LIVE MVP LAUNCH • BUILT FOR WINDOWS 10/11

Say what you mean.

Get it done at the speed of thought.

Local-first ambient voice dictation and context-aware copilot. Hold Alt (Left or Right) anywhere on your PC — speak naturally, and watch text paste instantly at your cursor without ever stealing window focus.

<ms Latency
Groq Whisper-large-v3
Native cursor injection
+ Languages
Automatic detection
% Privacy Guarded
Zero audio storage

INTERACTIVE SIMULATOR

Feelthespeedbeforeyouinstall.

Simulate QEVRA’s floating obsidian capsule across your favorite Windows apps. Hold to speak, release to paste.

cursor — auth.middleware.ts
// Composer Prompt (Ctrl+I)
1
2
Mode:

CORE CAPABILITIES

Engineered as a native Windows power-tool. Not another browser wrapper.

Wispr Flow Native

Hold-to-Talk (Sub-180ms Trigger)

Press and hold Right Alt (or F8 / Fn) for ≥180ms. The non-activating obsidian capsule floats at the top without stealing window focus.

Release the key to transcribe via Groq and paste directly at your exact cursor position across any Windows application.

Cursor IDEVS CodeGoogle ChromeSlackDiscordNotepadWindows Terminal
380ms Window

Double-Tap Persistent Mode

Double-tap Right Alt within 380ms to lock the HUD open for long-form dictation, meetings, or multi-step prompt drafting.

Features interactive on-screen controls for Record, Pause / Resume, Stop & Insert, and Discard.

Dual Architecture

Direct Dictation vs. Smart Assistant

DIRECT MODE (VERBATIM)

Pure verbatim transcription with strict capitalization, punctuation, and code snippet formatting. Zero hallucinations.

SMART COPILOT MODE

Contextual prompt engineering designed specifically for Claude 3.7, ChatGPT, and Cursor Composer workflows.

Sub-180ms STT

Groq Cloud Whisper-large-v3

Inference powered by Groq LPU™ systems running Whisper-large-v3 at sub-180ms speed.

Uses greedy zero-temperature decoding to eliminate hallucinations on background noise, with optional secondary Llama 3.3 filler-word suppression (strips “um”, “uh”).

Strict HITL Gate

Autonomous Screen Copilot & Safety

Multimodal visual inspection of active windows via low-latency native screen capture.

HUMAN-IN-THE-LOOP SAFETY GATE

Actions with side effects require explicit approval: press Enter to execute or Esc to reject.

Win32 Hooking

Global Windows Hotkey Architecture

Low-level native C++ hook (WinKeyServer.exe) captures hardware key presses without invoking PowerShell or triggering Windows UAC freezes.

Customizable bindings with support for Right Alt, F8, Fn, Right Ctrl, and Ctrl + Shift + Space overlay toggles.

DOWNLOAD & ONBOARDING

Zero-friction setup. Running in under two minutes.

Download the pre-compiled native installer, connect your free Groq key, and begin speaking anywhere in Windows.

QEVRA for Windows

MVP

v0.1.0 · ~170 MB · Windows 10 & 11 (x64)

Global CDN Mirror (GitHub Releases)
All Releases & Mirrors →
Beta Note: Because this is an unsigned community MVP build, Windows SmartScreen may show an initial prompt. Simply click More info → Run anyway.

Groq powers the ultra-low latency Whisper-large-v3 model. The free tier provides generous daily limits with no credit card required.

  1. Visit the Groq Console and sign up with GitHub or Google.
  2. Click API Keys → Create API Key.
  3. Copy your key (starts with gsk_).
Open Groq Console Keys

KEYBOARD ARCHITECTURE

Global Shortcuts Reference Table

Zero-delay low-level Win32 hardware hooks. Works across any active window without stealing caret position.

HoldRight Alt≥180ms
Transcribes audio on release and pastes directly at active cursor without losing focus.
Double-tapRight Alt≤380ms
Locks the floating capsule open with manual Record, Pause, Stop, and Discard controls.
CtrlShiftSpace
Summons the full workspace overlay, intent history, memory hub, and settings panel.
Esc
Instantly aborts in-progress recording without pasting, or closes active modals.
123
Selects numbered clarification pills when the AI detects ambiguous instructions.

PRODUCT ROADMAP

How we are moving forward.

From today’s live MVP to a fully autonomous, offline ambient intelligence platform.

PHASE 1 — TODAYLive MVP

Local-First Windows Hold-to-Talk & Context Engine

Full native Windows release with sub-180ms Groq Whisper-large-v3 transcription, zero-focus cursor pasting, and direct verbatim dictation.

Hardware-hooked Right Alt (WinKeyServer.exe)
Win32 foreground window preservation
Direct Dictation & Smart Copilot modes
Human-in-the-Loop safety approval system
PHASE 2 — NEXTIn Active Build

100% Air-Gapped Local Whisper (ONNX + Transformers.js)

Zero internet connection required. Run compact Whisper models entirely on local GPU/NPU hardware with quantized weights.

Local ONNX Runtime execution
Direct NVIDIA TensorRT & Intel NPU acceleration
Zero egress network guarantees for high-security environments
Local vector RAG semantic memory indexing
PHASE 3Planned

macOS & Linux Companion Architecture

Cross-platform parity with global accessibility hooks, native Wayland/X11 support, and per-app prompt templates.

macOS Apple Silicon (Metal) native binary
Application-specific prompt presets (e.g. Cursor vs Slack vs Word)
Global multi-monitor capsule positioning
PHASE 4Planned

Team Dictionary & Enterprise Vocabulary Sync

Collaborative domain dictionaries for engineering organizations, medical terminology, and proprietary codebase symbols.

Real-time synchronized custom vocabulary
Enterprise audit logging and compliance verification
Self-hosted team dictionary relays

PRIORITY ACCESS

Join the QEVRA Pro & Mac Beta Waitlist

Be the first to access on-device offline Whisper models, team custom dictionaries, and cross-platform companion releases. All submissions are indexed into an Excel-ready database.

Platform Interest

FREQUENTLY ASKED QUESTIONS

Straight answers from the builders.