GitHub - run-llama/liteparse: A fast, helpful, and open-source document parser
Short summary
LiteParse is an open-source PDF parser that extracts text with spatial positioning and bounding boxes entirely on your machine, running 2-5ms per page without cloud dependencies or LLMs. It handles PDFs, Word docs, spreadsheets, and images, with optional OCR for scanned content.
- Benchmarks show LiteParse outperforms free competitors (pdf-inspector, opendataloader, markitdown) across three major document parsing tests, scoring 0.389 on ParseBench vs 0.283 for the next best model-free tool.
- Speed comes from avoiding ML models — uses PDFium for text extraction, optional Tesseract or PaddleOCR for scanned pages, and spatial layout reconstruction via grid projection.
- Available in Python, Node.js, Rust, and browser (WASM) with CLI tools; includes worker pool mode for parallel parsing with per-document timeouts to prevent pipeline stalls.
Full summary
LiteParse is an open-source PDF parser built for speed and simplicity. It runs locally without cloud dependencies or LLM features, parsing text with spatial positioning at roughly 2-5ms per page. The tool handles multiple input formats (PDF, DOCX, XLSX, PPTX, images) and outputs structured markdown, JSON, or plain text with bounding boxes. It includes a flexible OCR system—Tesseract comes bundled by default, but you can plug in HTTP-based servers like EasyOCR or PaddleOCR. A complexity detection feature lets you check whether a document actually needs OCR before committing to a full parse, which saves time and resources. The project supports Rust, Python, Node.js/TypeScript, and browser-based WASM, plus a worker pool mode for true parallelism that prevents slow documents from blocking your pipeline.
The benchmarks show LiteParse outperforms other model-free tools across three major test suites. On ParseBench (2,049 documents), it scores 0.389 overall with PaddleOCR, beating pdf-inspector (0.283) and opendataloader (0.277). On opendataloader-bench, it hits 0.901 overall, slightly ahead of the commercial nutrient parser at 0.885. On olmOCR-bench, it reaches 42.2% pass rate versus pdf-inspector's 33.7%. Tables and reading order are its strengths—it handles structured content well. Charts are a known weakness across all tested tools (near-zero scores everywhere), and math rendering requires LaTeX output that none of these parsers produce.
The trade-off is clear: LiteParse excels at routine documents but hits a ceiling with complex layouts, dense tables, handwritten text, and scanned PDFs. The creators acknowledge this and point to LlamaParse, their cloud alternative, for documents that need heavier processing. If you're building a local pipeline that processes mostly native-text PDFs and standard formats, LiteParse gives you solid performance without external services. For messy real-world documents at scale, you'll likely need something heavier.