A Privacy-First Meeting Companion

Google AI Edge Foresight is an experimental app for Mac that automatically enriches hand-written short notes with live transcript details. The app records the microphone and system audio, so it works offline in Google Meet, Zoom, or in-person meetings. It supports natural-language search across different media types — images, documents, transcripts, and notes. The screen is split in two: short user notes on one side and full AI-generated notes on the other — a local-first alternative to Granola-style notetakers.

Powered by EmbeddingGemma 2

At the heart of the app is the 740-million-parameter multimodal embedding model EmbeddingGemma 2. It places text, images, video frames, and audio into a single shared vector space and is distributed under the Apache 2.0 license. For text, it runs in ~191 MB of active RAM, and the full multimodal variant in ~567 MB (Google Pixel 11 Pro measurements). On the MacBook M5 Pro GPU, computing the visual embedding of a single image takes as little as 37.3 milliseconds. The model is compressed with INT4 and INT8 quantization; the vector size is 768, reduced to 128–512 with Matryoshka representation learning. The assistant's chat capability is powered by the Gemma 4 language model.

"Google AI Edge Foresight is an experimental app for Mac: a context-aware meeting companion that runs directly on your device. With fully on-device AI processing, Foresight continuously assists with taking notes, indexing, and recalling conversation transcripts and your personal files," says the Google Developers blog.

Personal Knowledge Base and Live Assistance

Users can build a personal knowledge base from PDF, Google Docs, Microsoft Office formats, plain text, Markdown, and web bookmarks. The chat window answers questions with citations based on meeting notes and documents. The Live Assistance feature detects when a topic from the user's files is mentioned during a call and displays a context card with an exact citation on screen.

Pricing, Platform, and Market Position

According to Google's FAQ page, Foresight is optimized for Apple Silicon and works fully offline. The app is free, with no cloud subscription required: all transcripts and audio processing stay on the device. TechCrunch reports that Google may in the future release a consumer version that works on video calling platforms through Gemini apps and relies on cloud models. The app was developed by the same Google team that created the local AI dictation tool released in April 2026. The market is crowded: in recent months, dictation company Wispr and scheduling service Calendly launched their own notetakers, while Superhuman (formerly Grammarly) acquired Y Combinator-backed Fathom last month. Google is offering to download the app today via the AI Edge page. The company plans to introduce EmbeddingGemma 2 as a service for Android via ML Kit in the coming weeks.