LiteParse PDF Parser: Fast Local Document Parser

What is LiteParse

LiteParse is a standalone Rust PDF parser that extracts spatial text with bounding boxes, converts documents to Markdown, and includes bundled Tesseract OCR — all without a cloud dependency or API key. It targets developers who need fast, structured PDF text extraction in their own pipeline rather than a managed service.

Key features

• ✓Fast spatial text parsing via PDFium (~2-5ms per page for raw text extraction)
• ✓Bundled Tesseract OCR — zero setup required; also accepts any HTTP OCR server via a simple API spec
• ✓Markdown, JSON, and plain text output with precise bounding box coordinates per text item
• ✓Layout block classification (headings, paragraphs, tables, list items, code, figures) with per-block bounding boxes
• ✓Complexity detection — cheaply check whether a document needs OCR before committing to full parse
• ✓Screenshot generation for individual pages — useful for agent-based document workflows
• ✓Worker pool mode in Python and Node.js — rogue documents get killed by name with hard timeouts, preventing pipeline stalls
• ✓Multi-language: Rust library, Python (PyO3), Node.js/TypeScript (napi-rs), and browser WASM

Installation

Python: pip install liteparse | Node.js: npm i -g @llamaindex/liteparse | Rust: cargo install liteparse (CLI) or cargo add liteparse (library) | Browser: npm i @llamaindex/liteparse-wasm

Using LiteParse from the CLI

The `lit parse` CLI parses a PDF and outputs structured text: bash # Basic parse — outputs text to stdout lit parse document.pdf # Markdown output with layout reconstruction lit parse document.pdf --format markdown -o output.md # Specific page range lit parse document.pdf --target-pages "1-5,10,15-20" # JSON with bounding boxes and layout blocks lit parse document.pdf --format json --extract-blocks -o output.json # Disable bundled OCR lit parse document.pdf --no-ocr From Python: python from liteparse import parse result = parse("document.pdf", extract_blocks=True) From Node.js: javascript import { parse } from "@llamaindex/liteparse"; const result = await parse("document.pdf", { extractBlocks: true });

How LiteParse compares to alternatives

Frequently asked questions

The problem it solves

Local PDF parsing tools often produce flat text or require LLM-based pipelines, and it's hard to know whether a given document actually needs OCR before spending cycles on it. Teams also need spatial coordinates for downstream tasks like citation matching, but getting bounding boxes from local tools is either absent or requires heavyweight setups.

Best use cases

• •Extracting structured text with bounding boxes from native PDFs for citation matching or redaction pipelines
• •Converting clean PDFs to Markdown for ingestion into RAG systems without a cloud API call
• •Adding OCR to a document pipeline via the built-in Tesseract integration or plugging in PaddleOCR via HTTP
• •Batch-processing PDFs at speed in a Node.js or Python service where cloud dependencies are unacceptable
• •Generating page screenshots for LLM agents that need to 'see' a document alongside its text
• •Pre-filtering documents with complexity detection to decide whether to route to a heavier parser or reject early

Who should try it — and who should skip

Reach for LiteParse when you process clean-to-moderate PDFs and need speed, bounding box data, or an OCR-capable pipeline without cloud dependencies. The bundled Tesseract makes it particularly useful for teams that want OCR without infrastructure. If your documents are heavy on dense tables, complex layouts, charts, or scientific equations — or if accuracy on edge cases outweighs local processing — use LlamaParse instead and pay for the cloud. LiteParse is not trying to win on quality for the hardest documents; it is trying to win on speed and simplicity for the common case.

Related repositories