Why I built Yapper.

How I got here

For most of the past year I’ve dictated almost everything. Slack messages, emails, code comments, video scripts, draft essays, reminders to myself, half the prompts I send to AI tools, the lot. I started with the free tier of a well-known dictation app, found I liked it, then started paying for it.

By the time I started building Yapper, I’d dictated over half a million words through that competitor — 522,000 to be precise — at an average of 129 words per minute. For context: even fast typists top out around 60 wpm. Voice is comfortably two to three times faster, when it works.

I used it everywhere. 114 different applications — every chat tool, every text editor, every browser tab where my cursor blinked. My longest unbroken streak of daily use was 38 days. I sent 6,604 AI prompts through it — to ChatGPT, to Claude, to internal tools — which alone makes voice dictation the biggest single productivity shift I’ve adopted in a decade.

So when I say I used it: I used it.

What started bothering me

A couple of things, really.

First, the latency. There was always a small beat between releasing the dictation key and the text appearing on screen. Maybe 600 to 900 milliseconds — fast in the abstract, but enough that voice never quite stopped feeling like a feature. I’d press, talk, release, wait, see text. The wait was small but it was always there. After tens of thousands of dictations, that quiet beat starts to grate.

Second, feature bloat. Snippets. Style transforms. Scratchpads. Custom tone presets. AI editing commands. A browser extension. Folders for past dictations. Tools I never opened, cluttering the one tool I used every minute. I was paying for a Swiss Army knife when what I actually needed was the single blade.

Third — and this snuck up on me — the asymmetry. The price was fine. But every month I was paying for a long list of features to subsidise the one I actually used. That’s the trade you make with most well-rounded SaaS, and almost always it’s worth it. For something I touched fifty times a day, though, the asymmetry started to feel pointed.

The turning point

The competitor put out a major update. New buttons. New panels. A feature for organising past dictations into folders. The dictation itself got fractionally slower in the process, which I noticed because, again, I’d done it half a million times.

I sat with that for a few weeks before realising the obvious: the thing I wanted didn’t exist. I wanted the one feature, faster, with nothing else around it. So I started building it.

What I trimmed

The shape of Yapper is what’s left after a year of careful subtraction.

Out:

• Snippets — use a system text expander.
• Style transforms — the cleanup pass handles tone in context.
• Scratchpads — use Notes.
• Browser extensions — native apps first.
• Folders, tags, sharing — your transcripts live in a timeline. Search them. That’s it.
• Meeting transcription, podcast editing, AI agents — other tools are excellent at those. We are not.

In:

• A global hotkey.
• Audio capture.
• On-device Whisper transcription.
• LLM cleanup, in the background, after the text has appeared.
• Direct text injection at your cursor — no clipboard hijacks.
• A timeline of every dictation, searchable, on your machine.
• A personal dictionary.
• Stats that show you’re getting better.

That’s the whole product.

How we made it feel instant

Most dictation tools send audio to the cloud, wait for transcription, wait for an LLM cleanup pass, then drop the text in place. That’s the 600-900 millisecond beat I was feeling.

Yapper inverts the order. Whisper runs locally and fast on Apple Silicon. The raw transcript hits the screen the moment the key releases. The cleanup pass runs in the background and silently replaces the text two hundred milliseconds later — by which time you’re already reading what you said.

Our perceived latency target is under 150 milliseconds on M-series Macs. We hit that comfortably. The full technical writeup lives at /research if you want it.

What changes with Yapper

If you’ve done the same half-a-million words through the slow tool, the difference is small but cumulative. Every dictation saves a fraction of a second of waiting and a small amount of cognitive friction. Multiply across hundreds of dictations a day and the keyboard’s muscle-memory pull starts to fade. Voice stops feeling like a feature you reach for and starts feeling like the default input.

We’re also rewriting the economics. The free tier gives you 2,500 words on signup. Refer a friend who joins, get another thousand. Refer a friend who upgrades to Pro, get a free month — same on their side. Keep referring and you’ll never pay. Or pay $15 a month for Pro and never think about it again.

Where it goes from here

Yapper ships on macOS first. Windows, iPhone, and Android are on the roadmap — and every plan upgrade activates automatically on those platforms when they launch. There’s nothing to re-buy.

If you’re on the slow tool, the door’s open. The product is itself the proof — install it, hold the key, talk, and see whether it lands the way it should.