Technology

Google Made a Meeting App That Takes Notes Without Internet

Martin HollowayPublished 2month ago3 min readBased on 6 sources
Reading level
Google Made a Meeting App That Takes Notes Without Internet
Image by Pexels from Pixabay

Google has released Google AI Edge Foresight, a free test app for Mac that writes down meetings and audio files on the computer itself. Right now it runs best only on Macs with Apple Silicon, Apple's own chips. The Verge

TechCrunch called it a local-first competitor to Granola, which means everything stays on your Mac. TechCrunch The Verge, in Oct. 8 coverage, noted TechCrunch's earlier report. There is no cloud transcription. There is no upload queue. Recording and finding notes both happen on the machine.

Google's developer notes call Foresight "completely offline-first." Both the Gemma model, Google's open AI helper, and the local vector embeddings tool, which turns speech into numbers the computer can search, run on your hardware with no internet. Transcription, embedding, indexing and query all happen locally. For meetings, that removes the need for a network and keeps raw audio and text off outside servers.

The system is built around EmbeddingGemma 2. Google describes it as an open, lightweight multimodal model, which means it is shared openly, is small, and can work with different kinds of input. Google Blog The developer writeup lists it as a 740M multimodal model, open-weight, made for private, ultra-low-latency on-device retrieval, or quick private search on your own device. Google Developers Blog

EmbeddingGemma 2 allows semantic search, routing, and retrieval fully on-device, and brings robust capabilities directly to edge hardware, according to Google. Spoken words become searchable embeddings, like a library index that remembers ideas instead of titles, kept in a local vector store on your laptop. You find past meetings by meaning, not file name or keyword. Routing decides what is handled on the computer and which part to show.

The use is simple. For Mac, the app turns scribbles into full notes using offline recordings. 9to5Google A short note during a call becomes the retrieval key. The full transcript and generated notes do the rest.

The broader context here is that Foresight is also a reference implementation for the Google AI Edge stack, not just a useful tool. Read it that way. It puts an embeddings model, a local search engine, and the app together tightly enough to handle steady speech, different speakers, and quick questions within a laptop's power and heat limits. Waiting time, or latency, matters. So does persistence, or keeping notes searchable over time. A notes app tests both hard.

In my view, meetings are a practical choice. Work teams already live in recorded calls, but many avoid cloud transcription for legal, client, or security reasons. Keeping everything on the device lowers that worry. It also changes what can go wrong. Instead of wondering about storage rules on a company server, you manage disk space, index size, and moving to a new device. Those tradeoffs are familiar to engineers and easier to understand than unclear cloud handling.

Worth flagging is what this means for on-device search. A 740M open-weight model is small enough to include in an app, yet strong enough to search a growing personal collection. If it works well day to day, the same method fits more than notes. Support logs, field recordings, lab audio, and research interviews all fit the same pipeline. No new equipment is needed.

What makes this test worth watching is that success is clear to measure: transcription accuracy, retrieval quality, and resource footprint, or how much space and power it uses. Those can be tested. And if private meeting notes on your own computer become normal, private AI stops being an idea and becomes a daily habit.