Google launches offline dictation app powered by AI

Google has released Google AI Edge Eloquent on iOS, a free dictation app that works offline after users download Gemma-based speech models, and can use Gemini in the cloud for text cleanup.

The app shows live transcripts and removes filler words after users pause. It also offers rewrite modes such as key points or formal text.

It can import names and jargon from Gmail if users allow it, add custom words, and store searchable transcription histories with word count and speaking speed.

The App Store listing also points to an Android version with keyboard and floating button support, but Google has not provided details on its availability.

🔗 Source: TechCrunch

🧠 Food for thought

Implications, context, and why it matters.

Eloquent is a showcase for Google’s on-device AI strategy

Eloquent runs offline using downloaded, Gemma-based speech models on-device on the user’s iPhone, rather than leaning on cloud AI 1.
It likely relies on smaller Gemma 4 edge models such as E2B or E4B, built for mobile hardware with audio input support 2.
Arm, the chip architecture used in iPhones and many other mobile devices, says Gemma 4 can run faster on-device with Armv9 CPU instructions such as SME2 plus software tuning in XNNPACK (a library that speeds up neural-network math) through ecosystem collaboration 3.
Local transcription can keep processing on-device, which supports privacy and keeps the app usable without connectivity 3.

The free app could help seed an on-device AI developer ecosystem

Eloquent can act as a clean example of Google’s on-device stack, including Google AI Edge and related tools, and may nudge developers toward offline-first features 1.
Google released Gemma 4 under the Apache 2.0 license (a widely used open-source software license) 4.
Apache 2.0 allows commercial use and redistribution with required attribution and license notice, instead of Google’s earlier more restrictive terms 5.
This may encourage products that rely less on cloud AI APIs (application programming interfaces) where privacy and offline reliability matter, from consumer apps to industrial Internet of Things (IoT) deployments 5.