Private, local-first dictation for macOS
ToskVoice
Local dictation that stays on your Mac.
A native menu-bar dictation app with on-device speech recognition, visible live drafts and focused-field insertion.
Why it exists
When the right dictation app is missing, build it.
I wanted to experiment with current voice models, but the dictation applications I found for Mac were either not fully local or did not offer the experience I wanted. Eventually I stopped searching and built my own.
ToskVoice keeps the core dictation path on the Mac. It listens from a non-activating floating panel, shows the transcript as it forms and inserts the result into the application that already has focus. Spoken corrections and optional on-device polishing can shape the draft without sending the recording to a remote service.
Open source, ready to explore
Experiment with local dictation on your own Mac.
Under the hood
Built with
- Build approach
- Vibe-coded
ToskVoice was built through a fully vibe-coded workflow. I defined the experience I wanted from local dictation and used the project to explore what current voice and on-device language models can make possible. Most of the application was developed with GPT-5.6 and Claude Fable 5; the focused-field input logic came from Claude Fable 5 after GPT-5.6 could not make that part reliable.
AI tools and models
GPT-5.6 and Claude Fable 5
- Availability
- Early-stage native macOS app · build from source
- License
- Source available · project license pending
- Source
- GitHub repository