run-llama / liteparse

项目摘要

LiteParse is a fast, open-source PDF parsing tool built in Rust by the LlamaIndex team, designed for high-performance document ingestion into LLM/RAG pipelines. It offers spatial text extraction with bounding boxes, a pluggable OCR system (Tesseract built-in, with HTTP server support for EasyOCR/PaddleOCR), complexity detection, and multi-language bindings for Rust, Python, Node.js, and WASM.

为什么值得关注

With 12.6K stars and 93 releases in just 6 months, LiteParse demonstrates exceptional community traction and development velocity. Its ~2-5ms per page parsing speed, worker pool mode for true parallelism, and seamless multi-language support make it a standout choice for developers building document-heavy RAG pipelines who need fast, local parsing without cloud dependencies.

优势

• Exceptional parsing speed (~2-5ms per page) using PDFium with Rust backend
• Multi-language support (Rust, Python, Node.js, WASM) from a single codebase
• Flexible, pluggable OCR system with Tesseract built-in and HTTP server API for custom OCR backends
• Worker pool mode enabling true parallelism and per-parse timeouts for production workloads
• Active development with 93 releases in 6 months and strong community adoption

局限性

• No automated tests despite high production claims, reducing reliability confidence
• No code examples or tutorials in the repository, requiring users to rely solely on external docs
• Complex documents (dense tables, multi-column, charts, handwritten text) are explicitly deferred to the paid LlamaParse cloud service, limiting the OSS version's standalone utility

使用场景

• Fast document ingestion pipelines for RAG and LLM applications
• Batch PDF text extraction with spatial bounding box metadata
• OCR processing of scanned documents with pluggable engine selection
• Pre-processing documents to determine parsing complexity before committing to expensive cloud parsing