TechCrunch: Google quietly launches an offline AI dictation app for iOS

On April 6, 2026, TechCrunch reported that Google released a new iOS app called Google AI Edge Eloquent without a formal announcement. The app is an AI-powered dictation tool that runs entirely on-device using Google’s Gemma AI models, meaning it transcribes speech without sending audio to remote servers.

What it does

Eloquent converts spoken audio to text on the device itself. It removes filler words — “um,” “uh,” and mid-sentence corrections — automatically as part of the transcription. An Android version is referenced in the App Store description, though it had not launched as of the initial report. The app positions itself against cloud-based dictation tools like Wispr Flow, which process audio on remote servers.

The on-device distinction

The privacy angle is the product’s main differentiator. Writers who dictate notes, drafts, or source interview summaries generate audio that contains unpublished work, sensitive source information, and editorial judgment that may not be appropriate to route through third-party cloud infrastructure. An on-device model addresses this by keeping the audio local throughout the process.

The use of Gemma — Google’s open-weight model family — for on-device processing is also notable as a signal of how capable smaller models have become for specific tasks. Dictation and transcription with filler-word removal is a task that benefits from a fast, lightweight model running locally rather than a large model running remotely.

Why it matters for writers

Dictation has long been a faster way to get ideas into text for many writers, but cloud dependency and concerns about audio privacy have limited adoption in professional contexts. A reliable offline option from Google — available on iOS with an Android version planned — expands the practical range of writers who can use voice-to-text as part of their workflow without the tradeoffs that cloud-based tools require.