Google unveils EmbeddingGemma 2 and the AI Edge Foresight app for Mac

Google has unveiled EmbeddingGemma 2, a 740-million-parameter model built on the Gemma 4 architecture. It is open under the Apache 2.0 license and is intended for organizing, searching, and linking information directly on the user's device. The company demonstrated the model's capabilities using the AI Edge Foresight app for Mac as an example.

The model is designed to run on consumer hardware, including under tight resource constraints. According to Google, after quantization on the Pixel 11 Pro the text weights need about 191 MB of RAM, and the full multimodal version about 567 MB. Among the model's tasks are finding a specific video fragment from a voice note and searching multi-hour audio recordings by text query, with a single native multimodal model handling everything.

Meeting notebook

AI Edge Foresight works with notes during meetings. The app listens in the background to video meetings or in-person meetings, and the user can record short talking points. These jottings are then instantly turned into tidy notes based on the meeting transcript. Audio and transcript are processed entirely on the device, not in the cloud.

The app can be pointed at work folders, reference materials, diagrams, and the calendar. Local files and Google Drive are supported, after which you can ask questions about their contents and quickly get the information you need. There is also a chat function: conversational search and a brief summary of personal materials. A separate feature, Live Assistance, listens and automatically answers questions.

AI Edge Foresight is part of the Google AI Edge lineup for Mac alongside two other apps, AI Edge Gallery and AI Edge Eloquent for offline transcription. The app is available for download. Demos of the model are also available in Google AI Edge Gallery on Android and iOS.

Translation editor: Салтанов А.