In early October 2026 I used my X account @Saboo_Shubham_ to announce that Google’s Embedding Gemma 2 model had been repurposed into a super‑fast, fully‑local speech‑to‑text engine and bundled into a free macOS AI notetaker app.
On a base M3 MacBook Air it transcribes an hour of audio in about 13 seconds, achieving a 288× real time factor.
Both the model and the AI notetaker app are free, running 100% locally on the device.
I also introduced the Gemini Agent framework, noting that each task gets its own computer and that users can pick between Gemini 4 Argon or Claude Opus 5.5.