LangChain Components Guide

LangChain components are the building blocks of intelligent workflows. Each component has a specific purpose, and you combine them to create sophisticated AI applications without writing any code.

Think of components like specialized tools in a workshop - each one does something specific really well, and you choose the right combination for your project.

Core AI Components

Language Models (LLMs)

The “brain” that powers your AI workflows:

OpenAI Models

Best for: General intelligence, creative tasks, conversation

Available models:

• GPT-4: Most capable, best reasoning
• GPT-3.5 Turbo: Fast, cost-effective
• GPT-4 Turbo: Balance of capability and speed

Perfect for:

• Content creation and editing
• Complex analysis and reasoning
• Conversational interfaces
• Creative writing tasks

Anthropic Claude

Best for: Careful reasoning, safety-focused tasks

Available models:

• Claude 3 Opus: Most capable reasoning
• Claude 3 Sonnet: Balanced performance
• Claude 3 Haiku: Fast, efficient

Perfect for:

• Research and fact-checking
• Sensitive content analysis
• Detailed explanations
• Safety-critical applications

Local Models (Ollama)

Best for: Privacy, offline use, cost control

Popular models:

• Llama 2: General purpose, good performance
• Mistral: Efficient, multilingual
• Code Llama: Programming tasks

Perfect for:

• Personal data processing
• Offline workflows
• Cost-sensitive applications
• Privacy-required scenarios

Chains

Pre-built sequences for common AI tasks:

Chain Type	Purpose	When to Use
Basic LLM Chain	Simple AI text processing	Content analysis, summarization, generation
Q&A Chain	Answer questions about content	Document analysis, customer support
Summarization Chain	Create summaries of long content	Report generation, content curation
Sentiment Analysis	Determine emotional tone	Social media monitoring, feedback analysis

Agents

AI that can think, plan, and use tools:

graph TD
    Goal[User Goal] --> Think[Agent Thinks]
    Think --> Plan[Creates Plan]
    Plan --> Select[Selects Tools]
    Select --> Execute[Executes Actions]
    Execute --> Check[Checks Results]
    Check --> Success{Goal Achieved?}
    Success -->|No| Adapt[Adapts Plan]
    Success -->|Yes| Complete[Task Complete]
    Adapt --> Select
    
    style Think fill:#6d28d9,stroke:#fff,color:#fff
    style Plan fill:#6d28d9,stroke:#fff,color:#fff

Agent capabilities:

• Tool Selection: Automatically chooses the right tools for each task
• Planning: Breaks complex goals into manageable steps
• Adaptation: Adjusts approach when things don’t go as expected
• Reasoning: Explains its decisions and thought process

Memory Systems

Give your AI the ability to remember and learn:

Conversation Memory

Purpose: Remember chat history and context

How it works:

• Stores recent messages in conversation
• Provides context to AI for relevant responses
• Maintains conversation flow naturally

Best for:

• Chatbots and virtual assistants
• Customer support systems
• Interactive workflows

Example:

User: "What's the weather like?"
AI: "It's sunny and 75°F today."
User: "What about tomorrow?"
AI: "Tomorrow will be cloudy with a high of 68°F." (remembers location context)

Summary Memory

Purpose: Compress long conversations into key points

How it works:

• Summarizes older parts of conversation
• Keeps recent messages in full detail
• Balances context with memory efficiency

Best for:

• Long conversations
• Memory-constrained environments
• Cost optimization

Example:

Summary: "User is researching competitors in the SaaS space, particularly interested in pricing models."
Recent: Last 5 messages in full detail

Vector Memory

Purpose: Store and retrieve relevant memories by meaning

How it works:

• Converts memories into searchable vectors
• Finds relevant past interactions
• Provides context based on similarity

Best for:

• Knowledge building over time
• Complex, topic-based conversations
• Learning from past interactions

Example:

Current: "How do I handle customer refunds?"
Relevant memory: Previous discussion about refund policies from 2 weeks ago

Vector Stores & Embeddings

Smart document storage that understands meaning:

Vector Stores

Store Type	Best For	Key Features
Local Knowledge	Browser-based, private	Offline, no external dependencies
Pinecone	Production, scalable	Cloud-hosted, high performance
Supabase	Full-stack apps	Database + vector search
Qdrant	Self-hosted, flexible	Open source, customizable

Embeddings Models

Convert text into searchable vectors:

Model	Best For	Strengths
OpenAI Embeddings	General purpose	High quality, widely compatible
Ollama Embeddings	Local, private	Offline, no API costs
Cohere Embeddings	Multilingual	Strong non-English support

graph LR
    Text["'Customer is unhappy'"] --> Embed[Embedding Model]
    Embed --> Vector["[0.2, 0.8, 0.1, ...]"]
    Vector --> Search[Vector Search]
    Search --> Similar["'Client dissatisfied'<br/>'User frustrated'"]
    
    style Embed fill:#6d28d9,stroke:#fff,color:#fff
    style Similar fill:#e8f5e8

Tools & Utilities

Extend what your AI can do:

Web & Search Tools

• Wikipedia: Look up factual information
• SerpAPI: Perform web searches
• Calculator: Handle mathematical calculations
• Wolfram Alpha: Complex computations and data

Browser Integration Tools

• GetAllTextFromLink: Extract text content from web pages
• GetHTMLFromLink: Get structured HTML data
• GetImagesFromLink: Collect images from pages
• FormFiller: Automatically fill web forms

Data Processing Tools

• EditFields: Transform and format data
• Code Tool: Execute custom logic
• Filter: Remove unwanted data
• Merge: Combine multiple data sources

Document Processing

Handle different types of content:

Document Loaders

Loader Type	Handles	Use Cases
Default Document Loader	Plain text, web content	General content processing
GitHub Document Loader	Code repositories	Documentation analysis, code review
PDF Loader	PDF documents	Report analysis, research papers

Text Splitters

Break large documents into manageable chunks:

Splitter Type	How It Works	Best For
Character Text Splitter	Splits by character count	Simple, predictable chunks
Recursive Character Splitter	Respects structure (paragraphs, sentences)	Most content types
Token Splitter	Splits by AI token limits	Optimizing AI processing costs

Choosing the right components

Start with basic components (LLM + simple memory) and gradually add complexity. Each component adds capability but also complexity - only add what you need.

Component Combinations

Smart Content Analyzer

Components: GetAllTextFromLink + Basic LLM Chain + EditFields Purpose: Extract and analyze web content automatically

Intelligent Q&A System

Components: Document Loader + Embeddings + Vector Store + RAG Node Purpose: Answer questions about your documents accurately

Adaptive Research Agent

Components: Tools Agent + Web Search + Content Extraction + Memory Purpose: Conduct research that adapts to what it discovers

Conversational Assistant

Components: Chat Model + Conversation Memory + Multiple Tools Purpose: Helpful assistant that remembers context and can take actions

Performance Considerations

Model Selection

• Speed vs Quality: GPT-3.5 Turbo for speed, GPT-4 for quality
• Cost vs Capability: Local models for cost control, cloud models for capability
• Privacy vs Convenience: Local processing for privacy, cloud for convenience

Memory Management

• Short conversations: Simple conversation memory
• Long conversations: Summary memory to manage costs
• Knowledge building: Vector memory for learning over time

Vector Store Sizing

• Small datasets (< 1000 docs): Local Knowledge
• Medium datasets (1000-100k docs): Pinecone or Supabase
• Large datasets (100k+ docs): Specialized vector databases

Each component serves a specific purpose in building intelligent workflows. Understanding their strengths helps you choose the right combination for your specific needs.