Practical Docling Examples

Clean, readable Markdown from any digital PDF.

When to use: Use when you need clean, readable Markdown from a digital or mixed PDF.

CLI

Python

from docling.document_converter import DocumentConverter

converter = DocumentConverter()
result = converter.convert("report.pdf")
print(result.document.export_to_markdown())

Flags used

• --to md Selects Markdown output.

Expected output

# Annual Report

## Revenue

| Year | Revenue |
|---|---:|
| 2025 | $12M |
| 2026 | $15M |

Variations

Common mistake: Running on a scanned PDF and getting empty text. If the PDF has no text layer, use OCR (--ocr-mode full_page).

Official docs → · PDF to JSON →

The full DoclingDocument structure for custom pipelines.

When to use: Use when you need the lossless DoclingDocument structure, including bounding boxes and layout metadata.

CLI

Python

import json
from docling.document_converter import DocumentConverter

converter = DocumentConverter()
result = converter.convert("report.pdf")
print(json.dumps(result.document.export_to_dict(), indent=2))

Flags used

• --to json Exports the lossless JSON representation.

Expected output

{
  "schema_name": "DoclingDocument",
  "texts": [ ... ],
  "tables": [ ... ]
}

Variations

Common mistake: Expecting the CLI JSON and export_to_dict() to be byte-identical; they are equivalent representations of the same document.

Official docs → · PDF to Markdown →

Convert a document straight from an HTTP URL.

When to use: Use when the document is hosted online and you do not want to download it first.

CLI

Python

from docling.document_converter import DocumentConverter

converter = DocumentConverter()
result = converter.convert("https://arxiv.org/pdf/2408.09869")
print(result.document.export_to_markdown())

Expected output

## Docling Technical Report

...

Variations

Common mistake: Assuming any URL works. The source must be a supported document format reachable over HTTP.

Official docs → · Supported formats →

Recover text from image-only pages.

When to use: Use when pages are images and contain no selectable text.

CLI

Python

from docling.document_converter import DocumentConverter, PdfFormatOption
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import PdfPipelineOptions

pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = True

converter = DocumentConverter(
    format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)}
)
result = converter.convert("scan.pdf")
print(result.document.export_to_markdown())

Flags used

• --ocr-mode full_page Runs OCR on every page, even those that already contain text.

Expected output

Text reconstructed from the scanned page image.

Variations

Common mistake: Leaving OCR on for digital PDFs wastes time. Only enable it when the text layer is missing.

Official docs → · Compare OCR engines →

Dramatically faster conversion for digital PDFs.

When to use: Use when the PDF already contains a text layer and you want the fastest conversion.

CLI

Python

from docling.document_converter import DocumentConverter, PdfFormatOption
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import PdfPipelineOptions

pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = False

converter = DocumentConverter(
    format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)}
)
result = converter.convert("report.pdf")
print(result.document.export_to_markdown())

Flags used

• --no-ocr Skips the OCR stage entirely.

Expected output

Markdown produced without running OCR.

Variations

Common mistake: Disabling OCR on a scanned PDF results in little or no text.

Official docs → · All commands →

Turn tables into Markdown or HTML matrices.

When to use: Use when the document contains tables you want as Markdown or HTML matrices.

CLI

Python

from docling.document_converter import DocumentConverter, PdfFormatOption
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import PdfPipelineOptions

pipeline_options = PdfPipelineOptions()
pipeline_options.do_table_structure = True

converter = DocumentConverter(
    format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)}
)
result = converter.convert("report.pdf")
print(result.document.export_to_markdown())

Flags used

• --table-mode accurate Uses TableFormer for complex, merged-cell tables.

Expected output

| Region | Q1 | Q2 |
|---|---:|---:|
| EMEA | 4.2 | 4.8 |

Variations

Common mistake: Assuming every table is perfect. Merged cells and borderless tables can still need review; try --table-mode accurate.

Official docs → · Build a custom command →

Capture equations and code blocks as LaTeX.

When to use: Use for scientific or technical documents containing equations or code blocks.

CLI

Python

from docling.document_converter import DocumentConverter, PdfFormatOption
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import PdfPipelineOptions

pipeline_options = PdfPipelineOptions()
pipeline_options.do_code_enrichment = True
pipeline_options.do_formula_enrichment = True

converter = DocumentConverter(
    format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)}
)
result = converter.convert("paper.pdf")
print(result.document.export_to_markdown())

Flags used

• --enrich-code Detects code blocks.
• --enrich-formula Extracts LaTeX formulas.

Expected output

$$ E = mc^2 $$

Variations

Common mistake: Enabling enrichment on documents without formulas or code adds processing time for little benefit.

Official docs → · Build a custom command →

Structure-aware chunks ready for a vector store.

When to use: Use when you want chunks that preserve headings, tables and page metadata for a vector store.

CLI

Python

from docling.document_converter import DocumentConverter
from docling.chunking import HybridChunker

converter = DocumentConverter()
result = converter.convert("report.pdf")

chunker = HybridChunker()
for chunk in chunker.chunk(result.document):
    print(chunk.text)

Flags used

• --to chunks Emits chunked output.
• --chunks-type hybrid Uses HybridChunker.

Expected output

Chunks split on document structure rather than raw character counts.

Variations

Common mistake: Using a naive character splitter instead of structure-aware chunking, which breaks tables and headings.

Official docs → · RAG guide →

Load parsed documents into a LangChain pipeline.

When to use: Use when loading parsed documents into a LangChain pipeline.

CLI

Python

from langchain_docling import DoclingLoader

loader = DoclingLoader(file_path="report.pdf")
docs = loader.load()
print(docs[0].page_content[:200])

Expected output

LangChain Document objects with parsed page content and metadata.

Variations

Common mistake: Forgetting to install the integration package separately from docling.

Official docs → · LlamaIndex recipe →

Build LlamaIndex nodes from parsed files.

When to use: Use when building LlamaIndex document nodes from parsed files.

CLI

Python

from llama_index.readers.docling import DoclingReader

reader = DoclingReader()
documents = reader.load_data(file_path="report.pdf")
print(documents[0].text[:200])

Expected output

LlamaIndex documents ready for indexing.

Variations

Common mistake: Mixing versions of docling and the reader package; keep both current.

Official docs → · LangChain recipe →