How it works Features Pricing Blog Tools
About

Why LexaWrite exists.

I built LexaWrite because I was tired of voice dictation that required the cloud. Every time I dictated, my words were sent to someone's server - and I couldn't even use it offline on a plane. It felt wrong for something as personal as your voice.

When a breakthrough open-source speech recognition model became available, it changed everything. Here was a model accurate enough for real-world use, small enough to run on a MacBook, and fast enough to be useful - especially on Apple Silicon with Metal GPU acceleration.

LexaWrite is my attempt to build the dictation experience I always wanted: hold a key, speak, release - and your words appear, transcribed entirely on your Mac, never touching the internet. No account required to start. No subscription for basic use. No cloud dependency, ever.

What LexaWrite is built on

Under the hood, LexaWrite uses whisper.cpp, Georgi Gerganov's highly optimised C++ port of an open-source speech recognition model that runs efficiently on Mac hardware. On Apple Silicon, it uses Metal for GPU acceleration, making transcription noticeably faster than CPU-only approaches.

The app is built in Swift using SwiftUI, targeting macOS 12 and later. It works on both Apple Silicon and Intel Macs. The entire transcription pipeline - audio capture, format conversion, inference, post-processing - runs locally.

System requirements

  • macOS: 12 (Monterey) or later
  • Architecture: Apple Silicon (M1 and later) or Intel - both supported
  • Storage: 75 MB for the app; 75 MB–3 GB additional per AI model (Tiny to Large)
  • Microphone: Built-in or external - any microphone supported by macOS
  • Internet: Not required for transcription. Required only for Pro subscription management.

Apple Silicon Macs (M1 and later) benefit from Metal GPU acceleration, making transcription significantly faster than on Intel Macs. The Tiny and Base models run well on both. The Large model is best suited to Apple Silicon.

About me

I'm Salih Caglar Ispirli, a senior software engineer with over a decade of experience across the full stack - since 2015. LexaWrite is a side project I built to solve a real frustration: every voice dictation tool I tried either sent my audio to the cloud or had too much friction to use without thinking about it. I built the app I wanted to exist.

My background spans both native Apple platform development and backend systems - Swift, SwiftUI, AVFoundation on the client side, and server-side infrastructure on the other. The on-device AI integration, the 16 kHz audio pipeline, the Metal GPU path, the clipboard preservation mechanism, and the Cloudflare Workers backend - all of it is something I've worked through hands-on.

If you have feedback, found a bug, or just want to say hi - reach out on Twitter, GitHub, or LinkedIn.

Privacy, for real

"Privacy-first" is a claim a lot of apps make. Here's what it actually means for LexaWrite:

  • The AI model runs on your device. Your audio never leaves your Mac.
  • Transcripts are stored in a local SQLite database on your Mac.
  • The free tier requires no account - nothing to sign up for, no email to hand over.
  • The Pro subscription is managed via Apple (in-app purchase), so payment goes through Apple's infrastructure, not a custom server.

See the full Privacy Policy for details.