chunktopus 0.1.2

A multithreaded OCR API for document processing and chunking with simple interface

Chunktopus

A powerful document processing and chunking API by Unsiloed AI, featuring multithreaded OCR and intelligent document chunking using OpenAI's Vision models.

Installation

pip install chunktopus

Features

• Process PDF, DOCX, and PPTX documents
• Multiple chunking strategies:
  • Fixed size chunks
  • Semantic chunks (using OpenAI)
  • Page-based chunks
  • Paragraph-based chunks
  • Heading-based chunks
• Fast parallel processing
• RESTful API with FastAPI
• Simple JavaScript-like API

Configuration

OpenAI API Key

Important: This package requires your own OpenAI API key. You must provide your API key in one of the following ways:

1. Set it as an environment variable:

export OPENAI_API_KEY="your-api-key"

2. Create a .env file in your project directory:

OPENAI_API_KEY=your-api-key

3. Set it programmatically:

import os
os.environ["OPENAI_API_KEY"] = "your-api-key"

4. Provide it directly in the API call (recommended for the simple API)

Usage

Using the Simple API (Recommended)

import chunktopus

# Process a document with your own OpenAI API key
result = chunktopus.process_sync({
    "filePath": "https://example.com/document.pdf",
    "credentials": {
        "apiKey": "your-openai-api-key"
    },
    # Optional parameters
    "strategy": "semantic",  # or "fixed", "page", "paragraph", "heading"
    "chunkSize": 1000,
    "overlap": 100
})

# Print the results
print(f"Number of chunks: {result['total_chunks']}")
for i, chunk in enumerate(result['chunks']):
    print(f"Chunk {i+1}: {chunk['text'][:100]}...")

# Async version is also available
import asyncio

async def main():
    result = await chunktopus.process({
        "filePath": "path/to/document.pdf",
        "credentials": {
            "apiKey": "your-openai-api-key"
        }
    })
    print(f"Number of chunks: {result['total_chunks']}")

asyncio.run(main())

Using as a Python Package (Legacy)

from chunktopus import chunktopus

# Process a document with your own OpenAI API key
result = chunktopus(
    file_path="path/to/document.pdf",
    credentials={"api_key": "your-openai-api-key"}
)

# Or process a document from a URL
result = chunktopus(
    file_path="https://example.com/document.pdf",
    credentials={"api_key": "your-openai-api-key"}
)

# Or process raw text
result = chunktopus(
    text="Your text content to chunk semantically...",
    credentials={"api_key": "your-openai-api-key"}
)

# Print the results
for i, chunk in enumerate(result):
    print(f"Chunk {i+1}: {chunk['text'][:100]}...")

Using from Command Line

After installation, you can use the chunktopus command:

# Process a file with your OpenAI API key
chunktopus path/to/document.pdf --api-key "your-openai-api-key"

# Output to a file
chunktopus path/to/document.pdf --api-key "your-openai-api-key" --output results.json

# Format as text instead of JSON
chunktopus path/to/document.pdf --api-key "your-openai-api-key" --format text

Running as a REST API

from chunktopus.server import run_server

# Start the server on port 8000
run_server(port=8000)

Using the document processing functions directly

from chunktopus.services import process_document_chunking
from chunktopus.utils.chunking import ChunkingStrategy

# Set your OpenAI API key
import os
os.environ["OPENAI_API_KEY"] = "your-api-key"

# Process a document with semantic chunking
result = process_document_chunking(
    file_path="path/to/document.pdf",
    file_type="pdf",
    strategy=ChunkingStrategy.SEMANTIC,
)

# Print the results
print(f"Number of chunks: {result['total_chunks']}")
for i, chunk in enumerate(result['chunks']):
    print(f"Chunk {i+1}: {chunk['text'][:100]}...")

API Endpoints

/chunking

POST endpoint that processes a document and returns chunks.

Parameters:

• document_file: The document file (PDF, DOCX, PPTX)
• strategy: Chunking strategy (semantic, fixed, page, paragraph, heading)
• chunk_size: Size of chunks for fixed strategy (default: 1000)
• overlap: Overlap size for fixed strategy (default: 100)

License

MIT License