upstage-document-parser

Upstage Document Parsing Plugin for Dify

A powerful document parsing plugin for the Dify platform that leverages the Upstage Document Parse API to convert various document formats into structured Markdown, HTML, or plain text.

Features

Broad format support: Handles PDF, DOCX, and various image formats
Intelligent document understanding: Extracts text, tables, charts, and figures while preserving original structure
Multiple output formats: Converts documents to Markdown, HTML, or plain text
Efficient caching: Content-based caching prevents reprocessing of identical files
OCR capability: Extracts text from scanned documents and images
Chart recognition: Identifies and extracts charts from documents
Batch processing: Efficiently processes multi-page documents
Coordinate extraction: Obtains bounding box coordinates for document elements

Installation

This plugin is under active development.

Configure the plugin in the Dify platform.

Configuration

Required credentials

The plugin requires the following credentials:

: Upstage API key (obtain it from the Upstage Console)
: Base URL for your Dify instance (default: "https://cloud.dify.ai")

Parameter options

You can configure the following parameters when using the tool:

: Output format (options: "md", "html", "text")
: Whether to return the result as a file or text (options: "file", "text")

Usage

In a Dify application

Add the Upstage Document Parse tool to your application.
Configure the required credentials.
Use the tool within your application flow to process documents.

Direct usage in Python

You can also use the client directly in Python:

API Parameters

The plugin uses the following parameters when calling the Upstage Document Parse API:

Type	Description	Default
File	The document file to process	Required
String	Control OCR behavior: "auto" (applies only to images) or "force" (convert everything to images first)	"auto"
Boolean	Whether to return bounding box coordinates	false
Boolean	Whether to use chart recognition	true
List[String]	Formats for layout elements: "text", "html", "markdown"	["html", "markdown", "text"]
String	Model used for inference	"document-parse-250618"
List[String]	Layout categories to provide as base64-encoded strings	["table", "figure", "chart"]

Caching Mechanism

The plugin implements an efficient caching system:

File content hashing to identify duplicate documents
Result caching based on content hash and output format
TTL-based cache expiration (default: 1 hour)

Examples

Convert a PDF to Markdown

Handling large documents

Development

Project structure

: Main Dify plugin integration
: Core client that interacts with the Upstage API
: Python dependencies