Google has officially released Google AI Edge Foresight, an experimental offline AI note-taking app for Mac that directly competes with Granola. The application relies entirely on the device-side 740-million-parameter EmbeddingGemma 2 model for meeting transcription and summary generation, requiring no cloud connection. In addition to supporting split-screen manual notes and AI auto-note-taking, users can upload multi-format documents to build a local knowledge base and interact via the Gemma 4 assistant.
Cross-verified sources indicate that Foresight’s core breakthrough is its fully offline operation, using small-parameter models to achieve low-latency, high-privacy meeting transcription. Its feature set deeply mirrors Granola, offering split-view interfaces, real-time transcription, and intelligent Q&A based on local knowledge bases. Currently, the tool is optimized only for macOS devices with Apple Silicon chips, with no support yet for Windows or mobile platforms.
This move marks Google’s further expansion into edge-side AI productivity tools, aiming to enter the competitive AI note-taking market through localization advantages. The successful deployment of EmbeddingGemma 2 demonstrates the viability of small-parameter models for specific tasks, providing a technical paradigm for future offline AI applications.