Built quietly,
on purpose.
VoiceType is software for people who type for a living and would rather speak. Made by a small team — okay, one person — who lives inside a terminal and got tired of the friction between a thought and a command-line argument.
In 2025 the best voice-to-text on macOS still required either a cloud subscription that uploaded every word to someone else's GPU, or a stack of Python and CoreML files that took a weekend to wire up. Neither felt right for a tool you'd use forty times an hour.
Press ⌘ ⌘. Speak. Paste.
Audio stays on the machine.
VoiceType is the version that doesn't ask permission to think for you. It runs locally. It doesn't keep your audio. It works in the terminal, in Xcode, in a Cmd-Tab away in Slack. It costs less than three coffees a month and refunds you for a month if it doesn't earn its keep.
Five rules
we don't break.
Audio stays on your Mac.
Not "we have policies." Not "we delete it eventually." We don't have an endpoint that receives audio. Inspect the network traffic — the only outbound calls in default operation are model downloads from Hugging Face.
No telemetry by default.
Funnel events are off until you opt in, and even then, only event names and plan/channel — never your transcript content. Your dictation is none of our analytics' business.
Real prices on the website.
$8 / $72 / $149. The same numbers in your Stripe receipt. The same numbers on your App Store invoice. No "starting at" asterisks, no "talk to sales," no surprise upgrades.
30 days, no questions.
If it doesn't earn its keep in a month, email us, we send it back. The policy is in the refund page, written like a contract not a marketing pitch.
Your data is yours when you leave.
One email and we delete or de-identify what we can. We're a small company; this isn't an enterprise compliance theater move — it's just the right thing.
A small operation,
by design.
Boring tech, well-chosen.
SwiftUI for the macOS app. WhisperKit for transcription. llama.cpp + Qwen 3.5 4B for local LLM polish. Apple's Speech framework as a strict on-device fallback. Sparkle with EdDSA signatures for updates. Stripe for billing. Vercel + Upstash for the small server-side state we keep. No vendor lock-in we can't walk away from in a weekend.
WhisperKit (MIT, by Argmax). Whisper model weights (MIT, by OpenAI). Qwen 3.5 (Apache 2.0). llama.cpp (MIT). Sparkle (MIT). The shoulders are tall and we're grateful.
Talk to us
anytime.
Real human at the other end. We answer in days, not weeks.