Great question, let me break down the exact flow. The app processes files one by one, right from your machine.

Here's the step-by-step:

1. Setup: You point the tool to a local input folder (with your source files) and a local output folder (which starts empty). You select a model (like Gemini 2.5 Flash) and write your detailed system prompt. For example:

You are an experienced JSON content translator. Your task is to translate all English text in the provided JSON files into Spanish, while preserving the original JSON file structure and formatting. It is crucial that the file structure, keys, and everything else (except the text content that needs to be translated) remain unchanged. Do not add any comments or additional text, only the translated JSON file. Make sure the translation sounds natural to the target audience.

2. Processing: The tool takes file #1 from your input folder, sends its entire text content to the API along with your prompt, and waits for the response.

3. Output: It then saves that raw text response into a new file inside your output folder, using the exact same filename and extension as the original. Then it moves on to file #2, #3, and so on.

The "then what?" is you come back to a fully populated output folder that mirrors your input folder, but with every file processed by the AI according to your instructions.