Google quietly launched an AI dictation app that works offline

Google has quietly released a new experimental dictation app for iOS called “Google AI Edge Eloquent,” positioning it as an offline-first tool in a growing market for AI-powered transcription services. The app competes directly with existing players like Wispr Flow and SuperWhisper by offering on-device speech recognition that functions without an internet connection. Its release, though low-key, signals Google's intent to test and deploy more sophisticated, edge-computed AI features directly to consumers.

Functionally, Eloquent uses Google's Gemma models for its core automatic speech recognition (ASR), which are downloaded to the device upon installation. The app provides live transcription and, upon pausing, automatically filters out filler words and self-corrections to produce cleaner text. Users can then apply further transformations, such as summarizing into key points or adjusting the tone to be more formal. While the primary mode is local, an optional cloud-connected mode utilizes the more powerful Gemini models for text cleanup. The app also allows users to import custom vocabulary, including names and jargon from their Gmail account, and tracks dictation metrics like words-per-minute.

While the app is currently available only on iOS, its initial App Store description referenced “seamless Android integration,” including the ability to function as a system-wide keyboard. Though these references have since been removed, their inclusion suggests this app is a testbed for features that could eventually be integrated into the Android operating system. This approach allows Google to refine its on-device AI models in a competitive, real-world environment before a potentially wider rollout, which could significantly enhance Android's native voice input capabilities.

Google's 'Eloquent' app is a strategic probe, using the iOS ecosystem as a controlled environment to battle-test its on-device Gemma models against specialized competitors before committing to a deeper, and more complex, integration within its own Android platform.