Google has quietly released Eloquent, an iPhone-only voice-dictation app that processes speech into polished, publish-ready text using on-device Gemma AI models., an app which requires no internet connection after installation and keeps voice data on the device.
Eloquent converts casual speech into cleaned, grammatically corrected prose by removing filler words and restructuring sentences. It provides live transcription during dictation and a suite of post-processing tools that generate bullet-point summaries, alter tone (formal/short/long), and automatically copy final text to the clipboard.
As per informatio availabel in public domain, the app runs entirely on-device, leveraging Google’s Gemma models under the company’s “AI Edge” initiative. Processing is performed locally after initial download, eliminating the need for cloud calls and preserving user data on the phone.
Eloquent supports personal vocabulary—users can add custom names and jargon—and offers an option to import frequently used terms from Gmail to improve recognition accuracy.
Not just that, but Eloquent is free as well to use with no subscription or usage caps, positioning it as a direct competitor to paid dictation solutions such as Wispr Flow and SuperWhisper and potentially challenging built-in OS dictation services.
Launched quietly on the App Store without a major announcement; currently available only for iPhone, while Android support is expected at a later date.
The release is believed to underscores a broader industry shift from cloud-based AI to edge-first, on-device processing—promising faster response, improved privacy, and offline capability. It also signals growing momentum for voice-first productivity workflows in which users speak and AI produces publish-ready text.
However, Eloquent is an early-stage offering under Google’s “AI Edge” branding. Platform coverage is limited to iOS, and real-world performance may vary across accents and noisy environments.
