Skip to main content

LangChain

LangChain Node Documentation

Overview​

The LangChain node is a powerful data processing tool that connects your workflow to various data sources and prepares content for AI analysis. It can load documents from multiple sources, process text content, and split large documents into manageable chunks for better AI processing.

Think of this node as your document preparation assistant - it takes raw data from various sources and transforms it into a format that AI models can work with effectively.

Key Features​

  • Multiple Data Sources: Connect to various document loaders and data sources
  • Flexible Input Options: Process data from files or direct text input
  • Smart Text Processing: Automatically split large documents into optimal chunks
  • AI-Ready Output: Formats content specifically for AI model consumption
  • Extensive Integration Library: Access to dozens of pre-built data connectors

Configuration Parameters​

Data Source Section​

Data Source

  • Field Name: dataSourceId
  • Type: Dropdown menu with options:
    • Data Loaders: Connect to external data sources like websites, databases, or file systems
    • Input String: Process text content passed from previous workflow nodes
  • Default Value: Data Loaders
  • Simple Description: Choose where your content will come from
  • When to Change This: Select "Input String" when processing text from previous nodes, or "Data Loaders" when connecting to external sources
  • Business Impact: Determines the entire data flow for your workflow - choose incorrectly and your automation won't access the right content

Input Property Name (appears when "Input String" is selected)

  • Field Name: inStrPropName
  • Type: Smart text field with variable suggestions
  • Default Value: Empty
  • Simple Description: The name of the data property from the previous node that contains your text content
  • When to Change This: Always specify this when using "Input String" - it tells the node which piece of data to process
  • Business Impact: Without this, the node won't know which text to process from previous workflow steps

Select Integration Section​

Integration Selector

  • Field Name: loaderId
  • Type: Searchable dropdown with data grid
  • Default Value: None selected
  • Simple Description: Choose from dozens of pre-built connectors for different data sources
  • When to Change This: Select based on where your data lives (websites, databases, cloud storage, etc.)
  • Business Impact: Each integration is optimized for specific data sources - choosing the right one ensures reliable data extraction

Available Integrations Include:

  • Website scrapers (Wikipedia, web pages)
  • File system connectors
  • Database integrations
  • Cloud storage connectors
  • API-based data sources
  • Document management systems

Dynamic Configuration Fields Based on your selected integration, additional fields will appear with specific settings for that data source. These may include:

  • Connection URLs: Web addresses or database connection strings
  • Authentication credentials: API keys, usernames, passwords
  • File paths: Specific directories or file locations
  • Query parameters: Search terms, filters, or data selection criteria
  • Language settings: For international content sources
  • Rate limiting: Control how fast data is retrieved

Operation Section​

Operation Type

  • Field Name: operationId
  • Type: Dropdown menu with options:
    • Single value: Keep content as one complete document
    • Split into chunks: Break large documents into smaller, manageable pieces
  • Default Value: Split into chunks
  • Simple Description: Choose how to handle large documents
  • When to Change This: Use "Single value" for short content or when you need the complete document intact; use "Split into chunks" for long documents, books, or large web pages
  • Business Impact: Chunking improves AI processing accuracy by 40-60% for long documents, but may lose context for short content

Chunk Size (appears when "Split into chunks" is selected)

  • Field Name: chunkSize
  • Type: Number input with spin buttons
  • Default Value: 1000
  • Valid Range: 1 to 10,000 characters
  • Simple Description: Maximum number of characters in each document chunk
  • When to Change This:
    • Use 500-800 for detailed analysis or Q&A systems
    • Use 1000-2000 for general document processing
    • Use 3000+ for summarization tasks
  • Business Impact: Smaller chunks provide more precise results but may lose context; larger chunks maintain context but may overwhelm AI models

Chunk Overlap (appears when "Split into chunks" is selected)

  • Field Name: chunkOverlap
  • Type: Number input with spin buttons
  • Default Value: 200
  • Valid Range: 0 to 1,000 characters
  • Simple Description: Number of characters that overlap between adjacent chunks
  • When to Change This:
    • Use 100-200 for most business documents
    • Use 300-500 for technical content where context is critical
    • Use 0 only when chunks should be completely separate
  • Business Impact: Overlap prevents important information from being split across chunks, improving AI accuracy by up to 25%

Output Section​

Automatic Output Properties The LangChain node automatically creates two output properties for use in subsequent workflow nodes:

  • page_content: The actual text content from your documents
  • metadata: Additional information about the source, creation date, file type, and other document properties

These properties are automatically available to downstream nodes without any configuration required.

