, went off to someone else's server.

So I Built My Own

In November 2026 I started building the app I wanted. Speech turns into text on the Mac itself, so there's no round trip overseas.

When a sentence needs tidying, the text goes to Claude. Your audio never does.

I gave it what I used most. A dictionary for names, a style for each app, and a way to catch "two quotes, sorry, three quotes" so I don't have to.

Within a week it had replaced Willow on my Mac. I did 440 dictations with it in the first three days.

Then I Tried To Break It

Before selling it, I ran it on recordings of real people from around the world.

Australian, American, British and Indian accents all came out about the same. Fifteen minutes of talking turned into text in under four seconds.

The tests found six bugs. All six are fixed.

What Happens Next

UnType is in early access. I use it every day, and I read every note that comes in from the Feedback page in the app.

Something feel off? Email me at support@rudygoel.com. You'll get me, not a support team.

Try It Free For 3 Days

Built On

Speech recognition by Parakeet Ultra from Moondream, trained from NVIDIA's Parakeet and shared under CC BY 4.0, in the Core ML version by FluidInference. It runs on your Mac with FluidAudio, under Apache 2.0. Text tidying by Claude from Anthropic. None of these companies is connected to UnType.