InkvoxIntroducing Inkvox
Tap a key, talk, and your words land in whatever app you're using — transcribed by Whisper on your own GPU, never in the cloud.
You think at roughly 150 words per minute. You type at maybe 40. That gap — between what's in your head and what ends up on screen — is where ideas get shortened, emails get blunter than you meant, and notes never get written at all.
Dictation should close that gap. So why does almost every dictation tool ship your voice to someone else's server?
Local first, local only
Inkvox runs OpenAI's Whisper speech-to-text model directly on your machine, on your own GPU through Vulkan — NVIDIA, AMD or Intel. Your audio is processed in memory, on device, and the transcript is typed straight into the app you're using. Audio uploaded: 0 bytes. No account, no subscription required for the core app, no telemetry on what you say.
The whole pitch: one key. Tap Right Ctrl, speak, tap again — your words appear where your cursor is. Claude Code, Gmail, Notion, WhatsApp, your IDE: if you can type there, you can dictate there.
Fast enough to disappear
On-device used to mean slow. It doesn't anymore: on a mid-range GPU (think RTX 3070), Inkvox transcribes a sentence in around 0.3 seconds with Whisper large-v3-turbo. That's faster than the round-trip to most cloud dictation APIs — without the network dependency. No GPU? Inkvox falls back to a lighter model on CPU.
Built for the tools you live in
Inkvox isn't another app you switch to. It's a small pill that idles in your tray, written in native Rust, and wakes when you tap the key. Dictate a prompt into your AI agent, a reply into WhatsApp, a paragraph into Google Docs — the pill follows you everywhere, and the words land at your cursor.
What's in the beta
- Free beta for Windows 10/11 — join the waitlist for access; one ~800 MB model fetch on install, then fully offline.
- Any Vulkan-capable GPU — covers most machines from the last decade.
- Inkvox Pro is coming later, with power features on top. The private core stays free.
- macOS is next — same codebase, Metal backend — once the Windows beta hardens.
Why "Inkvox"
Vox is your voice. Ink is what it becomes. The whole product is the arrow between the two — as short, private and fast as we can make it.
If that sounds like how dictation should have worked all along, join the waitlist and tell us what breaks: contact@fortix.fr.