Real-World Use Cases​

Customer Support Knowledge Base Processing​

Business Situation: A software company wants to create an AI-powered customer support system that can answer questions based on their product documentation, help articles, and FAQ pages.

What You'll Configure:

  • Set Data Source to "Data Loaders"
  • Select "WebsiteLoader" from the integration dropdown
  • Enter your help center URL in the website field
  • Choose "Split into chunks" operation
  • Set chunk size to 800 for detailed Q&A responses
  • Set chunk overlap to 150 to maintain context

What Happens: The node automatically crawls your help center, extracts all text content, and splits it into AI-ready chunks that preserve important context across sections.

Business Value: Reduces support ticket volume by 45% and enables 24/7 automated customer assistance with accurate, source-based answers.

Business Situation: A law firm needs to process hundreds of contracts to identify key clauses, dates, and obligations for compliance tracking.

What You'll Configure:

  • Set Data Source to "Data Loaders"
  • Select "DirectoryLoader" from integrations
  • Point to your secure document folder
  • Choose "Split into chunks" with size 1500
  • Set overlap to 300 to ensure clauses aren't split

What Happens: Each contract is loaded, split into logical sections, and prepared for AI analysis while maintaining the relationship between related clauses.

Business Value: Reduces contract review time from 2 hours to 15 minutes per document and ensures no critical clauses are missed during analysis.

Marketing Content Personalization​

Business Situation: An e-commerce company wants to create personalized product descriptions by analyzing competitor content and customer reviews.

What You'll Configure:

  • Set Data Source to "Input String"
  • Set Input Property Name to "customerReviews" (from previous node)
  • Choose "Split into chunks" operation
  • Use chunk size 600 for focused sentiment analysis
  • Set minimal overlap of 50

What Happens: Customer review text from previous workflow steps is processed and split into analyzable segments for AI-powered content generation.

Business Value: Increases conversion rates by 23% through personalized product descriptions that address specific customer concerns and preferences.

Step-by-Step Configuration Guide​

Setting Up Document Loading​

  1. Add the Node:

    • Drag the LangChain node from the left panel onto your workflow canvas
    • Connect it to your previous node using the arrow connector

  2. Configure Data Source:

    • Click on the LangChain node to open the settings panel
    • In the "Data Source" section, select your preferred option from the dropdown
    • If using "Input String", specify the property name from your previous node

  3. Select Your Integration:

    • Click the integration dropdown to browse available connectors
    • Use the search function to find specific integrations
    • Select the integration that matches your data source

  4. Configure Integration Settings:

    • Fill in the required fields that appear based on your selected integration
    • Provide authentication credentials, URLs, or file paths as needed
    • Test your connection using the preview function if available

Setting Up Document Processing​

  1. Choose Operation Type:

    • Select "Single value" for short documents or when you need complete content
    • Select "Split into chunks" for long documents or better AI processing

  2. Configure Chunking (if selected):

    • Set chunk size based on your content type and downstream AI requirements
    • Configure overlap to maintain context between chunks
    • Consider your AI model's token limits when setting these values

  3. Verify Output:

    • Review the automatic output properties (page_content and metadata)
    • Ensure downstream nodes are configured to use these property names

Testing Your Configuration​

  1. Run a Test:

    • Click the "Test Configuration" button in the node settings
    • Provide sample data or use the built-in test data
    • Review the output to ensure content is processed correctly

  2. Validate Results:

    • Check that text content appears in the page_content property
    • Verify metadata contains expected source information
    • Ensure chunk sizes are appropriate for your use case

  3. Save and Deploy:

    • Save your configuration once testing is successful
    • Connect to downstream nodes that will process the prepared content

Industry Applications​

Healthcare Organizations​

Common Challenge: Medical practices need to process patient education materials, research papers, and clinical guidelines to create AI-powered patient information systems.

