Google has released Google AI Edge Foresight, an experimental Mac app that listens to your meetings and turns a few scribbled bullet points into full notes. The transcript, the audio and your files stay on the Mac, and the app keeps working without an internet connection.
Google announced the app today, alongside EmbeddingGemma 2, a new open model from Google DeepMind. Google AI Edge Foresight is the main showcase for that model, and the download page for Google AI Edge Foresight offers a macOS installer. Google says the app is optimized for Apple Silicon.
Google AI Edge Foresight turns shorthand bullets into full meeting notes
You type a few short bullets while a call is running. Google AI Edge Foresight transcribes the system audio and microphone in the background, then expands each bullet with details from the conversation. Because it reads audio from the Mac itself, it works with Google Meet, Zoom or a face-to-face meeting.
The app also learns from your own documents. You point it at project folders, reference material, diagrams and your calendar, and it indexes them without any manual sorting. Local folders and Google Drive both work, and sources can be tagged with custom labels and filtered by category.
Google lists support for these formats:
- PDFs
- Google Docs
- Microsoft Office files
- Plain text and Markdown files
- Web bookmarks
Once your files are indexed, a chat window answers questions from them with summaries. Live Assistance goes a step further. It listens during a call and, when a topic from your files comes up, shows a context card with exact citations. Google says this cuts prep time and the follow-up circle backs after a meeting.
Transcripts, audio and files never leave your Mac
Google says all inference in Google AI Edge Foresight runs locally on Gemma models, and that no recordings, transcripts or files are sent to cloud servers. The app is built to work offline, so it keeps going on a flight or inside an air-gapped office. Google’s launch post also says there are no cloud subscription costs.
The search side runs on EmbeddingGemma 2, which has 740 million parameters, uses the Gemma 4 architecture and ships under an Apache 2.0 license. It maps text, images, video and audio into a single vector space, so a text query can pull up a matching audio clip or image. Google’s AI Edge developer blog puts visual embeddings at 37.3 milliseconds per image on the GPU of a MacBook with an M5 Pro chip, which works out to about 26.9 images per second.
Google measured the full multimodal version of EmbeddingGemma 2 at about 567MB of active RAM on a Google Pixel 11 Pro, with the text-only weights at about 191MB.
iPhone users get a smaller taste of the same model. Google added two demos to the Google AI Edge Gallery app, Instant Media Search and Video Moments Finder, and the app is available from the App Store. Instant Media Search finds photos and videos in your library from a typed description or an example image. Video Moments Finder locates specific moments in local video, such as kids laughing or a dog catching a frisbee.
