The app lets you bring your own API key (OpenAI, Groq, etc.) and make direct calls. If you want complete privacy, it also supports local transcription. Either way, your audio never goes through any middleman servers. It's super lightweight (~22MB), built with Svelte 5 and Tauri, and works on Mac, Windows, and Linux.
I've been using it daily for the past few months and just released v7 last night. I spent a considerable amount of time developing a clean architecture that's hopefully educational to read. This is one of the most complex Svelte 5 apps in production, with extensive use of runes and TanStack Query.
I'm happy to answer questions about implementation or how to build desktop apps with this stack!
braden-w•7mo ago
The three-layer architecture has been particularly helpful:
- Services: https://github.com/braden-w/whispering/tree/main/apps/app/sr...: Pure functions with platform abstraction, no UI dependencies
- Query layer: https://github.com/braden-w/whispering/tree/main/apps/app/sr... : Adds reactivity, caching, and runtime dependency injection
Voice-activated mode is particularly nice when coding—you can keep your hands on the keyboard while dictating. The Svelte 5 runes + TanStack Query combination has been fantastic for managing real-time audio state.