Skip to content

Latest commit

 

History

History
98 lines (75 loc) · 2.67 KB

README.md

File metadata and controls

98 lines (75 loc) · 2.67 KB

LangChain RAG Tutorial (Updated for v1.2.4 - January 2026)

A comprehensive, production-ready RAG (Retrieval Augmented Generation) tutorial using the latest LangChain patterns.

✨ What's New in This Update

Feature Description
🤖 Agent-Based RAG Uses create_agent instead of chains - LLM decides when to retrieve
🏠 Local Model Support Ollama integration for free, local inference (no API key needed)
📄 Advanced PDF Processing Tables (PDFPlumber) + Diagrams (GPT-4o Vision)
💬 Conversation Memory Multi-turn chat with InMemorySaver
🌊 Streaming Responses Real-time token streaming for better UX
🛠️ Multi-Tool Agents Easily extend with custom tools

🚀 Quick Start

Option A: With OpenAI API

pip install langchain langchain-openai langchain-chroma langgraph
export OPENAI_API_KEY="your-key-here"

Option B: Fully Local (No API Key)

pip install langchain langchain-ollama langchain-chroma langgraph
# Install Ollama: https://ollama.com/download
ollama pull llama3.2
ollama pull nomic-embed-text

Then set USE_OLLAMA = True in the notebook.

📦 Requirements

pip install -qU \
    langchain>=1.2.4 \
    langchain-openai \
    langchain-ollama \
    langchain-chroma \
    langchain-community \
    langgraph>=1.0.0 \
    chromadb \
    pypdf \
    pdfplumber \
    pymupdf \
    pillow

🔑 Key API Patterns (v1.2.4)

# Agent Creation (NEW - replaces create_retrieval_chain)
from langchain.agents import create_agent
from langchain_core.tools import create_retriever_tool
from langgraph.checkpoint.memory import InMemorySaver

agent = create_agent(
    model=llm,                    # ChatOpenAI or ChatOllama instance
    tools=[retriever_tool],
    system_prompt="You are...",   # NOT state_modifier!
    checkpointer=InMemorySaver()  # For conversation memory
)

result = agent.invoke({"messages": [HumanMessage(content="question")]})
answer = result["messages"][-1].content

📊 Why Agents > Chains?

Chain-Based RAG Agent-Based RAG
Always queries database LLM decides when to retrieve
Wastes tokens on "hi", "thanks" Answers simple prompts directly
Fixed pipeline Flexible tool selection
Hard to extend Easy to add more tools

📁 Files

  • langchain_rag_tutorial_updated.ipynb - Main tutorial notebook
  • CHANGELOG.md - Detailed list of all changes made

🧪 Tested With

  • Python 3.12
  • LangChain 1.2.4
  • LangGraph 1.0.6
  • OpenAI GPT-4o-mini / GPT-4o
  • Ollama llama3.2 / llava

📝 License

MIT

Updated January 2026 for LangChain v1.2.4