Google releases AI Edge Foresight, an offline meeting notes app for Mac

Google released Google AI Edge Foresight, an experimental meeting notes app for Mac, on October 6, 2026. According to the announcement on Google's developers blog, the app listens to a meeting and transcribes it, enriches the short notes a user types with details from that transcript, and searches the user's own files. All AI processing happens on the computer itself, and the app also works without an internet connection.

Google AI Edge is Google's developer platform for running AI models on devices such as phones and computers. Google presents Foresight as one of the apps that show what can be built with this platform. Under the hood it uses EmbeddingGemma 2, the embedding model Google DeepMind recently released with its trained model files open for anyone to download, together with Google's Gemma 4 language model.

Short notes are completed with the meeting transcript

According to the product page, Foresight listens to the meeting in the background. The user only jots down the points they want to remember as a few words each, and the app turns these bullets into organized notes using details from the transcript. Because Foresight works with the computer's microphone and system audio, it can run alongside any meeting platform. Google gives Google Meet and Zoom as examples on the page, and the app can also be used in face-to-face meetings.

With the Live Assistance feature turned on, Foresight listens to the conversation, detects questions as they are asked and suggests answers drawn from the user's files. When a topic from the imported files comes up, context cards appear on screen with citations showing where the information came from. Users can also ask questions about their files and meeting transcripts in a chat window and request summaries.

The app builds a personal knowledge library from the user's files

Users can point Foresight to project folders, reference materials, diagrams and calendar details, and connect Google Drive in addition to local folders. Supported formats include PDF, Google Docs, Microsoft Office files, plain text, Markdown and web bookmarks. Sources can be given custom labels and filtered by category. Search works by meaning: because EmbeddingGemma 2 represents text, images, video and audio in the same numerical space, a single query can search across documents, images, transcripts and notes together. Gemma 4 then writes answers and summaries from what is found. According to Google, this setup forms a search and answer pipeline that runs entirely on the device with no network latency.

Data stays on the Mac and the app works offline

According to Google's FAQ, all AI inference runs locally with on-device Gemma models, and meeting audio, notes, transcripts and knowledge sources are not sent off the computer. Google says the app keeps all of its functions without internet access, for example on a plane or in secure facilities cut off from outside networks. The app is optimized for Mac computers with Apple Silicon, and Google says compressed Gemma models keep memory use and battery drain low. The installer can be downloaded from the product page, and according to Google the app requires no cloud subscription.

Foresight is offered as an experimental app. The product page does not say which languages it can transcribe, nor the minimum macOS version or amount of memory required, so it is worth testing with your own recordings before using it in meetings held in other languages. The page also does not explain in detail how connecting online sources such as Google Drive works. As with any tool that listens to and transcribes meeting audio, informing participants and following workplace policies is the user's responsibility. For a fuller picture, follow evaluations from independent researchers as well.

Sources: Google Developers Blog: Google AI Edge and EmbeddingGemma 2 announcement, Google: AI Edge Foresight product page.