Affiliate links on Android Authority may earn us a commission. Learn more.
On-device AI is becoming increasingly capable, with companies exploring ways to run more AI workloads directly on phones and other consumer devices. Perplexity, for instance, has launched on-device AI assistants, while Google has released models such as Gemma 4 12B specifically designed for on-device use. Now, Google appears to be pushing that idea even further.
Today, the company announced a new multimodal on-device model called EmbeddingGemma 2. Google says the new model has been designed for privacy-first applications and can act as an “ultra-low-latency on-device decision engine.” Since it is multimodal, EmbeddingGemma 2 can look up information from within text files, images, video frames, and audio.
To showcase the capabilities of the new EmbeddingGemma 2 model, Google has also launched a brand-new macOS app called AI Edge Foresight. The app uses EmbeddingGemma 2 to help users take notes during online or in-person meetings. Users can also add knowledge sources, and the app will use the imported knowledge base to answer questions.
AI Edge Foresight keeps all files, meeting audio, and everything else completely on-device. It can integrate with the mic and system audio on Mac for a fully offline experience. Users can write shorthand bullet points in a meeting, and the app uses on-device AI to create “polished notes using the meeting transcript.”
It also supports Live Q&A, which uses the meeting transcript (and any additional knowledge sources) to answer questions posed during a live meeting. AI Edge Foresight can use EmbeddingGemma 2 to detect questions in a meeting, go through the meeting transcript, and generate answers.
I tried using the AI Edge Foresight app while pretending to be in a meeting. The app quickly and successfully turned my bullet points into notes. It also detected questions and provided answers from the meeting transcript.
The AI Edge Foresight app runs on Apple Silicon, and you can download it for free right now. It takes a minute to download the required on-device models, but then you’re good to go.
Thank you for being part of our community. Read our Comment Policy before posting.
