what is shuttleslide fr?

Shuttleslide is a Python library and command line tool that converts PowerPoint files into HTML and back again, aiming to preserve as much of the original formatting as possible on each trip between the two formats. Unlike tools that turn slides into flat images, it keeps real, editable elements on both sides, such as text boxes, tables, and native PowerPoint shapes, and it embeds extra metadata in the HTML so that converting back to PowerPoint restores the original coordinates, fonts, and structure. The project offers three separate capabilities. The first turns a pptx file into a single HTML file, with a choice of three layouts: a scrollable page suited to search indexing or feeding into a retrieval system, an editor style preview meant to look close to the original slide, and an interactive slideshow with keyboard navigation and transitions, which is the default. The second direction converts HTML back into a pptx file using a rule based pipeline with no AI involved, producing shapes that stay fully clickable and editable inside PowerPoint itself. This direction only works reliably on a limited subset of HTML the project calls Slide HTML, described in its own documentation. The third capability skips source files altogether: given just a topic, it can generate a full slide deck using any OpenAI compatible language model, including options like Zhipu GLM, DeepSeek, or a locally run Ollama model, with an optional web interface that pauses at each step for a person to approve the result. Installation is through pip, with optional extras for the AI generation feature and the web review interface, plus a one time setup step to install a headless Chromium browser that the tool uses internally for layout work. Python 3.9 or newer is required. The README positions the project as different from similar tools mainly because it supports both conversion directions with format preserving metadata, rather than just one direction, and because the core pptx and html conversion pipeline works entirely offline without needing any language model. It is released under the MIT license.