Google’s Eloquent AI App Aims to Transcribe Meetings Offline
Google launched Eloquent, an offline AI transcription app using Gemma models that keeps voice data on-device, addressing privacy concerns in legal and medical sectors.
Jersey City, N.J., April 13th, 2026
Google has silently launched “Google AI Edge Eloquent,” a new artificial intelligence (AI) dictation app that processes speech offline. Built on Google’s lightweight Gemma-based speech recognition models, the app is designed to allow users to transcribe sensitive conversations directly on their devices. This move appears to address the privacy concerns that have historically limited AI adoption in the legal, medical, and financial sectors.
By keeping transcripts on a local phone, the tool mitigates the risk of proprietary information being intercepted or used for model training. This on-device approach also reflects a broader industry shift away from routing voice data through cloud infrastructure. An optional cloud mode is available, which routes text through Gemini models for additional cleanup, but the app’s default behavior keeps all audio on-device.
For professionals, the launch reflects a growing enterprise interest in privacy-focused AI. As offline tools aim to automate the capture and summarization of spoken work, the role of the employee may shift further toward high-level strategy. Smaller, disciplined models like the one powering Eloquent appear to be demonstrating that cloud-based systems may not always be necessary for daily tasks. However, it remains to be seen how the app will perform against tools with deeper real-time and cloud integrations.
While Google is clearly aiming to capture demand for on-device AI, the app currently launched exclusively on iOS with no announced Android release date, and it is too early to tell if the market will fully embrace Eloquent as a replacement for established subscription alternatives.
