zerox-sync 0.1.0

Synchronous OCR using Gemini Vision API - A rewrite of pyzerox without async/litellm

Zerox Sync

A synchronous Python library for OCR and document extraction using Google's Gemini Vision API. This is a rewrite of pyzerox that removes async wrappers and replaces litellm with direct Gemini API integration.

Features

• Synchronous API: No async/await complexity, simple function calls
• Direct Gemini Integration: Uses Google's Gemini API directly without litellm dependency
• PDF to Markdown: Convert PDFs to structured markdown using vision models
• Concurrent Processing: Process multiple pages in parallel using ThreadPoolExecutor
• Selective Page Processing: Extract specific pages from PDFs
• Format Consistency: Maintain formatting across pages
• Simple Setup: Just set GOOGLE_API_KEY and go

Installation

Using uv (Recommended)

uv is a fast Python package installer:

# Install uv if you haven't already
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install zerox-sync
uv pip install zerox-sync

Using pip

pip install zerox-sync

System Dependencies

You'll need poppler installed for PDF processing:

macOS:

brew install poppler

Ubuntu/Debian:

sudo apt-get install poppler-utils

Windows: Download and install from poppler releases

Quick Start

from zerox_sync import zerox
import os

# Set your Gemini API key
os.environ["GOOGLE_API_KEY"] = "your-api-key-here"

# Process a PDF
result = zerox(
    file_path="path/to/document.pdf",
    model="gemini-3-pro",
)

# Access the results
for page in result.pages:
    print(f"Page {page.page}:")
    print(page.content)
    print(f"Length: {page.content_length} chars\n")

print(f"Total time: {result.completion_time}ms")
print(f"Input tokens: {result.input_tokens}")
print(f"Output tokens: {result.output_tokens}")

API Reference

zerox()

Main function to perform OCR on a PDF document.

def zerox(
    cleanup: bool = True,
    concurrency: int = 10,
    file_path: str = "",
    image_density: int = 300,
    image_height: tuple = (None, 1056),
    maintain_format: bool = False,
    model: str = "gemini-3-pro",
    output_dir: Optional[str] = None,
    temp_dir: Optional[str] = None,
    custom_system_prompt: Optional[str] = None,
    select_pages: Optional[Union[int, List[int]]] = None,
    **kwargs
) -> ZeroxOutput:

Parameters:

• cleanup (bool): Whether to cleanup temporary files after processing (default: True)
• concurrency (int): Number of concurrent threads for page processing (default: 10)
• file_path (str): Path or URL to the PDF file
• image_density (int): DPI for PDF to image conversion (default: 300)
• image_height (tuple): Image dimensions as (width, height) (default: (None, 1056))
• maintain_format (bool): Maintain consistent formatting across pages (default: False)
• model (str): Gemini model to use (default: "gemini-3-pro")
• output_dir (Optional[str]): Directory to save markdown output (default: None)
• temp_dir (Optional[str]): Directory for temporary files (default: system temp)
• custom_system_prompt (Optional[str]): Override default system prompt (default: None)
• select_pages (Optional[Union[int, List[int]]]): Specific pages to process (default: None)
• **kwargs: Additional arguments passed to Gemini API

Returns:

ZeroxOutput object with:

• completion_time (float): Processing time in milliseconds
• file_name (str): Processed file name
• input_tokens (int): Number of input tokens used
• output_tokens (int): Number of output tokens generated
• pages (List[Page]): List of Page objects containing:
  • content (str): Markdown content
  • page (int): Page number
  • content_length (int): Content length in characters

Advanced Usage

Process Specific Pages

result = zerox(
    file_path="document.pdf",
    select_pages=[1, 3, 5],  # Only process pages 1, 3, and 5
)

Maintain Format Consistency

result = zerox(
    file_path="document.pdf",
    maintain_format=True,  # Process pages sequentially to maintain formatting
)

Save to File

result = zerox(
    file_path="document.pdf",
    output_dir="./output",  # Markdown saved to ./output/{filename}.md
)

Custom System Prompt

result = zerox(
    file_path="document.pdf",
    custom_system_prompt="Extract only tables from this document in markdown format.",
)

Process from URL

result = zerox(
    file_path="https://example.com/document.pdf",
)

Adjust Concurrency

result = zerox(
    file_path="document.pdf",
    concurrency=5,  # Process 5 pages concurrently (default: 10)
)

Available Models

Zerox Sync supports various Gemini models:

• gemini-3-pro (default): Most intelligent model
• gemini-3-flash-preview: Fast with frontier-class performance
• gemini-2.5-pro: Powerful reasoning model
• gemini-2.5-flash: Balanced model with 1M token context
• gemini-2.5-flash-lite: Fastest and most cost-efficient

Environment Variables

• GOOGLE_API_KEY: Your Google AI Studio API key (required)
  • Get your key from: https://aistudio.google.com/apikey

Differences from pyzerox

1. Synchronous: No async/await - uses standard function calls
2. Gemini Direct: Direct Gemini API integration instead of litellm
3. Simple Dependencies: Fewer dependencies, no aiofiles/aiohttp/aioshutil
4. ThreadPoolExecutor: Uses standard library threading instead of asyncio
5. Requests: Uses requests library for HTTP instead of aiohttp

Error Handling

from zerox_sync import zerox
from zerox_sync.errors import (
    FileUnavailable,
    MissingEnvironmentVariables,
    ResourceUnreachableException,
    PageNumberOutOfBoundError,
)

try:
    result = zerox(file_path="document.pdf")
except MissingEnvironmentVariables:
    print("Please set GOOGLE_API_KEY environment variable")
except FileUnavailable:
    print("File not found or invalid path")
except ResourceUnreachableException:
    print("Could not download file from URL")
except PageNumberOutOfBoundError:
    print("Invalid page numbers specified")

Development

Setup

# Clone the repository
git clone https://github.com/yourusername/zerox-sync.git
cd zerox-sync

# Install uv if needed
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install with dev dependencies
uv pip install -e ".[dev]"

Running Tests

pytest

Code Formatting

# Format code
black zerox_sync tests

# Lint
ruff check zerox_sync tests

License

MIT License - see LICENSE file for details

Credits

This project is a synchronous rewrite of pyzerox by the getomni-ai team. The original project is an excellent async implementation with litellm support.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.