Why it mattersPush-to-talk dictation that rewrites in real time when you change your mind, and never touches a third-party server, is the first Mac dictation option a developer can hand a security review.
Jot is a macOS dictation app that uses Google's Gemini 3.5 Transcribe model. It shipped on 19 August in the google-gemini GitHub organisation under Apache 2.0, is at 422 stars as of today, and its README states plainly that "this is not an officially supported Google product." The author is Ammaar Reshi.
Hold the fn key, speak, and Jot types the transcript into whatever app the cursor is in. The key is rebindable if fn is spoken for. That is the whole interaction, and the onboarding walks a first-time user through it in about two minutes: paste a Gemini API key from Google AI Studio, allow the microphone, allow Accessibility, then hold and talk.
Four things it does that the built-in macOS dictation does not
The README lists four differences worth reading against the built-in dictation. Each is small in isolation and useful together.
Change of mind. Say "let's meet at 1pm, actually, no, make it 2pm" and Jot writes "Let's meet at 2pm." The onboarding forces the user to try this once so the pattern is obvious.
Crash-safe capture. Audio is written to disk from the first millisecond, so a crash, a kill -9, or a flat battery does not cost the dictation. Offline dictations queue and land when the machine reconnects. Every failure retries from History.
Private by architecture. The user's voice goes from the Mac straight to the Gemini API with the user's own key. No middleman server, no account, no analytics, no screenshots, no keystroke logging. The PRIVACY.md file in the repository documents the single network host the app talks to.
Jargon in a dictionary. Names and product terms go in a Dictionary that rides along with the audio, so the model sees "Kubernetes" instead of a phonetic guess. Tone matching for email, chat, and code is a separate setting.
The billing runs through the user's own API key
Cost is the standard Gemini API price for what the user dictates, billed to the key in Google AI Studio, with a free tier that the app relies on for onboarding. The person paying can see the usage in the Google AI Studio console, because it is their own key.
For a developer who already writes email, Slack, and code comments by voice on a Mac, the change-of-mind rewrite and the Dictionary are the two features the built-in macOS dictation fails on. Inside a company, the single-host architecture is short enough to describe to a security review in one paragraph.
The project is under Google's google-gemini organisation on GitHub, and the README is clear that Google does not officially support it. That is the arrangement to check with a security or procurement team before rollout: it is a supported model with an unsupported demo client, released by an author under his own name, with the code readable in the repository. The install path is a Jot-x.y.z.dmg from Releases, dragged into Applications, and the setup requires Microphone and Accessibility permissions on macOS.
Source
google-gemini/jot-gemini-transcribe-macOS on GitHub. Star count and repository dates verified via the GitHub REST API, 2026-09-06. The model is documented at ai.google.dev/gemini-api/docs/models/gemini-3.5-transcribe.
Source: GitHub
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually shipping with them. Short, and only when there is something worth reading.