Pdfstract

PDFStract is a data preparation tool for retrieval augmented generation pipelines. It takes PDFs through three steps, extract, chunk and embed, and exposes them through a single Python API and command line, so a document can go from PDF to vector ready output in one command. A single parameter switches between libraries, chunkers and embedding providers. For extraction it wraps more than ten PDF conversion libraries, including pymupdf4llm, markitdown, marker, docling, unstructured, pytesseract, paddleocr, deepseek OCR and mineru, and the documentation compares their speed, quality, OCR support and best use, from simple text PDFs to scanned documents, complex layouts and scientific papers. An auto setting selects the best available library. Chunking offers more than ten methods, from simple token based splitting to semantic chunking powered by AI. Embedding supports several providers including OpenAI, Sentence Transformers and local models. Installation comes in tiers, from a base package to standard, advanced and all, installed with pip, and a convert chunk embed command handles the whole flow. It is meant for developers and data engineers building document search, question answering and other AI applications who want to test several extraction and chunking strategies without rewriting their code for each one. The project is also available on GitHub. The site publishes no pricing and lists no paid plans. It is installed with pip and run by the user on their own machine or server.