GOOGLE has released an experimental offline note‑taking and personal knowledge base tool for Mac called Google AI Edge Foresight. The software is optimised for Apple Silicon and runs entirely on the user’s local machine, with no cloud processing or ongoing subscription fees. According to the piece, Foresight captures audio from meetings (system and microphone audio) and automatically drafts, expands, and structures notes in real time. It supports online meetings as well as in‑person discussions and is presented as an offline-first alternative to cloud transcription services.
At the core, Google describes a local edge‑computing pipeline built around two models: EmbeddingGemma 2 for retrieval and Gemma 4 for content generation. EmbeddingGemma 2 is a multimodal embedding model with 740 million parameters designed to vectorise transcripts and imported documents, enabling semantic search without exact keyword matches. After retrieval, Gemma 4 handles generating fluent text and structured notes.
The article stresses that all processing occurs on the user’s Mac, with sensitive data never uploaded to Google’s or any cloud servers, addressing enterprise privacy concerns often raised about AI tools. The piece positions Foresight as a technology showcase for the feasibility of on‑device RAG and semantic search, and suggests potential future integrations with Google Gemini for Workspace. There is no discussion of observed exploits or CVEs tied to the tool itself in the article.