An AI voice input that actually gets you — technical jargon, slang, code snippets, mixed languages. Far more accurate than Mac’s built-in dictation. Works in every app. Open source.
brew install --cask tyun08/tap/audio-input
or download from Releases · macOS Ventura+ · Windows 10+
Features
Built to stay out of your way. Invoke it, speak, and get back to work.
Injects text directly into any focused input. macOS Accessibility API and Windows key simulation. Slack, Notion, Terminal, browsers — anywhere.
Understands domain jargon, slang, and mixed-language speech that trips up Mac’s built-in dictation. An optional LLM pass also removes filler words and fixes punctuation. Screenshot context makes it even smarter.
Groq Whisper (free API key) or Google Vertex AI Gemini (enterprise ADC). Switch providers with one click. Extensible to new providers.
Audio goes directly to your chosen provider — no intermediary server. Fully open source, BYOK. You own your data.
Whisper large-v3-turbo and Gemini handle Chinese, English, Japanese, Korean, Spanish, and many more — auto-detected.
~20 MB RAM. Built with Tauri + Rust. No Electron, no background bloat. It just sits in your menu bar.
Providers
Switch providers any time. Adding a new one takes 3 files.
gcloud ADCInstall
Install via Homebrew on macOS, or grab the installer from GitHub Releases for macOS & Windows.
Prefer a manual download? GitHub Releases →
Windows: Download the .msi installer from
Releases. Windows 10+ required.
Setup
Pick a provider and start talking.
Sign up at console.groq.com. The free tier is generous enough for daily use.
Launch Audio Input, select Groq as your provider, and paste the API key. Done.
Hold the hotkey, dictate, release. Your words appear at the cursor — polished and ready.