you say
Quill writes
Every feature,
demonstrated.
A live, interactive tour of everything QuillVoice does — the floating overlay, the text pipeline, per-app profiles, licensing, models, the privacy guarantees, and the full settings window. Every mock here is built to match the real desktop app, pixel for pixel. Nothing is uploaded; nothing is faked beyond the data.
01
Capture & dictation
Press a global hotkey anywhere and speak. The overlay records your voice and transcribes it without ever stealing focus from the app you're typing into.
Ready
Listening…
Transcribing…
Busy
Global hotkey dictation Free
System-wide
Ctrl/Cmd+Space. On an OS conflict it falls back to an alternate and persists the choice; re-registers live when changed.Activation — Toggle Free
Press once to start, again to stop. The default mode.
Activation — Push-to-talk (hold) Pro
Record only while the key is held; VAD auto-stop is disabled in this mode.
Voice-activity auto-stop (VAD) Free
Stops automatically after a short pause (about 1.5 s of silence by default), with a brief grace window so it never cuts off before you start.
Whisper mode (quiet-speech gain) Pro
Boosts gain and detector sensitivity so near-silent whispering still transcribes.
Microphone selection Free
Pin capture to a chosen input; defaults to your system mic, with automatic fallback if it's unplugged.
Mic audio ring / waveform Free
Live waveform and orb visualisation while recording (on by default).
02
Transcription engine
Recognition runs locally on your device with an on-device Whisper engine. No audio ever leaves your machine, and nothing is left behind afterward.
Custom prompt / boost dictionary Free
Your custom words gently bias recognition toward the names and jargon you actually use.
GPU acceleration Pro
Turn on GPU acceleration for near-instant transcription; otherwise it runs on CPU.
Robust failure reporting Free
A non-zero exit surfaces a real message instead of the misleading "No speech detected".
03
Text post-processing pipeline
A single, deterministic path turns the raw transcript into final text. Each stage below shows the same before→after the desktop pipeline produces.
you say
Quill writes
you say
Quill writes
you say
Quill fixes it
you say
Quill expands it
raw
tidied
04
Output / text injection
Clean text lands in whatever app has focus — typed in as real keystrokes, or pasted via the clipboard. The strategy is chosen automatically.
Toteam@company.com
SubjectSync timing
Press your hotkey to start
Two strategies, auto-chosen Free
Short text is typed directly; longer text is placed on the clipboard and pasted with
Ctrl/Cmd+V to stay instant and avoid per-key stutter.Type · Auto-paste · Auto-copy Free
Choose whether text is typed into the active app, just copied to your clipboard, or shown in the overlay transcript panel for manual paste.
05
Per-app profiles & focus context
QuillVoice notices which app is in focus and picks the matching processing profile automatically. Watch the active profile follow the focused app.
VS Code
Gmail
Notes
Other
Active profile Default
06
Overlay widget
An always-available floating widget — always on top, off the taskbar, and it remembers where you drag it. Three modes, a history panel, focus-mode auto-hide, and full appearance theming.
Hey team, can we move the sync to Thursday?
07
History & statistics
Everything is stored locally. Below is the real History and Stats UI from the app — Free sees the latest 3 entries (the rest stay on disk and return on upgrade); export, clear and unlimited history are Pro.
Hey team, can we move the sync to Thursday?
The beta is basically ready, I think we ship Friday!
// Checks whether the token is still valid before refreshing it.
Free shows the latest 3. Upgrade to Pro for full, searchable, exportable history.
Usage
3h 12m
Recording time
1,284
Transcriptions
61,402
Words typed
Activity (local only)
2,140
Words today
11,907
Words this week
8
Sessions
Reset statistics
Permanently clears all recorded usage data.
Fully local: your usage data stays on your device and powers the stats and the learning suggestions. Click the tabs above to switch.
08
Correction learning
When you edit a transcript, QuillVoice compares it to the original and remembers the change. Recurring ones surface in a review queue — the real "Dictionary suggestions" UI.
Dictionary suggestions
QuillVoice notices words you frequently correct in your history and suggests adding them to Text Replacements.
kubernetis→kubernetes
postgress→Postgres
Auto-add corrections
Skip the review and add recurring corrections to replacements automatically.
How suggestions work: a correction you make repeatedly across multiple sessions becomes a suggestion here. Try Add or Dismiss — anything you dismiss won't come back. An optional auto-add mode (off by default) skips the review entirely.
09
Licensing & activation
Offline license keys — signed by us and verified right on your device, so activation works fully offline. The app can verify a key but never create one. Paste a key and watch the edition flip to Pro live, no restart.
License
FreeCore dictation
Core dictation with the small models. Pro unlocks all models, GPU acceleration, and offline AI cleanup.
This machine's hardware ID
Needed only for per-seat / enterprise (node-locked) licenses. Share it with your administrator to bind a seat to this device.
7F3A·E91C·44B0·2D6E
Try it: the field is pre-filled with a sample key — click Activate and watch Pro-only features unlock instantly, no restart. Licenses are perpetual. Per-seat (node-locked) licenses bind to a single machine — and never lock out a paying customer if their machine can't be identified.
10
Security & privacy guarantees
Local-first isn't a slogan here — it's enforced in the architecture.
100% local processing
No audio or transcript ever leaves the device. The only network access is the consented update check and explicit model downloads.
Hardware fingerprint
A privacy-preserving device fingerprint is computed once and cached locally. Raw hardware identifiers never leave your machine.
