ProductCreator & Lead Engineer2025shipped
Vaani
Press a hotkey, speak, and your words appear — privacy-first macOS voice dictation.
Stack
ElectronReactTypeScriptTailwind CSSwhisper.cppGroqNode-API (C++ / Obj-C)
Outcomes
OfflineLocal whisper.cpp inference
5 modesContext-aware text injection
0Telemetry, cloud storage
What it is
A native macOS voice dictation app. Hit a global hotkey, speak, and the words appear at the cursor in any application — with a choice of cloud or fully on-device transcription. No subscription, no cloud storage, no telemetry.
Key points
- Multi-provider STT — Groq Whisper, OpenAI Whisper, Deepgram Nova-2, or local
whisper.cpp(offline). Switch per-app or globally. - Multi-provider LLM formatting — optional cleanup pass via Groq Llama, OpenAI GPT, Anthropic Claude, or OpenRouter — removes filler words, fixes punctuation, applies smart formatting.
- Context-aware injection — detects the active app and picks the best of 5 insertion strategies (Accessibility API, clipboard, keystroke, …) with per-app override policies.
- Per-app profiles — different provider, language, and formatting settings depending on which app you're typing into.
- Native macOS integration — global hotkey, audio capture, and text injection are written as C++/Obj-C Node-API addons; the rest is Electron + React + TypeScript.
- Snippets, dictionary, and history — slash-command snippets, custom word replacements, and a browseable history of past dictations.
Result
Shipped to GitHub Releases with a notarised macOS DMG. Cloud mode achieves end-to-end latency competitive with paid alternatives; local mode keeps every byte of audio on-device. Watchdog and capsule-overlay reliability fixes shipped in v1.0.4.