byaldi

Why we included this project

Byaldi is for anyone who needs to search inside scanned documents, PDFs, and images that lack a reliable text layer. It skips OCR entirely and indexes pages as full images using late-interaction models such as ColPali and ColQwen2, then returns the pages that match your query through a small, familiar API. Loading a pretrained model, building an index from a folder, finding matching pages, and even adding documents to an existing index all fit within a few method calls. That makes it a quick way to test visual retrieval on your own documents before investing in a heavier RAG stack. Just note that it is still a pre-release library with uncompressed indexes and a recommended GPU for encoding, so it is best treated as a prototype path rather than a tuned production system.