Convert PDFs to Markdown and JSON locally without the cloud
Extract text and layout from PDFs on your machine without complex setup. LiteParse accepts PDF, DOCX, XLSX, PPTX, and images, converting them into Markdown, JSON, or text that includes headings, tables, lists, images, and links. It focuses on supplying clean data to LLM agents or RAG pipelines.
Local execution and fast processing speed
The tool uses the PDFium C library to extract text at a speed of approximately 2–5ms per page. All processing happens locally, so there is no cloud dependency or need for private LLM features. Written with a Rust core, it runs on Linux, macOS (Intel/ARM), and Windows. Bindings are available for Node.js, Python, Rust, and browsers (WASM), allowing easy integration into existing projects.
OCR and complexity detection features
You can perform OCR without additional configuration using the built-in Tesseract engine, or connect external OCR servers like EasyOCR or PaddleOCR via HTTP API. The lit is-complex command allows you to quickly determine if a document requires OCR or heavier processing before parsing the entire file. This enables efficient workflows such as routing documents to different pipelines, estimating processing costs, or rejecting unprocessable documents.
Benchmark results and comparisons
In public benchmarks measured with the latest release version as of September 9, 2026, LiteParse showed strong performance among 'model-free' tools that are not model-based. It recorded an average score of 0.364 on ParseBench (2,049 documents), rising to 0.380 with Tesseract OCR and 0.389 with PaddleOCR. Under the same conditions, competing tools pdf-inspector, opendataloader, and markitdown scored 0.283, 0.277, and 0.185, respectively. It also recorded higher scores than other free local tools on opendataloader-bench and olmOCR-bench.
Installation and usage
Install via npm i -g @llamaindex/liteparse for Node.js/TypeScript, pip install liteparse for Python, or cargo install liteparse for Rust. A browser WASM version is available via npm i @llamaindex/liteparse-wasm. After installation, use lit parse document.pdf --format markdown to convert to a Markdown file, or --format json for structured JSON. Options are provided to extract specific pages, disable OCR, or include additional metadata such as vector graphics, annotations, and form fields.
Suitable targets and limitations
This tool is suitable for environments where documents must be processed quickly locally, or where cloud costs and data privacy are concerns. It is particularly useful for developers generating Markdown data for LLMs or needing document layout block and bounding box information. However, for difficult documents such as very complex tables, multi-column layouts, charts, handwriting, or scanned PDFs, the cloud-based LlamaParse may provide more accurate results. Additionally, chart data reconstruction and formula (LaTeX) output are currently not supported, and these items are evaluated near 0 points in benchmarks.