Encrypted secret vault
Your custom dictionary, replacements and shortcuts are stored with AES-256 encryption, bound to your device — unreadable if copied to another machine.
Signed model manifest
Each model's contents are cryptographically verified before it loads, so a tampered or corrupted model file is rejected.
Self-integrity check
QuillVoice checks its own program files against a first-run baseline to catch tampering — while being careful to never block you over a legitimate update.
Atomic config writes
Settings are saved safely, so a crash can never leave you with a half-written, corrupted config.
11
Compliance attestation PDF
One click exports a professional, self-contained PDF attesting that everything was processed locally — aimed at enterprise and regulated buyers.
🔒
QuillVoice
Privacy & Processing Attestation
✓ No audio was transmitted
✓ No data left this device
✓ Processed locally using base.en
✓ No third-party services contacted
QuillVoice · Local attestation
12
Updates
QuillVoice never touches the network for updates without explicit consent. The flow: you ask → a consent dialog → on accept it checks for a newer, verified version.
Updates
Check for a newer version. QuillVoice only connects to the internet after you agree.
13
Model management
Every model is free for everyone. Download, resume, discard and delete — the real Models tab, with a live progress bar. Click Download on a model below.
Installed models
base.en142 MB
Best everyday default — fast, accurate enough for real-time.
small.en466 MB
More accurate, still quick on CPU.
large-v3-turbo1.6 GB
Top accuracy, multilingual. GPU recommended.
medium1.5 GB
Partial download found — 0.6 GB done.
GPU runtime
CUDA acceleration 🔒 Pro
One-click install of the NVIDIA runtime for near-instant transcription, with automatic CPU fallback.
Reliable downloads: if your connection drops mid-download, it resumes automatically instead of restarting a multi-gigabyte download from scratch.
14
System integration
A first-class desktop citizen: system tray, autostart, a settings window that survives closing, and rotating logs for support.
System tray Free
Show/Hide overlay, Settings, and Quit. Left-click re-opens the overlay.
Autostart on login Free
Off by default; synced to your OS startup list — no admin rights required.
Persistent settings window Free
Closing hides rather than destroys it, so it re-surfaces instantly with the right taskbar icon.
Rotating file logs + system info Free
Rotating, size-limited logs plus basic system info help with support — all stored locally on your device.
15
The settings window
The whole dashboard, faithfully rebuilt and fully clickable. Switch between all six tabs — Dashboard, Settings, Stats, History, Models, About — exactly as in the app.
Welcome
Free edition
2,140
Words today
11.9k
This week
61k
All time
2h 41m
Time saved
8
Sessions today
Recent dictations
Hey team, can we move the sync to Thursday?
The beta is basically ready, I think we ship Friday!
// Checks whether the token is still valid…
Quick settings
Filler word removal
Auto-capitalize
Active model
Active profile
Recognition
Model
Larger models are more accurate but slower.
Language
Auto-detect needs a multilingual model.
GPU acceleration 🔒 Pro
Use the CUDA runtime when available.
VAD silence
Stop recording after this much silence.
Whisper mode 🤫 🔒 Pro
Amplify near-silent input.
Microphone
Input device
Defaults to your system default.
Level
Speak to confirm the mic picks you up.
Hotkey
Hotkey
Global shortcut to start dictation.
Activation mode
Toggle, or hold-to-talk (Pro).
Behaviour
Auto-copy
Auto-paste
Insert method
Type keeps your clipboard untouched.
Start on login
Focus mode
Pulse the overlay after inactivity.
Text processing
Voice commands 🔒 Pro
Auto-capitalize 🔒 Pro
Filler word removal 🔒 Pro
Appearance
Theme
Accent & gradient 🔒 Pro
Recolour the waveform, mic widget and highlights.
→
Overlay opacity
Always on top
Above all windows, including games.
Widget mode
Mic audio ring
Show hotkey in widget
Custom dictionary
Text replacements · 3 free
→
→
Usage
3h 12m
Recording time
1,284
Transcriptions
61,402
Words typed
Privacy
100% local processing
Your voice never leaves this machine. Recognition runs on your CPU.
Compliance attestation
Export a signed PDF confirming no audio or data left this device.
Reset statistics
Permanently clears all recorded usage data.
Hey team, can we move the sync to Thursday?
The beta is basically ready, I think we ship Friday!
// Checks whether the token is still valid before refreshing it.
Free shows the latest 3. Upgrade to Pro for full history.
Installed models
base.en142 MB
Best everyday default.
small.en466 MB
More accurate, still quick.
large-v3-turbo1.6 GB
Top accuracy, multilingual.
QuillVoice — Local voice to text overlay
v2.0.0 · built 2026-06-19
Log file · stored locally
Updates
Check for a newer version. QuillVoice only connects after you agree.
Tips
1Hold the hotkey to record, release to transcribe and paste.
2Drag the overlay by its title bar to reposition it anywhere.
3
base.en is the best default for real-time use.16
Cross-platform packaging
Windows and macOS build from one codebase. The overlay tags the platform it's running on, and each platform gets its own packaging so its resources and permissions stay correct.
Windows
macOS
Per-platform builds Free
Windows and macOS each get their own build configuration, bundling the right runtime and requesting only the permissions that platform needs — like microphone access on macOS.
Signing & notarization Free
Both platforms ship properly code-signed — Windows code signing and macOS notarization — plus the microphone and accessibility permissions macOS asks for on first run.
That's the whole app.
Try it offline.
Every feature above runs on your machine. No account, no telemetry, no subscription.