iceener/voice-to-notes

This is a small background program for Node.js that watches a folder where voice recordings are saved, such as the ones created by the Superwhisper app on a Mac, and turns each new recording into a Markdown note using Google's Gemini AI model. It works with any app that drops audio files into a folder, not just Superwhisper. The tool is organized around categories that the user defines in a configuration file. Each category has a description used to sort incoming recordings, one or more prompts that tell Gemini what to do with the audio, and a folder where daily notes get saved. If only one category is set up, every recording goes straight to it with no extra cost. With multiple categories, an extra Gemini call listens to the recording and decides which category fits based on the descriptions provided. Some categories can also keep a rolling memory, a Markdown file that gets updated after every recording and is included in future requests for that category. This lets feedback build on what came before instead of starting fresh each time. The memory file has a size limit, and when it grows too large, the app asks Gemini to shorten it while keeping the important parts. Out of the box, the tool comes configured as an English speaking coach. Each recording gets personalized feedback on pronunciation, rhythm, and phrasing, with a running profile of the user's progress that helps the coaching improve over time. The same underlying pipeline, though, could just as easily route recordings toward capturing ideas, extracting tasks, or keeping a journal, since all of that is controlled through the prompts and category descriptions rather than fixed behavior. Setup involves installing the Node dependencies, adding a Gemini API key, and pointing the tool at the folder where recordings are saved.