featured · github
MinerU: Turn Messy PDFs Into LLM-Ready Markdown
Open source document parser that extracts text, tables, and layout from complex PDFs and Office files so your agents can actually understand them.
opendatalab/MinerU ↗PDFs are a mess for AI. A scanned contract, a spreadsheet embedded in a presentation, a multi-column research paper, tables with merged cells, handwritten notes in margins, figures that span pages, footnotes scattered across layouts. Most document parsers choke or strip out the structure that matters. MinerU is built to handle exactly this chaos. It reads complex layouts and outputs clean markdown or JSON that your LLM can parse without confusion. Feed it a document, get back structured text with tables intact, images referenced, columns separated. For anyone building RAG systems, document automation workflows, or agents that need to reason over real-world PDFs, this cuts the preprocessing friction significantly. Open source, so you own the pipeline.
Share kit
MinerU: PDF parsing that actually preserves document structure
Complex PDFs are poison for RAG and agentic workflows. MinerU extracts text, tables, and layout into LLM-ready markdown/JSON. Open source, designed for the messy real-world documents that other parsers fail on. If your pipeline is drowning in preprocessing, worth testing today.
PDFs are chaos for AI. Merged table cells, multi-column layouts, embedded images, footnotes scattered everywhere. MinerU is an open source parser built for exactly this. Outputs clean markdown or JSON your agents can actually reason over. Cuts preprocessing friction for RAG pipelines and document automation.
Document-heavy workflows (legal, finance, research) live in PDFs. Most parsers strip structure. MinerU preserves it, outputting LLM-ready markdown and JSON from complex layouts, tables, and multi-column text. If your agents are struggling with preprocessing, this is builder-useful. Open source. Worth a sprint to evaluate.
Just integrated MinerU into a document workflow and it cut preprocessing time by half. PDFs and Office docs go in, LLM-ready markdown comes out. No more manual formatting before feeding into RAG systems or agents. The real win: complex layouts (tables, multi-column text, mixed media) parse cleanly instead of turning into garbage tokens. Built on deep document understanding, not just regex and string splits. If you're building anything that touches legal docs, financial statements, or research papers, this saves the debugging tax. One command, clean output. GitHub: https://github.com/opendatalab/MinerU
spent too long hand-parsing PDFs into LLM format. tried MinerU. complex layouts actually work now. tables stay tables. columns don't merge. output is clean markdown/JSON. if your workflow touches documents, this is the move. https://github.com/opendatalab/MinerU