# LlamaIndex Rewrites LiteParse in Rust for v2.0 — 5–100× Faster Open-Source PDF and Office Parser, 457-Page Doc in 0.777 Seconds, Native + Python + Node + WASM

> LlamaIndex rewrote its open-source PDF parser in Rust — same CLI, 5–100× faster, runs on Linux, macOS, Windows, Python, Node, and in the browser via WASM.

- Org: LlamaIndex
- Date: 2026-05-27
- Category: repo, tool
- Source: https://www.llamaindex.ai/blog/liteparse-v2-0-runs-everywhere
- Canonical page: https://ai-tldr.dev/releases/llamaindex-liteparse-v2/

## In short

LiteParse v2.0 swaps a TypeScript parser for a Rust core wrapping a custom PDFium build. Same CLI, 5–100× speedup on small docs, 0.777s for a 457-page PDF, with native, Python, Node, and WASM bindings shipping side-by-side.

## What it is

LiteParse is a local, model-free document parser from LlamaIndex covering PDFs plus 50+ Office and image formats with built-in OCR and screenshot output for agents. Version 2.0 is a full rewrite from TypeScript to a Rust core that ships as a CLI, a Rust library, a Python package, a Node.js binding, and a WASM build — all sharing the same parser. It's a sibling to the hosted LlamaParse but stays entirely on your machine: no LLM calls, no API keys, no cloud.

## How it works

The new core wraps a forked-and-rebuilt PDFium for text + spatial coordinates and uses tesseract-rs for OCR; the Rust binary auto-detects when OCR is needed via text-coverage metrics so it skips unnecessary work. Office documents are converted via LibreOffice on Linux/macOS/Windows. Python and Node bindings are generated through PyO3 and napi-rs; the WASM target exposes the same parser with a callback-based OCR hook so browsers and edge runtimes (Cloudflare Workers, Vercel Edge) can parse PDFs without spawning a process.

## Why it matters

Every retrieval pipeline, agent, and ingest job has a PDF problem. LiteParse v2.0 turns a slow ETL step into a sub-second one — a 457-page, 100MB PDF parses in 0.777 seconds — without sending the document to a cloud parser or feeding pages to an LLM. The WASM build also unlocks parsing fully in the browser, which matters for privacy-sensitive RAG and on-device coding agents.

## Who it's for

RAG engineers, indie devs shipping document-aware agents, anyone running a self-hosted ingest pipeline.

## Try it

cargo install liteparse  # or:  pip install liteparse

## Links

- [Announcement Post](https://www.llamaindex.ai/blog/liteparse-v2-0-runs-everywhere)
- [GitHub Repo](https://github.com/run-llama/liteparse)
- [Docs](https://developers.llamaindex.ai/liteparse/)
- [PyPI Package](https://pypi.org/project/liteparse/)