Google AI Edge Foresight: Offline Meeting Notes on a Mac

Google AI Edge Foresight is a new Mac app that takes meeting notes without sending anything to the cloud. TechCrunch reported the release on October 8, 2026, describing a note-taker that runs Google's small open models directly on Apple Silicon. For anyone following on-device AI, it is a practical demonstration of how much everyday work a laptop can now handle locally.

• What it is: a Mac meeting-notes app from Google's AI Edge team.
• Models: EmbeddingGemma 2, at 740 million parameters, plus an assistant powered by Gemma 4.
• Offline: Google's FAQ, as quoted by TechCrunch, says it is optimized for Apple Silicon and can work completely offline.
• Knowledge base: accepts PDFs, Google Docs, Microsoft Office files, plain text, Markdown and web bookmarks.

What Does Google AI Edge Foresight Do?

According to TechCrunch, Foresight uses a split-screen layout. On one side you type your own shorthand notes during a meeting. On the other, the app produces fuller AI-generated notes from the conversation. You can open the transcript when you need the exact wording, and you can ask the built-in assistant questions about what was said.

The app is not tied to one video platform. TechCrunch notes it captures notes across apps and also works for in-person meetings, which makes it useful for interviews, lectures and workshops as well as calls.

The feature that stands out is the knowledge base. You can load reference material in several common formats, and when a related point comes up in a meeting, the app surfaces the matching information in real time. That is retrieval running live on the laptop, not a search you run afterward.

Which Models Run on the Device?

Two models share the work. EmbeddingGemma 2, a 740-million-parameter embedding model, handles the matching between what is being said and what is in your documents. We covered that model's design in our article on EmbeddingGemma 2 and on-device multimodal search. The chat assistant is powered by Gemma 4, Google's open model family that we have also seen running faster on Apple Silicon through Ollama.

This division of labor is a sensible pattern for local software. A compact embedding model is fast and light enough to run continuously, while a larger generative model is called only when the user asks a question or needs a summary. TechCrunch's report does not specify which Gemma 4 variant the assistant uses.

Why Does Offline Meeting Transcription Matter?

There are three clear benefits to keeping this work on the device.

• Privacy: meeting audio and notes do not need to leave the computer.
• Availability: the app works on a plane or anywhere without a connection.
• Cost: there is no per-minute cloud processing involved.

TechCrunch frames the app as an entry into a category made popular by note-takers such as Granola. The local-first approach is what distinguishes it, and it comes from the same team that released an offline dictation app in April 2026.

What Is Not Yet Known?

Several practical details are still open. TechCrunch's report does not state a price, and it does not list the minimum macOS version, memory requirements or supported languages. Google's developer page for Foresight is the place to check before installing.

It is also worth being clear about scope: this is a release from Google's edge AI team for the Mac, not an announcement of a note-taker inside the consumer Gemini apps.

The broader point holds regardless. A year ago, real-time retrieval and summarization during a live conversation was firmly a cloud task. Seeing it packaged as a desktop app built on open models suggests that private, offline assistants are becoming an ordinary software category. More on local models and tools is in our AI coverage.

Sources: TechCrunch — October 8, 2026; Google for Developers — accessed October 12, 2026; AI Weekly — October 2026.