Google lists free AI transcription app “Edge Eloquent”: runs on-device with Gemma, no internet connection required, launches on iOS first

Google has quietly released a free offline AI dictation app, “AI Edge Eloquent,” on the iOS App Store this week. It uses an on-device Gemma ASR model, with all audio processing staying entirely on the device, and—rarely—arriving on iOS before Android.
（Backgrounder: Google unveiled its smartest open model, Gemma 4! It fully switched to the Apache 2.0 license, with performance that rivals large closed-source AI models.）
（Additional context: Breaking Nvidia’s dominance? Google’s strongest AI model on the ground, “Gemma 3,” uses 1 H100 to beat DeepSeek and OpenAI.）

Table of contents

Toggle

• Speech to text—start talking and you can use it
• iOS leads before Android—Google’s unusual move
• Gemma moves into everyday life: real-world validation of on-device AI

Without any press event or official announcement, Google has just, earlier this week, quietly pushed “AI Edge Eloquent” to the iOS App Store. This completely free AI dictation app focuses on offline speech recognition. Behind it runs Google’s own on-device Gemma on-device ASR (automatic speech recognition) model, with all audio processing completed locally on the device, with no need to connect to the internet.

Speech to text—start talking and you can use it

The “AI Edge Eloquent” operating logic is straightforward: after opening the app, talk into the microphone, and the screen displays text and the waveform in real time. After you stop recording, the app automatically filters out spoken filler words like “um” and “ah,” corrects sentence breaks, organizes what you said into clean, ready-to-use text, and automatically copies it to the clipboard.

Privacy design is also one of the app’s core selling points. In a pure offline mode, audio data doesn’t leave the user’s iPhone and isn’t sent to any server. Google also provides a “cloud mode” option: speech recognition is still completed on the device, and only the recognized text (not the original audio) is sent to the Gemini cloud model for further cleanup, resulting in higher quality.

The entire service is completely free—no subscription fees and no usage limits.

iOS leads before Android—Google’s unusual move

At the same time, tech media noticed a rare phenomenon: almost all of Google’s products launch on Android first, especially features involving on-device AI. They typically roll out first in tandem with Pixel devices and the Gemini Nano model. This time, “AI Edge Eloquent” goes against the grain: even in the App Store description, it directly states that an Android version is coming.

Analysts have two interpretations. The first is a market test: Google may be intentionally choosing iOS—an “off-home” market—for an experiment, assessing consumer acceptance and usage habits for on-device AI dictation without disrupting the Android ecosystem.

The second is that the technology is ready: the Gemma ASR model on iOS may have reached a usable standard earlier than its Android version in terms of optimization progress. Google chose to publish directly rather than waiting to launch both platforms simultaneously.

Gemma moves into everyday life: real-world validation of on-device AI

From a technical perspective, the significance of “AI Edge Eloquent” isn’t just that it’s a dictation tool. Gemma is Google’s open-source lightweight AI model family, designed to run efficiently in resource-constrained environments like mobile devices. AI Edge is the framework Google provides for developers to run machine learning models on-device.

In effect, this app brings Gemma’s on-device inference capabilities into everyday consumer use scenarios.

For developers, AI Edge Eloquent is also a reference template: it shows how Gemma models can be deployed in mainstream consumer products, and how a hybrid offline-and-cloud architecture can strike a balance between privacy and quality.