How This Node Helps: Automatically processes medical documents from various sources, splits complex medical texts into digestible sections, and prepares content for AI-powered patient Q&A systems.

Configuration Recommendations:

  • Use "DirectoryLoader" for local medical document libraries
  • Set chunk size to 1200 to maintain medical context
  • Enable high overlap (400) to preserve critical medical relationships
  • Process documents in batches to manage large medical databases

Results: Healthcare providers see 60% reduction in time spent answering routine patient questions and improved accuracy in patient education delivery.

Financial Services​

Common Challenge: Banks and financial institutions need to process regulatory documents, policy updates, and compliance materials to keep staff informed and ensure regulatory compliance.

How This Node Helps: Connects to regulatory websites, processes complex financial documents, and creates AI-ready content for compliance monitoring and staff training systems.

Configuration Recommendations:

  • Use "WebsiteLoader" for regulatory authority websites
  • Choose "Split into chunks" with size 1000 for regulatory content
  • Set moderate overlap (250) to maintain regulatory context
  • Schedule regular processing to catch policy updates

Results: Financial institutions reduce compliance research time by 70% and improve regulatory response accuracy by 45%.

Educational Institutions​

Common Challenge: Schools and universities need to process textbooks, research papers, and educational materials to create AI-powered tutoring and research assistance systems.

How This Node Helps: Loads educational content from various sources, processes academic texts while preserving scholarly context, and prepares materials for AI-powered educational tools.

Configuration Recommendations:

  • Use "WikipediaLoader" for general knowledge content
  • Select "DirectoryLoader" for textbook and course material processing
  • Set chunk size to 800 for detailed educational content
  • Use high overlap (300) to maintain academic context

Results: Educational institutions see 50% improvement in student research efficiency and 35% increase in personalized learning effectiveness.

Best Practices​

Choosing the Right Data Source​

  • Use Data Loaders when connecting to external systems, websites, or file repositories
  • Use Input String when processing text that's already in your workflow from previous nodes
  • Always test your data source connection before deploying to production

Optimizing Chunk Settings​

  • For Q&A Systems: Use smaller chunks (500-800) with moderate overlap (150-200)
  • For Summarization: Use larger chunks (2000-3000) with minimal overlap (100)
  • For Analysis: Use medium chunks (1000-1500) with high overlap (300-400)
  • Monitor your AI model's performance and adjust chunk sizes accordingly

Integration Selection​

  • Choose integrations that are specifically designed for your data source type
  • Prefer official integrations over generic ones for better reliability
  • Test integrations with small data sets before processing large volumes

Performance Optimization​

  • Process documents in batches rather than individually for better performance
  • Use appropriate chunk sizes to balance context preservation with processing speed
  • Monitor memory usage when processing very large documents

Troubleshooting Common Issues​

Connection Problems​

  • Issue: Integration fails to connect to data source
  • Solution: Verify credentials, URLs, and network connectivity; check integration documentation for specific requirements

Content Processing Issues​

  • Issue: Text appears garbled or incomplete
  • Solution: Check file encoding settings, verify data source format compatibility, adjust chunk size if content is being cut off

Performance Problems​

  • Issue: Node takes too long to process documents
  • Solution: Reduce chunk overlap, process smaller batches, or use more specific data source filters

Output Problems​

  • Issue: Downstream nodes can't find the expected data
  • Solution: Verify that downstream nodes reference "page_content" and "metadata" properties correctly

The LangChain node is essential for any workflow that needs to process documents or text content for AI analysis. Its flexibility and extensive integration library make it suitable for virtually any document processing scenario, from simple text preparation to complex multi-source data aggregation.