Chatterfox
Building voice-to-text transcription without the subscriptions.
Voice and context
Voice has become increasingly important in the age of AI. It's about speed, but it's also about context. The more context I can give agents and workflows, the more I can get from them, and speaking has fundamentally changed the way I work.
I was an avid Wispr Flow fan and had seen how much it changed what I was able to do. That made me curious about what was possible with open-source voice tools. I looked at what the community had already built and found Handy, an incredible open-source project for running voice-to-text transcription on your own device.
My first open-source fork
I'd never forked an open-source project before. I wanted to try building on someone else's work and add my own touch and twist. With Chatterfox, that meant a fun, friendly interface, with multiple colour themes to match your mood. I stripped back the model choices and focused on NVIDIA's Parakeet Unified, a ridiculously fast English-language model.
Ready when I want to dictate
The app lives in the Mac's menu bar, ready when I want to dictate. Everything runs on my device, so my speech stays private without needing a cloud transcription service.
A personal piece of software
Chatterfox gives me a fun, easy-to-use and free alternative to subscription voice-to-text tools. It's fast, tailored to me and open source. It saves me a subscription, but it's also become a highly personal piece of software.
Built with: Cursor, NVIDIA Parakeet, Claude Code (Sonnet 4.6 and Opus 4.6), Tauri, Remotion
Part of the Rolling Waves work archive.