$ show pymupdf/pymupdf4llm
pymupdf4llm is a library and CLI tool that converts PDFs (and other documents) into clean Markdown, JSON, plain text, or layout-aware chunks specifically optimized for LLM and RAG workflows. It provides a high-level API (`to_markdown`, `to_json`, `to_text`, `to_chunks`) that internally uses either a legacy heuristic approach or an optional GNN-based layout parser (via `pymupdf-layout`) to extract text, tables, images, and reading order. The repository also includes a deterministic chunking pipeline that uses only PDF-native layout signals (box boundaries, font jumps, vertical gaps) to produce context-preserving chunks without embeddings. OCR support is built-in, with a hybrid decision model (ONNX-based) that selectively applies Tesseract, RapidOCR, or PaddleOCR to pages that need it. Framework integrations for LlamaIndex and LangChain are provided. The architecture is primarily a monolithic library with two parallel conversion paths (legacy and layout-based), a chunking subpackage, an OCR subsystem with abstract backends, and a batch converter for large-scale processing. While the code is generally well-structured, the presence of two parallel conversion paths and tight coupling to PyMuPDF internals introduce some maintenance risk.
The LLM identified 5 files worth highlighting for reuse:
A self-contained subpackage for splitting PDF documents into layout-aware chunks (sections, tables, figures) without embeddings. Consists of models (`Chunk`, `ChunkMetadata`, etc.), a sentence builder, boundary scorer, chunk assembler, and serializer. Can be extracted and reused in any project that needs structured chunking from parsed document output.
Defines a clean interface (`exec_ocr`) that abstracts over multiple OCR backends (Tesseract, RapidOCR, PaddleOCR). The module auto-detects installed engines and routes calls accordingly. Isolates OCR logic from the rest of the PDF pipeline, making it easy to swap or extend backends.
Provides a standalone function `to_html(pdf, page_index)` that converts PDF tables into HTML markup. Uses PyMuPDF's `find_tables` with layout-aware settings. The output is suitable for rendering in browsers or LLM context. The module is independent of the rest of the library and can be reused in any PDF processing workflow.
Contains the `find_reading_order` function that reconstructs a sensible reading order from layout boxes (text, tables, figures) after the GNN-based layout parser. Implements a custom algorithm that clusters stripes and handles multi-column layouts. This is a non-trivial geometric algorithm that could be reused in other document layout analysis projects.
Implements a column detection algorithm that identifies multi-column layouts on a page using geometric analysis of text blocks. It is used by the reading order logic to properly order content across columns. The algorithm is self-contained and could be extracted for other layout analysis tasks.