By name: docling-cli description: Convert documents to structured formats with OCR, table extraction, and AI-powered features. Use when working with PDF/DOCX/PPTX/images to Markdown/JSON/HTML conversion, table extraction, OCR on scanned PDFs, vision models for better extraction, or audio transcription.
Docling CLI
Convert documents to structured formats with intelligent parsing, OCR, and AI enhancement.
Installation
uv tool install docling[asr,vlm]
Directory Setup
Create working directories in your project root:
mkdir -p import export
import/- Source files to convertexport/- Converted output files
Note: You can specify any output directory with --output, but examples below use export/ for consistency.
Quick Start
Check complete options before using:
docling --help
Basic conversion (most parameters have sensible defaults):
# Convert PDF to Markdown
docling --output export/ import/document.pdf
# Convert to JSON
docling --to json --output export/ import/document.pdf
# Batch convert entire directory
docling --output export/ import/
Default behavior:
- Input format: auto-detects (PDF, DOCX, PPTX, images, etc.)
- Output format: Markdown
- Output directory: current directory
- OCR: enabled
- Table extraction: enabled
- Image export: embedded
Image Export Mode
Choose image export mode based on your needs:
| Mode | Description | When to use |
|---|---|---|
placeholder |
Only mark image positions | When user doesn't need images |
embedded |
Base64-encoded images (default) | When user needs images but wants single-file output |
referenced |
Export as PNG files, reference in document | When user needs separate image files |
docling --to json --image-export-mode referenced --output export/ import/document.pdf
Platform Optimization
macOS / Apple Silicon
Recommend using VLM pipeline for better performance, especially for scanned PDFs or documents with images:
docling --pipeline vlm --output export/ import/document.pdf
VLM pipeline with default model works well for documents with Chinese image content.
Audio Transcription
Convert audio files to text using ASR models:
# Use default whisper_tiny model
docling --pipeline asr --output export/ import/audio.mp3
# For better accuracy with Chinese
docling --pipeline asr --asr-model whisper_large --output export/ import/audio.mp3