macOS · Apple Silicon · Completely Offline

Your voice.
Instant text.

Native macOS menu bar app that turns speech into text in under a second. Push to talk, release to paste. No cloud, no latency, no compromise.

<1s
End-to-end latency
190×
Realtime factor
25
Languages
0 B
Data sent to cloud

Built for speed
and nothing else

No browser. No Electron. No Python runtime. Just native Swift talking directly to the Apple Neural Engine. Every millisecond matters.

Push to Talk

Hold the Fn key to record, release to transcribe. Natural as a walkie-talkie. No clicking, no UI.

Auto-Paste

Text appears at your cursor in whatever app is focused. TextEdit, browser, terminal, Slack—anywhere.

100% Local

Everything runs on-device via CoreML. Your audio never leaves your machine. No account, no API key, no internet.

Neural Engine

The encoder runs on Apple’s dedicated Neural Engine via CoreML. 190× realtime on M-series chips.

Invisible Until Needed

Lives in your menu bar. No dock icon, no window. A translucent overlay appears only while recording.

25 Languages

Powered by NVIDIA Parakeet TDT v3 supporting English, German, French, Spanish, and 21 more European languages.

AI-powered
transcript cleanup

Optionally clean up transcripts using GitHub Copilot—fix grammar, remove filler words, and polish punctuation while preserving meaning, tone, and language.

GPT-4.1

Default Model

High-quality grammar and punctuation cleanup. Included with any GitHub Copilot subscription—no premium tokens required.

GPT-5 Mini

Lightweight & Fast

Quick enhancement for simple corrections. Also included with GitHub Copilot at no extra cost—no premium tokens.

Claude Haiku

Premium Option

Claude Haiku 4.5 for alternative enhancement style. Uses premium tokens from your GitHub Copilot allowance.

Requires the GitHub Copilot CLI installed and authenticated. Works with all 25 supported languages.

Three steps.
Under one second.

The entire pipeline—from microphone to pasted text—completes before you finish lifting your finger.

01

Hold Fn

Press and hold the Fn (Globe) key. The mic activates instantly—audio streams into a 16kHz buffer via AVAudioEngine. A floating pill overlay confirms you’re recording.

02

Release

Let go of Fn. The audio buffer is passed to the Parakeet TDT model running on the Neural Engine via CoreML. Transcription finishes in milliseconds.

03

Pasted

The transcribed text is placed on the clipboard and a Cmd+V is simulated. Text appears at your cursor. Your original clipboard is silently restored.

Built on proven
foundations

OpenWritr stands on the shoulders of exceptional open-source work.

FluidAudio

CoreML-optimized speech recognition framework by FluidInference. Provides the Parakeet TDT 0.6b v3 model compiled for Apple Neural Engine with ~190× realtime performance.

NVIDIA Parakeet TDT v3

State-of-the-art 600M parameter ASR model using Token-and-Duration Transducer architecture. Non-autoregressive decoding for minimum latency. 25 European languages.

Apple CoreML & Neural Engine

Hardware-accelerated inference on Apple Silicon’s dedicated Neural Engine. The encoder model runs entirely on the ANE for maximum throughput and minimal power draw.

Swift 6 & SwiftUI

Pure native implementation with strict concurrency safety. MenuBarExtra for the system tray, AVAudioEngine for low-latency capture, CGEventTap for global hotkeys.

Up and running
in 30 seconds

Requires macOS 14+ and Apple Silicon. The ASR model (~460 MB) downloads automatically on first launch.

Terminal
# Clone and build
git clone https://github.com/trsdn/OpenWritr.git
cd OpenWritr
swift build -c release

# Run
.build/release/OpenWritr

# Grant Microphone + Accessibility permissions when prompted
# Hold Fn to speak, release to paste