LiteParse
github.com/run-llama/liteparse- Category
- AI Tools
- Rank
- No. 123Tools index
- Type
- APP
- Builder
- run-llama
- GitHub
- 12.3k stars
- Latest release
- wasm-v2.14.5
- Date
About
A fast, open-source document parser from LlamaIndex for turning files into clean, model-ready text.
What it does
LiteParse reads PDFs locally, preserves page geometry and bounding boxes, and reconstructs spatial content as text, JSON, or Markdown. It can convert Office files and images before extraction, apply bundled or remote OCR, render page screenshots, and expose structured PDF details.
Why it's ranked here
LiteParse offers an unusually broad local parsing surface without requiring cloud services, API keys, or language models. Its Rust core reaches command-line, server-side, and browser workflows. The verdict is tempered by explicit quality limits on dense tables, multiple columns, charts, handwriting, and scans.
What's good
Complexity detection can flag pages needing OCR or heavier parsing before the full job runs. Selective OCR supports bundled Tesseract and compatible HTTP servers. Optional outputs include embedded images, vector graphics, annotations, form fields, tagged structure, word boxes, and page-level errors.
Tradeoffs
Markdown reconstruction relies entirely on rules and heuristics, so difficult layouts may render incorrectly. Office documents and images require conversion before PDF extraction. Detailed options increase payload or processing costs: word boxes roughly double text-item payloads, while complexity signals require an extra detection pass.
How to use it well
Use LiteParse for local document ingestion, coding agents, real-time applications, and retrieval pipelines where speed, privacy, geometry, or predictable offline operation matters. Run complexity detection first, route simple pages through local extraction, and reserve stronger external parsing for dense tables, multiple columns, charts, handwriting, or stubborn scans. It does not provide proprietary language-model interpretation.
Technical notes+
Cargo.toml defines an Apache-2.0 Rust workspace containing the core parser plus PDFium, N-API, PyO3, and WASM crates. packages/node/src/lib.ts wraps the native binding with typed configuration and structured results, while packages/node/native.d.ts exposes generated N-API types. packages/node/src/cli.ts implements parsing, complexity checks, screenshots, stdin input, configuration overrides, and JSON, text, or Markdown output. packages/python/liteparse/__init__.py exports the Python parser and result types, and packages/python/liteparse/cli.py delegates the command-line entry point to the native extension. Makefile runs workspace tests, Clippy, formatting, and builds.
Observed
- License
- Apache-2.0
- Primary implementation
- Rust workspace with PDFium-based extraction
- Install surface
- Published through Cargo, npm, PyPI, and a separate npm WASM package
- Interfaces
- Command-line tool and libraries for Rust, Node.js and TypeScript, Python, and browser WASM
- Platform support
- Linux, macOS on Intel and ARM, and Windows
- Output formats
- Markdown, structured JSON, layout-preserved text, and PNG page screenshots
- Runtime model
- Runs locally without cloud dependencies, language models, or API keys
- OCR surface
- Bundled Tesseract plus a standard HTTP interface for external OCR servers
Read from README.md, Makefile, Cargo.toml, packages/node/native.d.ts, packages/node/src/cli.ts, packages/node/src/lib.ts, packages/python/liteparse/cli.py, packages/python/liteparse/__init__.py, docs/src/content/docs/liteparse/index.md, docs.config.mjs, eslint.config.js, scripts/bump-version.py, scripts/strip-impls-from-api-docs.py, scripts/generate_filled_acroform_fixture.py, ocr/easyocr/server.py.
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.