Google expands its AI footprint with EmbeddingGemma 2, an open, multimodal model designed for local inference on consumer devices, boosting privacy, speed, and offline functionality with its new AI Edge Foresight application.
Google has expanded its on-device artificial intelligence push with EmbeddingGemma 2, a new multimodal embedding model built for local inference rather than cloud processing. According to Google, the model is designed for privacy-sensitive applications and can act as an ultra-low-latency decision engine on consumer hardware. The company says it can work across text, code, images, video and audio, placing those inputs into a shared embedding space for search, classification and retrieval tasks.
The release reflects a wider industry shift towards running more AI features directly on devices, where latency is lower and user data can remain local. Google says EmbeddingGemma 2 is open and released under the Apache 2.0 licence, and that it is built on the Gemma 4 architecture. Other reports say the model has 740 million parameters and is aimed at efficient on-device inference, making it suitable for phones and other consumer devices where power and memory remain constrained.
To show what the model can do, Google has also introduced AI Edge Foresight, a macOS application that keeps meeting notes and audio processing on device. According to Google and Android Authority, the app can take shorthand bullet points and turn them into polished notes, while also detecting questions raised during meetings and generating answers in real time. It can use imported knowledge sources as well as meeting transcripts, and Google says it works entirely offline, including with microphone and system audio input on Apple Silicon Macs.
The practical value of the app is in how it combines note-taking, retrieval and live question handling without sending material to the cloud. That makes it a demonstration not only of the model’s multimodal abilities, but also of the broader case for local AI: lower latency, stronger privacy controls and fewer dependencies on network connectivity. Google has positioned EmbeddingGemma 2 as a foundation for that approach, with AI Edge Foresight serving as a visible proof of concept.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
