Inputalk

Free, open-source dictation for macOS. Hold Fn, speak, release — text appears wherever your cursor is.

No cloud. No API keys. No cost. Transcription runs entirely on your Mac via WhisperKit.

How it works

• Hold Fn — push-to-talk dictation. Release to transcribe and paste.
• Double-press Fn — hands-free dictation. Press Fn again to stop.
• Works in any app — Slack, VS Code, Terminal, Messages, browser, email, anything with a text field.

Install

Download the latest .dmg from inputalk.com.

Open the DMG, drag to Applications. On first launch, grant Microphone and Accessibility permissions when prompted.

Models

Whisper models download on first use and run on Apple Neural Engine.

Model	Size	Speed	Quality
tiny	75 MB	~10x realtime	Basic
base	142 MB	~7x realtime	Good (default)
small	466 MB	~4x realtime	Very good
medium	1.5 GB	~2x realtime	Excellent

Switch models in Settings. Model data is stored in ~/Library/Application Support/com.inputalk.app/Models/ and removed on uninstall.

Requirements

• macOS 15+
• Apple Silicon or Intel

Build from source

cd macos
swift build
swift run

Release build (signed + notarized)

# Set up .env with CODE_SIGN_IDENTITY
cd macos
./scripts/release.sh          # build + sign + notarize + DMG
./scripts/publish-release.sh  # upload to S3 + update latest.json

Project structure

macos/          Swift app (SPM, macOS 15+)
  Sources/      AppDelegate, services, views
  Resources/    Info.plist, entitlements, icons
  scripts/      build, release, publish
web/            Landing page (Next.js 15 + Tailwind)
video/          Launch video (Remotion)

Tech

• WhisperKit — on-device speech-to-text via CoreML
• Swift 6, Swift Package Manager
• CGEvent tap for global Fn key capture
• Clipboard + simulated Cmd+V for text insertion
• Liquid Glass floating indicator (macOS Tahoe)

License

MIT