CLI Usage¶

The Kreuzberg CLI provides a convenient command-line interface for text extraction from documents.

Installation¶

Install Kreuzberg with CLI support:

1

pip install kreuzberg[cli]

Or install all optional dependencies:

1

pip install kreuzberg[all]

Basic Usage¶

Extract from a file¶

1

kreuzberg extract document.pdf

Extract to a file¶

1

kreuzberg extract document.pdf -o output.txt

Extract from stdin¶

1
2
3

cat document.pdf | kreuzberg extract
# or
kreuzberg extract -

Command-Line Options¶

General Options¶

• -o, --output PATH: Output file path (default: stdout)
• --output-format [text|json|markdown|tsv|hocr]: Output format for extraction
• --show-metadata: Include metadata in output
• -v, --verbose: Verbose output for debugging

Processing Options¶

• --force-ocr: Force OCR processing
• --chunk-content: Enable content chunking
• --extract-tables: Enable table extraction
• --enable-table-detection: Enable table extraction from scanned documents (with TSV format)
• --max-chars INTEGER: Maximum characters per chunk (default: 2000)
• --max-overlap INTEGER: Maximum overlap between chunks (default: 100)

OCR Backend Options¶

• --ocr-backend [tesseract|easyocr|paddleocr|none]: OCR backend to use

Tesseract Options¶

• --tesseract-lang TEXT: Language(s) (e.g., 'eng+deu')
• --tesseract-psm INTEGER: PSM mode (0-13)
• --tesseract-output-format [text|markdown|tsv|hocr]: OCR output format (default: markdown)

EasyOCR Options¶

• --easyocr-languages TEXT: Language codes (comma-separated, e.g., 'en,de')

PaddleOCR Options¶

• --paddleocr-languages TEXT: Language codes (comma-separated, e.g., 'en,german')

Configuration File¶

Kreuzberg can load configuration from a pyproject.toml file:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14

[tool.kreuzberg]
force_ocr = false
chunk_content = true
extract_tables = false
max_chars = 5000
ocr_backend = "tesseract"

[tool.kreuzberg.tesseract]
language = "eng+deu"
psm = 3

[tool.kreuzberg.gmft]
verbosity = 1
cell_required_confidence = 50

Use a specific config file:

1

kreuzberg extract document.pdf --config custom-config.toml

OCR Output Format Examples¶

Extract tables from scanned documents¶

1
2

# Extract tables using TSV format with table detection
kreuzberg extract invoice.png --tesseract-output-format tsv --enable-table-detection

Fast text extraction¶

1
2

# Use text format for fastest extraction
kreuzberg extract document.pdf --tesseract-output-format text

Get structured markdown output¶

1
2
3
4
5

# Markdown is the default format
kreuzberg extract document.jpg

# Or explicitly specify markdown
kreuzberg extract document.jpg --tesseract-output-format markdown

Extract with position information¶

1
2

# Use hOCR format to get word positions
kreuzberg extract document.pdf --tesseract-output-format hocr -o output.html

Examples¶

Basic text extraction¶

1

kreuzberg extract report.pdf -o report.txt

OCR with specific language¶

1

kreuzberg extract scan.jpg --force-ocr --tesseract-lang deu

Extract tables to JSON¶

1

kreuzberg extract spreadsheet.pdf --extract-tables --output-format json -o tables.json

Extract with metadata¶

1

kreuzberg extract document.pdf --show-metadata --output-format json

Using EasyOCR backend¶

1

kreuzberg extract image.png --ocr-backend easyocr --easyocr-languages en,de

Extract with chunking¶

1

kreuzberg extract large-document.pdf --chunk-content --max-chars 1000

Module Execution¶

You can also run Kreuzberg as a Python module:

1

python -m kreuzberg extract document.pdf

Command Reference¶

kreuzberg extract¶

Extract text from a document.

Usage:

1

kreuzberg extract [OPTIONS] [FILE]

Arguments:

• FILE: Path to document or '-' for stdin (optional, defaults to stdin)

kreuzberg config¶

Show current configuration.

Usage:

1

kreuzberg config [OPTIONS]

Options:

• --config PATH: Configuration file path

kreuzberg --version¶

Show version information.

Error Handling¶

The CLI provides clear error messages:

• Exit code 0: Success
• Exit code 1: General error (parsing, validation, etc.)
• Exit code 2: Missing dependency error

Use --verbose for detailed error information and stack traces.