LlamaParse Dify Plugin

LlamaParse is a GenAI-native document parser that can parse complex document data for any downstream LLM use case (RAG, agents). This plugin integrates LlamaParse capabilities into the Dify platform.

Features

✅ Broad file type support: Parse various unstructured file types (.pdf, .pptx, .docx, .xlsx, .html) with text, tables, visual elements, and complex layouts.

✅ Table recognition: Accurately parse embeddLed tables into text and semi-structured representations.

✅ Multimodal parsing: Extract visual elements (images/diagrams) into structured formats using the latest multimodal models.

✅ Custom parsing: Customize output through custom prompt instructions.

✅ Advanced LLM Mode: Enhanced parsing with LLM-powered layout reconstruction, target pages, max pages, and custom prompts.

Tools Included

This plugin provides two tools:

1. Llama Parse (Standard)

The standard LlamaParse tool with basic parsing capabilities.

2. Llama Parse Advanced (LLM Mode)

An advanced version with enhanced features:

• LLM Mode: Uses for better layout reconstruction
• Target Pages: Parse specific pages using comma-separated page numbers (e.g., "0,2,7")
• Max Pages: Limit the maximum number of pages to parse
• System Prompt: Override default LlamaParse system prompt (use with caution)
• User Prompt: Transform or modify parsed content (e.g., translation, summarization)

Getting Started

API Key Setup

1. Visit https://cloud.llamaindex.ai/api-key to create an account
2. Generate your API key
3. Configure the API key in your Dify plugin settings

Configuration Setup

For file upload functionality, configure the environment variable in your Dify file:

• Docker Compose:
• Other deployments:

Usage Limits

• Free plan: Up to 1000 pages per day
• Paid plan: 7000 free pages per week + $0.003 per additional page

Parameters

Standard Tool Parameters

Parameter	Type	Required	Default	Description
files	files	Yes	-	Files to be parsed
result_type	select	No	markdown	Output format (txt or md)
num_workers	number	No	4	Number of parallel workers for processing multiple files
verbose	boolean	No	false	Enable detailed output logging
language	string	No	"en"	Output language (e.g., "en" for English)

Advanced Tool Parameters

The advanced tool includes all standard parameters plus:

Parameter	Type	Required	Default	Description
target_pages	string	No	-	Specific pages to parse (e.g., "0,2,7")
max_pages	number	No	-	Maximum number of pages to parse
system_prompt	string	No	-	Custom system prompt (use with caution)
user_prompt	string	No	-	Custom user prompt for content transformation

Usage Examples

Standard Tool

Advanced Tool

Output Format

The plugin provides three types of output for each processed file:

1. Text Message

   • Plain text concatenation of all parsed documents, separated by "---"

2. JSON Message

   • Structure:
   • Contains parsed text and associated metadata for each document

3. Blob Message

   • Binary content with appropriate MIME type:
     • JSON: "application/json"
     • Markdown: "text/markdown"
     • Text: "text/plain"

Credits

This plugin is powered by LlamaParse, a powerful document parsing service by LlamaIndex.