Engine Reference

Engines registered per capability in the engine registry (natural_pdf.engine_registry.list_engines). “Requires” shows the install hint carried by the registration metadata; “included” means the registration declares no extra install requirement (availability may still depend on optional packages — run npdf doctor to check your environment).

This page lists the built-in engines only. Third-party plugins registered through the natural_pdf.engines entry point extend these lists at runtime in your environment; they are deliberately not part of the generated reference.

Capability: checkbox

Engine	Requires	Notes
default	included	options: DefaultCheckboxOptions
onnx	pip install onnxruntime huggingface_hub	options: OnnxCheckboxOptions
vector	included	options: VectorCheckboxOptions
vlm	pip install openai	options: VLMCheckboxOptions

Capability: classification

Engine	Requires	Notes
default	included

Capability: deskew (also deskew.apply, deskew.detect)

Engine	Requires	Notes
hough	included
projection	included
standard	included

Capability: guides.detect

Engine	Requires	Notes
builtin	included
builtin.content	included
builtin.headers	included
builtin.lines	included
builtin.stripes	included
builtin.whitespace	included

Capability: layout

Engine	Requires	Notes
doclayout	pip install transformers torch	options: DocLayoutOptions
gemini	included	deprecated alias for vlm; options: VLMLayoutOptions
paddle	pip install “natural-pdf[paddle]” or pip install paddleocr	options: PaddleLayoutOptions
surya	pip install “surya-ocr<0.15”	options: SuryaLayoutOptions
tatr	included	options: TATRLayoutOptions
vlm	included	options: VLMLayoutOptions
yolo	pip install “natural-pdf[all]” or pip install doclayout_yolo	options: YOLOLayoutOptions

Capability: ocr (also ocr.apply, ocr.extract)

Engine	Requires	Notes
chandra2	pip install chandra-ocr[hf]
doctr	pip install python-doctr
easyocr	pip install easyocr
paddle	pip install “natural-pdf[paddle]” or pip install paddleocr
paddlevl	pip install “natural-pdf[paddle]” or pip install paddleocr
rapidocr	pip install “natural-pdf[all]” or pip install rapidocr
surya	pip install surya-ocr

Capability: selectors

Engine	Requires	Notes
native	included

Capability: tables

Engine	Requires	Notes
lattice	included
pdfplumber	included
pdfplumber_auto	included
stream	included
tatr	included
text	included

Capability: tables.detect_structure

Engine	Requires	Notes
tatr	included

OCR dispatch registry

The unified OCR registry behind page.apply_ocr(engine=...). Named VLM-family entries resolve to local vision-language models instead of classic OCR engines; the generic vlm entry can use either a local model or a configured remote client.

Engine	Type	Requires	Notes
chandra	vlm_shorthand	included	VLM family: chandra
chandra2	classic	pip install chandra-ocr[hf]	options: ChandraOCROptions
doctr	classic	pip install python-doctr	options: DoctrOCROptions
dots	vlm_shorthand	included	VLM family: dots_mocr
easyocr	classic	pip install easyocr	options: EasyOCROptions
glm_ocr	vlm_shorthand	included	VLM family: glm_ocr
paddle	classic	pip install “natural-pdf[paddle]” or pip install paddleocr	options: PaddleOCROptions
paddlevl	auto_platform	pip install “natural-pdf[paddle]” or pip install paddleocr	VLM family: paddlevl; options: PaddleOCRVLOptions
rapidocr	classic	pip install “natural-pdf[all]” or pip install rapidocr	options: RapidOCROptions
surya	classic	pip install surya-ocr	options: SuryaOCROptions
vlm	vlm_generic	included