google-gemini/jot-gemini-transcribe-macOS

What it solves

Jot provides a smart dictation experience for macOS that converts spoken words into polished, punctuated text and inserts it directly at the cursor. It eliminates the need to switch windows or manually copy transcripts, while handling common speech patterns like self-corrections (e.g., changing a time from 1pm to 2pm) and specialized jargon that standard dictation often misses.

How it works

Jot uses the gemini-3.5-transcribe model via the Gemini API. When a user holds a specific key (like fn), the app captures audio to disk immediately for crash safety. Once the key is released, the audio is converted to FLAC and sent to the Gemini API. The resulting text is processed through a formatting pipeline—which can include tone matching and a custom dictionary for technical terms—before being inserted into the active application using the macOS Accessibility API or a guarded paste mechanism.

Who it’s for

macOS users who want a high-accuracy, low-friction dictation tool that integrates seamlessly into any application they are using without requiring a separate account or middleman server.

Highlights

• Intelligent Editing: Automatically removes filler words and handles "changes of mind" in speech to produce clean final text.
• Privacy-First: Voice data is sent directly from the Mac to the Gemini API using the user's own API key, with no intermediate servers or analytics.
• Crash-Resistant: Audio is written to disk from the first millisecond, ensuring recordings are recovered even after a crash or power loss.
• Custom Dictionary: Allows users to add specific names and product terms to ensure technical jargon is transcribed correctly.
• Low Latency: Pre-warms the capture graph while idle to minimize the delay between key press and recording start.