Standard voice dictation is deeply frustrating. Standard system dictation often fails to recognize specialized terms, ignores natural phrasing, or forces you to go back and edit dozens of spelling mistakes. If you try to use cloud-based alternatives, your private voice recordings are uploaded directly to external servers.
Our April 2026 update solves these constraints entirely by embedding our state-of-the-art local transcription and LLM correction pipelines directly into the iOS and macOS keyboard systems.
1. System-Wide AI Input: iOS & macOS Integration
On Device AI now runs as a native system component on your Apple hardware:
- iOS Keyboard Extension: Switch keyboards in any iPhone or iPad application to access our high-performance voice interface instantly.
- macOS Input Method: Tap a global hotkey to summon a floating, elegant glassmorphic voice typing panel that inputs text directly at your cursor in any desktop app.
This deep OS integration operates within secure sandbox boundaries, letting you input private voice transcripts directly into text fields without switching apps.
2. Real-Time Whisper Core & Offline Speech Processing
The entire transcription engine runs natively on Apple Silicon. By utilizing optimized Whisper speech-to-text models running on your device's Neural Engine, On Device AI transcribes your spoken words locally at incredible speeds.
Because the processing is offline, there are no transmission delays or server outages. It works perfectly on an airplane, in a remote workspace, or in a deep basement—all with zero data leaving your hardware.
3. Smart, Verbatim, and No-Correction Modes
Spoken language is naturally messy. We repeat words, stutter, use filler phrases like "um" and "ah", and occasionally make grammatical mistakes. On Device AI utilizes small, highly efficient local LLMs (like Qwen 2.5 1.5B or Gemma 2 2B) to clean up transcripts instantly via three specialized correction modes:
- Verbatim Mode: Transcribes every spoken word exactly as uttered, ideal for strict dictation or code commentary.
- Smart Mode: Intelligently strips stutters and filler words, adds appropriate punctuation, fixes grammar, and structures the output into polished, professional prose.
- No-Correction Mode: Bypasses LLM styling entirely for raw speech-to-text output, maximizing speed and minimizing battery consumption.
4. Unified Cross-Platform Settings
Your custom terms, jargon, and preferences are synchronized securely and privately across your devices. Set up your default correction preferences, preferred Whisper model sizes, and hotkey bindings on your Mac, and experience the exact same responsive voice workflow on your iPad and iPhone.
Upgrade to On Device AI today via the App Store and transform how you write, communicate, and create—simply by speaking.