Write 5x faster. Zero servers.

The private dictation app for Mac. Speech to text in a tenth of a second, entirely on your device. No cloud, no accounts, no data collection.

v1.4.0 — macOS 14+ — Built for Apple Silicon

Voxpen
Transcribing

Private voice to text, transcribed right on your Mac. Your audio never leaves your device. Completely offline.

LocalParakeet v30.1s latency

Works with every app on your device

Slack
VS Code
Cursor
Terminal
Chrome
Notion
Mail
Messages
Notes
Pages
Voxpen Engine

From speech to polished text

What you said

So um I wanted to uh follow up on the meeting we had yesterday and um basically the key takeaway was that we need to uh finalize the Q3 budget by Friday and also like make sure the design team is um aligned on the new dashboard layout

What Voxpen types

Following up on yesterday’s meeting the key takeaway is that we need to finalize the Q3 budget by Friday and ensure the design team is aligned on the new dashboard layout.

Filler words removed/Punctuation added/Grammar polished/Automatically
Speed

5x faster than typing

Keyboard
40 WPM
Voxpen
200+ WPM
0.1s
to transcribe a short dictation with Parakeet v3
110x
realtime — a minute of speech in about half a second
0.3W
power draw on the Neural Engine. Silent fans, full battery

Spend less time typing, more time thinking.

How it works

Four steps. Zero friction.

Press a key, speak, and your words appear. No switching apps, no copying, no pasting.

01

Press Fn

Tap the globe key or your custom hotkey from any app to start recording.

02

Speak naturally

Voxpen captures crystal-clear audio through your device's microphone.

03

Local transcription

Voxpen Engine processes everything locally on your device, instantly.

04

Auto-paste

Transcribed text appears instantly wherever your cursor is.

Features

Everything you need to dictate

Built for speed, privacy, and seamless integration with your workflow.

Completely private

All processing happens on your device. No servers, no cloud uploads, no data leaves your machine. Ever.

Network requests
Last 24 hours
0

Frontier models on your Neural Engine

Choose NVIDIA Parakeet v3 for blazing speed or OpenAI Whisper for 99 languages — both running on Apple's dedicated neural hardware at ~0.3W.

Voxpen Engine active
Local neural processing

Pastes anywhere

Transcribed text goes directly into the active app. Slack, VS Code, browsers, anything.

Slack
Let me check the design files and get back to you...

Global hotkey

Press Fn from anywhere to start dictating. No app switching required.

fn

Instant results

Parakeet v3 transcribes a short dictation in about a tenth of a second — up to 110x realtime. No waiting, no server round-trip.

Optional cloud APIs

Bring your own Deepgram or OpenAI keys for cloud transcription when you need it. No middleman markup.

No account required

Download, open, dictate. No accounts, no subscriptions, no tracking.

Under 10 MB

Lightweight app that stays out of your way. Minimal disk space, minimal memory usage.

Models

Pick your engine

Other dictation apps lock you into one model. Voxpen ships seven — switch anytime in Settings → Models, downloads happen once, and everything stays on your device.

Parakeet v3New — Fastest
NVIDIA Parakeet TDT 0.6B · ~500 MB · English + 24 European languages

The fastest on-device speech model available. Transcribes a short dictation in ~0.1s — up to 110x realtime on the Neural Engine.

Speed~10x Whisper large
distil-large-v3Recommended
OpenAI Whisper (distilled) · 600 MB · English-optimized

The balanced default. Near large-v3 accuracy at a fraction of the size, tuned for fast English dictation.

SpeedFast
large-v3-turbo
OpenAI Whisper (turbo decoder) · 632 MB · 99 languages

Whisper's multilingual flagship with a pruned decoder — around 8x faster than large-v3 with near-identical accuracy.

SpeedFast, multilingual
large-v3
OpenAI Whisper (full) · 1.5 GB · 99 languages

Maximum accuracy for challenging audio, heavy accents, and rare languages. When quality matters more than speed.

SpeedMost accurate

Plus compact Whisper models — tiny.en (75 MB), base.en (140 MB), and small.en (460 MB) — for minimal footprints.

Polish Mode

Speak naturally. Get polished text.

Voxpen Engine automatically removes filler words, adds punctuation, and fixes grammar — so you can focus on what you're saying, not how you're saying it.

Removes filler words
Auto-punctuation
Grammar correction
Context-aware numbers
Raw transcription

so basically um this has been like one of our greatest quarters we grew revenue by twenty percent and uh make sure we're hitting the the monthly targets

Polish Mode
Polished output

This has been one of our greatest quarters. We grew revenue by 20% — let's make sure we're hitting the monthly targets.

Use Cases

Built for how you work

Developers

Dictate code comments, documentation, and commit messages without leaving your editor.

Writers & Creators

Draft articles, scripts, and emails at the speed of thought. Edit later, capture now.

Students

Take lecture notes instantly. Transcribe study sessions. Write essays faster.

Professionals

Compose client emails, meeting summaries, and reports hands-free.

Accessibility

Break free from the keyboard. Full voice control for users with RSI or motor disabilities.

Teams

Fewer meetings, faster alignment. Dictate updates and feedback across tools.

Privacy

Your audio never leaves your device

No cloud. No servers. Your voice stays on your machine.

No network requests

Disconnect WiFi and Voxpen works perfectly. Voxpen Engine processes everything locally on your device.

No data collection

Zero analytics, zero telemetry, zero crash reports. Your transcriptions stay on your machine. Period.

Verifiably local

All processing runs locally. You can verify with Activity Monitor — zero outbound network calls.

Your Device
Audio
Voxpen Engine
Text
No cloud. No servers. No exceptions.
Compare

How Voxpen stacks up

See how Voxpen's local, private approach compares to the alternatives.

Feature
Voxpen
Cloud dictation apps
Built-in dictation
Privacy
100% on-device
Sent to servers
Sent to servers
Works offline
Yes
No
Limited
Account required
No
Yes
Yes
Pastes into any app
Yes
Varies
Yes
Smart text polishing
Yes
Varies
No
99 languages
Yes
Varies
Limited
Choice of AI models
7 models incl. Parakeet v3
Fixed
Fixed
Lightweight (< 10 MB)
Yes
No
System
Open hotkey system
Fully customizable
Limited
Fixed
Under the hood

Under the hood

voxpen --specs
PlatformmacOS 14+ (built for Apple Silicon)
EnginesNVIDIA Parakeet TDT v3 + OpenAI Whisper (WhisperKit)
Models7 local models, 75 MB to 1.5 GB — switch anytime
ProcessingApple Neural Engine via CoreML, ~0.3W
Latency~0.1s short dictations, up to 110x realtime
Languages99 via Whisper, 25 European via Parakeet
RAM usage< 150 MB active
StorageLocal only — nothing leaves your device
FAQ

Frequently asked questions

Everything you need to know about Voxpen.

Start dictating in 30 seconds

Download Voxpen and start speaking. No accounts, no setup, no cloud.