Panoptikon
A local search engine that indexes images, video and other hard-to-search files using OCR, speech-to-text, image embeddings and automated tagging, letting several models' results sit side by side at search time.
Panoptikon indexes local files with a combination of machine learning models, making media such as images and video that are otherwise difficult to search findable through natural queries. It combines OCR, speech-to-text transcription, image and text embeddings, full-text search, automated tagging and automated image captioning into one system, and is designed to be fully configurable, so a custom model or a newer fine-tune can be added with a few lines of configuration once it is supported by one of the built-in model classes.
It deliberately keeps index data from multiple different models, or different configurations of the same model, side by side rather than replacing one with another, letting a search choose which results to draw on and compare methods against each other directly. It never copies, moves or otherwise modifies the user's original files; it only needs the directories added to its allowed paths and tracks files by hash so renaming or moving them afterward does not break the index.
It is written in Rust.