OmniVoxtryomnivox.com

Speak your intent. Ship a prompt.

OmniVox is a local-first voice workflow — a faster way to capture ideas, instructions, and workflows. It listens, cleans, and types, all on your own machine. Whisper for speech, a local Qwen3 for structure, output shaped for the app you're aiming at.

LiveOpen OmniVoxSee it work

Local-first · Windows · macOS soon · v0.4.0

Flat isometric illustration: a microphone's waveform dissolves into pixel squares and lands as a row of keycaps on a document plane

Fig.01 — Voice in, keystrokes out

Local Whisper · Qwen3

Live dictation

Hold a key. Talk. It's typed.

The floating pill opens on your hotkey, shows the live waveform while Whisper listens, cleans the transcript, and keys the words into whatever window has focus — no cloud round-trip in between.

Deterministic demo · loops

Focused field · login.tsidle

Press Ctrl + Alt and talk — your words land right here.

Fig.02 — Dictation, end to end

Ported app UI

How it works

From thought to shipped in four keypresses.

The whole flow is one hotkey, your voice, and an optional word that turns plain dictation into a structured prompt. Nothing else to learn.

  1. 01Ctrl + AltPress the hotkeyOne global hotkey, remappable to any one- or two-key combo. It works from any window.
  2. 02speakTalk naturallyWhisper transcribes on your device, biased by what's on your screen so file paths and identifiers land verbatim.
  3. 03voxifySay “voxify” (optional)A local Qwen3 classifies intent and slots your words into a clean, agent-ready prompt. Skip it for plain dictation.
  4. 04↵Lands in your appType-simulation keys it into the focused field, then restores focus. No copy-paste dance.

Fig.03 — Press · speak · voxify · land

Structured Mode

Rambling thought, agent-ready prompt.

Say “voxify” at the end of any thought. A local Qwen3 recognises whether you're asking to implement, explore, or get advice, and slots your speech into a grammar-validated prompt — file paths intact, fillers gone, invented filenames refused.

  • Slots stay slots — no drift into prose
  • Files must appear in your speech, or not at all
  • Only structures when you ask; otherwise it's clean dictation

Fig.04 — Voxify — speech to structured prompt

Ported app UI

Built to fit your work

Context modes

A different brain for every domain

Four seeded modes ship with the app — and you can build your own, each with its own vocabulary, dictionary, snippets, writing style, and app bindings. OmniVox switches the moment you switch windows.

  • Programmingdev dictionary · file-path biasAuto
  • Business & Salessales vocabulary · formalAuto
  • Medicalclinical terms · formalAuto
  • Legallegal terms · citationsAuto

Vocabulary

Teach it your words, three ways

Bias Whisper toward the terms you use, post-correct phonetic mishears, and expand short phrases into full text. Scope each entry globally or to one profile.

Vocabularyseed Whisper

OmniVoxClaude CodeQwen3verifyJWT

Dictionaryfix mishears

omni vox → OmniVoxtow ree → Tauri

Snippetsexpand phrases

ty → thank youtldr → too long; didn't read

Voice commands

Speak the keystroke

A short, focused command vocabulary that turns into real keystrokes. Toggle each command independently — no accidents, no surprises.

  • “new line”Shift + Enter
  • “delete last word”Ctrl + Backspace
  • “new paragraph”Shift + Enter ×2
  • “send”Enter ↵

Local-first

Runs entirely on your machine

Whisper and a small Qwen3 run on your hardware with optional Vulkan or CUDA acceleration. Air-gap the laptop and OmniVox still works — nothing leaves the device.

whisper.cppqwen3 · 1.7Bvulkancuda99 languages
no cloudno API keysno telemetry

Speed

4×

faster than typing

You think at ~400 words a minute and speak at ~180, but type at 45. OmniVox closes the gap between a thought and the prompt you meant to write.

Words per minute · higher is faster

Thinkthe ceiling — faster than any input
400wpm
Speak · OmniVoxnatural speech, captured locally
180wpm
Typethe keyboard tax on every idea
45wpm

What changes when nothing leaves your machine

CapabilityOmniVoxCloud dictation
Runs entirely on your deviceYesNo
Works offline / air-gappedYesNo
No account or API keysYesNo
No monthly subscriptionYesNo
Structured output for agentic toolsYesNo
Per-app context profilesYespartial
Screen-context bias for codeYesNo
Voice commands → keystrokesYespartial

Based on public docs of major cloud dictation tools

Flat isometric illustration: a compact computer tower with its side panel open, a microphone and two model chips inside, and an unplugged cable coiled beside it

Fig.05 — Everything happens in the box

Offline · air-gap OK

Every app you use

Lands cleanly in every app.

OmniVox simulates keystrokes the moment a transcript is ready, so your words land in the focused window — clipboard, type, or both, with focus restored after.

Shaped for

  • Claude Code
  • Codex
  • Qwen3 · local

OmniVox · structured

goal
files
urgency
type-simulation

Focused window

Add a rate limiter to the login route — cap it at five attempts a minute.

keyed in · focus restored

Lands in the focused window of

  • Claude Code
  • Cursor
  • VS Code
  • Terminal
  • Codex
  • Notion
  • Linear
  • Obsidian
  • Slack
  • Gmail
  • Google Docs
  • ChatGPT
  • Word
  • Telegram
  • + any text field

Fig.06 — Type-simulation into the focused window

Clipboard · type · focus-restore

The overlay, up close

Fig.07 — Right-click — modes and quick toggles

Ported app UI

Hold Right Ctrl · speak a command

Command

Fig.08 — Command Mode — speak an action

Hold Right Ctrl

The interface

The OmniVox desktop app — dictation profiles, vocabulary, and structured-mode settings

Fig.09 — The OmniVox app

tryomnivox.com

Get OmniVox

Get OmniVox

Your voice. Your machine.

No cloud round-trip, no API keys, no telemetry. Whisper and Qwen run on your hardware — air-gap the laptop and OmniVox still works.

Open OmniVoxBack to the bench

tryomnivox.com · Windows 10 / 11 · macOS soon