LLMWhisperer CLI Test Script

A simple command-line client for LLMWhisperer, a powerful document extraction service from Unstract that converts complex documents (PDFs, images, scanned files) into LLM-ready text.

Features

• Extract text from PDFs, images, and scanned documents
• Multiple extraction modes for different document types
• Table structure preservation with optional border recreation
• Page-specific extraction
• Save output to file or display in terminal
• Environment-based API key configuration

Installation

Prerequisites

• Python 3.7 or higher
• pip package manager

Setup

1. Clone this repository:

git clone https://github.com/Zipstack/llmwhisperer-cli-test-script.git
cd llmwhisperer-cli-test-script

2. Create and activate a virtual environment:

# Create virtual environment
python -m venv venv

# Activate virtual environment
# On Linux/Mac:
source venv/bin/activate
# On Windows:
# venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Configure your API key:

Create a .env file in the project directory:

LLMWHISPERER_API_KEY=your_api_key_here

Alternatively, set it as an environment variable:

export LLMWHISPERER_API_KEY=your_api_key_here

Get your API key from Unstract LLMWhisperer

Usage

Basic Usage

Extract text from a document:

python llmwhisperer_cli.py document.pdf

Save Output to File

python llmwhisperer_cli.py document.pdf -o extracted_text.txt

Extraction Modes

LLMWhisperer supports different extraction modes optimized for various document types:

• native_text: For digitally created PDFs with embedded text (fastest)
• low_cost: For clean, printed documents with good scan quality
• high_quality (default): For challenging documents including handwritten text
• form: For documents with forms, checkboxes, and structured layouts
• table: For documents with dense table structures

Example:

python llmwhisperer_cli.py document.pdf -m table

Page Selection

Extract specific pages or page ranges:

# Extract pages 1-5
python llmwhisperer_cli.py document.pdf -p "1-5"

# Extract pages 1-5 and page 7
python llmwhisperer_cli.py document.pdf -p "1-5,7"

# Extract from page 21 to end
python llmwhisperer_cli.py document.pdf -p "21-"

Table Border Recreation

For better table structure preservation:

# Add vertical borders
python llmwhisperer_cli.py document.pdf --vert

# Add both vertical and horizontal borders
python llmwhisperer_cli.py document.pdf --vert --horiz

Note: --horiz requires --vert to be enabled

Complete Example

Extract tables from specific pages with borders and save to file:

python llmwhisperer_cli.py financial_report.pdf \
  -m table \
  -p "10-15" \
  --vert --horiz \
  -o tables_output.txt

Command-Line Options

Option	Description	Default
file_path	Path to the document to process	Required
-o, --output	Output file to save extracted text	None (prints to console)
-m, --mode	Extraction mode (see modes above)	high_quality
-p, --pages	Pages to extract (e.g., "1-5,7,21-")	All pages
--vert	Recreate vertical table borders	False
--horiz	Recreate horizontal table borders	False
-h, --help	Show help message	-

Output

The client provides:

• Extracted text (to console or file)
• Total number of pages processed
• Processing status and progress indicators

API Configuration

The client uses the following environment variables:

• LLMWHISPERER_API_KEY: Your API key (required)
• LLMWHISPERER_BASE_URL_V2: API endpoint (optional, defaults to US region)

For EU region, set:

LLMWHISPERER_BASE_URL_V2=https://llmwhisperer-api.eu-west.unstract.com/api/v2

Examples

Extract text from a scanned document

python llmwhisperer_cli.py scanned_document.pdf -m low_cost

Process a form with checkboxes

python llmwhisperer_cli.py application_form.pdf -m form -o form_data.txt

Extract tables with structure preservation

python llmwhisperer_cli.py data_tables.pdf -m table --vert --horiz

Extract specific pages from a large document

python llmwhisperer_cli.py manual.pdf -p "1-10,50-55" -o summary.txt

Troubleshooting

1. "LLMWHISPERER_API_KEY not found": Ensure your .env file is in the same directory as the script or set the environment variable.

2. "--horiz requires --vert": Horizontal borders can only be added when vertical borders are enabled.

3. Timeout errors: For large documents, the default timeout is 200 seconds. The script will wait for processing to complete.

License

This project is provided under the MIT license.

Additional Resources

• LLMWhisperer Documentation
• API Reference
• Python Client Documentation