aicapture 0.2.1

A Python module to capture knowledge from documents using Vision Language Models (VLMs)

AI Vision Capture

A powerful Python library for extracting and analyzing content from PDF, Image, and Video files using Vision Language Models (VLMs). This library provides a flexible and efficient way to process documents with support for multiple VLM providers including OpenAI, Anthropic Claude, Google Gemini, and Azure OpenAI.

Features

• 🔍 Multi-Provider Support: Compatible with major VLM providers (OpenAI, Claude, Gemini, Azure, OpenSource models)
• 📄 Document Processing: Process PDFs and images (JPG, PNG, TIFF, WebP, BMP)
• 🎥 Video Processing: Extract and analyze frames from video files (MP4, AVI, MOV, MKV)
• 🚀 Async Processing: Asynchronous processing with configurable concurrency
• 💾 Two-Layer Caching: Local file system and cloud caching for improved performance
• 🔄 Batch Processing: Process multiple documents in parallel
• 📝 Text Extraction: Enhanced accuracy through combined OCR and VLM processing
• 🎨 Image Quality Control: Configurable image quality settings
• 📊 Structured Output: Well-organized JSON and Markdown output

Installation

pip install aicapture

Environment Setup

1. Set your chosen provider and API key:

# For OpenAI
export USE_VISION=openai
export OPENAI_API_KEY=your_openai_key

# For Anthropic
export USE_VISION=anthropic
export ANTHROPIC_API_KEY=your_anthropic_key

# For Gemini
export USE_VISION=gemini
export GEMINI_API_KEY=your_google_key

2. Optional performance settings:

export MAX_CONCURRENT_TASKS=5      # Number of concurrent processing tasks
export VISION_PARSER_DPI=333      # Image DPI for PDF processing

Core Capabilities

1. Document Parsing

The VisionParser provides general document processing capabilities for extracting unstructured content from documents.

from aicapture import VisionParser

# Initialize parser
parser = VisionParser()

# Process a single PDF
result = parser.process_pdf("path/to/your/document.pdf")

# Process a single image
result = parser.process_image("path/to/your/image.jpg")

# Process multiple documents asynchronously
async def process_folder():
    return await parser.process_folder_async("path/to/folder")

Parser Output Format

{
  "file_object": {
    "file_name": "example.pdf",
    "file_hash": "sha256_hash",
    "total_pages": 10,
    "total_words": 5000,
    "pages": [
      {
        "page_number": 1,
        "page_content": "extracted content",
        "page_hash": "sha256_hash"
      }
    ]
  }
}

2. Structured Data Capture

The VisionCapture component enables extraction of structured data from images using customizable templates.

1. Define your data template:

# Example template for technical alarm logic
ALARM_TEMPLATE = """
alarm:
  description: string  # Main alarm description
  destination: string # Destination system
  tag: string        # Alarm tag
  ref_logica: integer # Logic reference number

dependencies:
  type: array
  items:
    - signal_name: string  # Name of the dependency signal
      source: string      # Source system/component
      tag: string        # Signal tag
      ref_logica: integer|null  # Logic reference (can be null)
"""

2. Use with OpenAI Vision:

from aicapture import VisionCapture
from aicapture import OpenAIVisionModel

vision_model = OpenAIVisionModel(
    model="gpt-4o",
    max_tokens=4096,
    api_key="your_openai_key"
)

capture = VisionCapture(vision_model=vision_model)
result = await capture.capture(
    file_path="path/to/image.png",
    template=ALARM_TEMPLATE
)

3. Or use with Anthropic Claude:

from aicapture import AnthropicVisionModel

vision_model = AnthropicVisionModel(
    model="claude-3-sonnet-20240620",
    max_tokens=4096,
    api_key="your_anthropic_key"
)

capture = VisionCapture(vision_model=vision_model)
result = await capture.capture(
    file_path="path/to/example.pdf",
    template=ALARM_TEMPLATE
)

3. Video Processing

The VidCapture component enables extraction of knowledge from video files by extracting frames and analyzing them with VLMs.

from aicapture import VidCapture, VideoConfig

# Configure video capture with custom settings
config = VideoConfig(
    frame_rate=2,                         # Extract 2 frames per second
    max_duration_seconds=30,              # Process up to 30 seconds of video
    target_frame_size=(768, 768),         # Resize frames for optimal processing
    supported_formats=(".mp4", ".avi", ".mov", ".mkv")
)

# Initialize video capture
video_capture = VidCapture(config)

# Process a video file with a custom prompt
result = video_capture.process_video(
    video_path="path/to/your/video.mp4",
    prompt="Describe what is happening in this video."
)

# Or extract frames for custom processing
frames, interval = video_capture.extract_frames("path/to/your/video.mp4")
print(f"Extracted {len(frames)} frames at {interval:.2f}s intervals")

# Analyze the extracted frames with a custom prompt
result = video_capture.capture(
    prompt="Analyze these video frames and describe key objects and actions.",
    images=frames
)

Advanced Usage

Custom Vision Model Configuration

from aicapture import VisionParser, GeminiVisionModel

# Configure Gemini vision model with custom settings
vision_model = GeminiVisionModel(
    model="gemini-2.0-flash",
    api_key="your_gemini_api_key"
)

# Initialize parser with custom configuration
parser = VisionParser(
    vision_model=vision_model,
    dpi=400,
    prompt="""
    Please analyze this technical document and extract:
    1. Equipment specifications and model numbers
    2. Operating parameters and limits
    3. Maintenance requirements
    4. Safety protocols
    5. Quality control metrics
    """
)

# Process PDF with custom settings
result = parser.process_pdf(
    pdf_path="path/to/document.pdf",
)

Development Setup

For local development:

1. Clone the repository
2. Copy .env.template to .env
3. Edit .env with your settings
4. Install development dependencies: pip install -e ".[dev]"

See .env.template for all available configuration options.

Documentation

For detailed configuration options and examples, see:

• Configuration Guide
• Advanced Usage Examples

Coming Soon

• 🔗 Cross-Document Knowledge Capture: Capture structured knowledge across multiple documents

Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/tiny-but-mighty)
3. Commit your changes (git commit -m 'feat: add small but delightful improvement')
4. Push to the branch (git push origin feature/tiny-but-mighty)
5. Open a Pull Request

For detailed guidelines, see our Contributing Guide.

License

Copyright 2024 Aitomatic, Inc.

Licensed under the Apache License, Version 2.0. See LICENSE for details.