Author: abesticode
Version: 0.0.6
Type: tool Repo: https://github.com/abesticode/dify-plugin-knowledge-pro

Description

Knowledge Pro is a comprehensive AI Agent tool designed for managing Dify Knowledge Bases (Datasets) via the Dify API. It provides a complete set of operations for knowledge base maintenance, document management, content chunk manipulation, and metadata filtering.

Features

📁 Dataset Operations

📄 Document Operations

🧩 Chunk/Segment Operations

👶 Child Chunk Operations (Hierarchical Mode)

🏷️ Metadata Operations

Installation

1. Install the plugin from the Dify Marketplace or upload the .difypkg file
2. Navigate to your Dify workspace settings
3. Add the Knowledge Pro tool to your agent
4. Configure the required credentials:
   • API Key: Your Dify Knowledge Base API key
   • Base URL: Your Dify API endpoint (e.g., https://api.dify.ai/v1)

Getting API Credentials

1. Go to your Dify Knowledge Base
2. Navigate to the API ACCESS page from the left navigation
3. Generate a new API key from the API Keys section
4. Copy the API key and base URL for configuration

How to Use

Step 1: Authorization

The API_URL and API_KEY credentials come from the Dify Knowledge Base API Access page:

• API Key: Copy from the API Keys section
• Base URL: Your Dify API endpoint (e.g., https://api.dify.ai/v1)

Step 2: Get Dataset ID

The Dataset ID can be found in the Knowledge Base URL or settings:

Copy the Dataset ID to use in the tool parameters.

Step 3: Create Metadata Field

Before assigning metadata to documents, you need to create metadata fields first:

Use the Add Metadata Field tool to define custom fields like product_id, category, etc.

Step 4: Get List of Metadata IDs

To assign metadata values, you need the metadata field IDs:

Use the List Metadata tool to get the UUIDs of your metadata fields.

Step 5: Configure Create/Update Document

When creating or updating documents, configure the following parameters:

Basic Configuration:

Chunk Method & Process Rules:

Metadata Assignment:

Step 6: Example Output

Successful Document Creation:

Document with Metadata:

Usage Examples

Creating a Knowledge Base

Create a new knowledge base called "Product Documentation" with team access

Adding Documents with Metadata

Add a document named "Getting Started Guide" with the following content to the Product Documentation knowledge base. 
Set the product_id metadata to "PROD-001".

Searching Knowledge

Search for information about "installation requirements" in the Product Documentation knowledge base using semantic search

Managing Metadata

Add a metadata field called "category" to the Product Documentation knowledge base

Advanced Usage

Process Rules

The PROCESS_RULE parameter controls how documents are segmented and processed. The default is {"mode": "automatic"}, but you can customize it based on your chunk method.

Automatic Mode (default):

Custom Mode for General or Q&A Chunk Method:

When your CHUNK METHOD is General or Q&A, use this format:

{
  "mode": "custom",
  "rules": {
    "pre_processing_rules": [
      {"id": "remove_extra_spaces", "enabled": true},
      {"id": "remove_urls_emails", "enabled": true}
    ],
    "segmentation": {
      "separator": "\n\n",
      "max_tokens": 1024,
      "chunk_overlap": 50
    }
  }
}

Hierarchical Mode for Parent-child Chunk Method:

When your CHUNK METHOD is Parent-child, use this format:

{
  "mode": "hierarchical",
  "rules": {
    "pre_processing_rules": [
      {
        "id": "remove_extra_spaces",
        "enabled": true
      },
      {
        "id": "remove_urls_emails",
        "enabled": true
      }
    ],
    "segmentation": {
      "separator": "\n\n",
      "max_tokens": 1024,
      "chunk_overlap": 50
    },
    "parent_mode": "full-doc",
    "subchunk_segmentation": {
      "separator": "\n\n",
      "max_tokens": 1024,
      "chunk_overlap": 50
    }
  }
}

Process Rule Parameters:

Metadata Assignment

To assign metadata when creating/updating documents:

1. First, use List Metadata to get the metadata field IDs
2. Then provide the metadata as JSON in the format:

[
  {"id": "actual-field-uuid", "name": "field_name", "value": "field_value"}
]

Configuration

Error Handling

The plugin provides descriptive error messages for common issues:

• Invalid API key or authentication failures
• Missing required parameters
• Resource not found errors
• API rate limiting

Best Practices

1. Use List Tools First: Before updating or deleting resources, use the corresponding list tools to get the correct IDs
2. Monitor Indexing: After creating documents, use Get Indexing Status to track embedding progress
3. Organize with Metadata: Use metadata fields to categorize and filter documents effectively
4. Choose Search Methods Wisely:
   • Use semantic_search for meaning-based queries
   • Use keyword_search for exact matches
   • Use hybrid_search for best of both

Changelog

v0.0.1

• Initial release
• Full CRUD operations for Datasets, Documents, and Chunks
• Metadata management support
• Multiple search methods for retrieval
• Auto create/update document feature

v0.0.2

• Bug fixes and improvements
• Enhanced error handling

v0.0.3

• Added animated SVG icon with floating cubes and search effect
• Updated documentation

v0.0.4

• Added Get Chunk Details tool for retrieving complete chunk information
• Added Child Chunk Operations for hierarchical document structures:
  • List Child Chunks
  • Create Child Chunk
  • Update Child Chunk
  • Delete Child Chunk
• Updated API utility module with new endpoints
• Enhanced documentation

v0.0.5

• Added Keyword Search to List Documents tool for filtering documents by name
• Improved List Documents output with document names and IDs
• Added rag tag to plugin manifest

v0.0.6

• ✨ Embedding Cost Calculation Feature:
  • Added configurable Embedding Model selection in provider credentials
  • Added Custom Cost input for custom embedding models (cost per 1M tokens)
  • Supports preset models: OpenAI (ada-002, 3-small, 3-large), Cohere, Voyage
  • All cost calculations now use the configured model rate
• 📊 JSON Response Enhancement:
  • Added cost_info object to JSON response for easy data extraction
  • Fields include: tokens, cost_usd, embedding_model, cost_per_1m_tokens_usd
  • No more regex needed - directly select cost values from JSON!
• 🛠️ Tools Updated with Cost Info:
  • Create/Update Document - shows estimated tokens and cost
  • Add Chunks - shows actual/estimated tokens and cost
  • Get Indexing Status - shows actual tokens and cost after indexing
• 📁 New utility: CostCalculator for centralized cost calculations

Support

For issues and feature requests, please contact abesticode.

License

This plugin is provided as-is for use with the Dify platform.