Project · 2026

Voice Dictation

A menu-bar dictation app for Apple Silicon Macs. Hold a hotkey, speak, release, and the text appears at your cursor in whatever app is focused. Runs a polished batch mode (local Whisper transcription plus Bedrock cleanup) or a fully-offline live streaming mode.

Python
mlx-whisper
AWS Bedrock
macOS

View code

Problem

Dictation tools that live in the cloud send your microphone audio off the machine, which rules them out for anything private or technical. I wanted push-to-talk dictation on my Mac that put text at the cursor in any app, kept audio on-device, and still cleaned up the raw transcript into something readable.

Approach

A menu-bar tool built for Apple Silicon with two modes:

Batch mode

Stage 1 (transcription) runs 100% locally via mlx-whisper on Metal, so audio never leaves the machine.
Stage 2 (cleanup) optionally sends only the raw transcript text (never audio) to AWS Bedrock (Claude Haiku) for grammar, punctuation, and technical-term correction before pasting at the cursor.

Streaming mode

A fully-offline live mode that types words as you speak, with no clipboard and no network.

Highlights

Hold a hotkey, speak, release, and the text lands at the cursor in whatever app is focused.
Audio stays on-device; only text is ever sent anywhere, and only when batch cleanup is enabled.
A fully-offline path for when you want zero network involvement at all.