Upstage Document AI

Developer: omiya0555 Organization: Fusic Version: 0.0.1 Type: Tool Plugin Contact: omiya@fusic.co.jp Repository: https://github.com/omiya0555/dify-upstage-plugin.git

Description

Advanced Document AI plugin using Upstage API. Provides two powerful tools for document processing: parse documents into text/HTML/Markdown format, or extract structured data using custom JSON schemas.

Tools Included

1. Upstage Document Parser

Parse PDFs and images into text, HTML, or Markdown format with high accuracy OCR and layout detection.

Features:

• Multiple Output Formats: Markdown, HTML, Plain Text
• Advanced OCR: Automatic text recognition with chart detection

2. Upstage Information Extract

Extract structured data from documents using custom JSON schemas. Perfect for automating data entry from invoices, receipts, forms, and contracts.

Features:

• Custom Schema Definition: Define extraction fields in simple JSON format
• Structured JSON Output: Get clean, structured data ready for database integration
• Flexible Extraction: Works with any document type without templates or training

Common Features

• Intelligent Memory Cache: In-memory caching reduces API calls and improves performance
• Simple Configuration: Only requires Upstage API key

Setup

1. Get your API key from Upstage Console
2. Install the plugin in Dify
3. Configure the API key in plugin settings

Usage

Document Parser

1. Upload a document file
2. Select your preferred output format (Markdown, HTML, or Text)
3. The tool will parse the document and return the extracted content

Information Extract

1. Upload a document file
2. Define your extraction schema in JSON format:

{
  "invoice_number": "Invoice number from the document",
  "total_amount": "Total amount to be paid",
  "issue_date": "Date when invoice was issued",
  "company_name": "Name of the company"
}

3. The tool will extract the specified fields and return structured JSON data

Requirements

• Upstage API key
• Supported file formats: PDF, images
• Network connection to Upstage API

Performance Optimization

Memory Cache

The plugin implements an intelligent memory cache system that:

• Reduces API calls for identical documents
• Uses MD5 hashing for efficient cache key generation
• Automatically expires cached content after 1 hour
• Maintains a maximum of 100 cached items with LRU eviction

API Reference

This plugin uses two Upstage APIs:

Document Parse API

• Endpoint: https://api.upstage.ai/v1/document-digitization
• Model: document-parse
• OCR Mode: auto (optimized for each file type)
• Chart Recognition: Enabled
• Timeout: 300 seconds (5 minutes)

Information Extract API

• Endpoint: https://api.upstage.ai/v1/information-extraction
• Model: information-extract
• Input: Base64-encoded documents with JSON schema
• Output: Structured JSON matching your schema
• Timeout: 300 seconds (5 minutes)

For more details, visit Upstage API Documentation.