LangChain vs LlamaIndex: Which to Use for Your Next AI App?

Published on: 2024-06-22

📝 Introduction

When building AI apps that integrate with external data or tools, two popular libraries emerge: LangChain and LlamaIndex. Both help connect LLMs to your data, but they serve slightly different goals.

This blog explains what each library does, includes setup and sample codes, and shows how they differ visually to help you choose wisely.

🔍 Overview Comparison

Feature	LangChain	LlamaIndex
Focus	Chains & agents orchestration	Indexing, retrieval, RAG pipelines
Design	Modular workflows with tools & agents	Data indexes + query engines
Use Cases	Chatbots, AI agents, tool calls	Knowledge retrieval, RAG apps
Learning Curve	Steeper, more flexible	Simpler for retrieval tasks
Integration	Models, tools, databases, APIs	Data + LLM retrieval integrations

🏗️ How They Work (Architecture Diagram)

✅ Explanation:

• LangChain orchestrates prompts, chains, and agents calling tools for complex workflows.
• LlamaIndex indexes your data and retrieves context chunks to enhance LLM responses.

💻 Setup Instructions

1. Create a new folder for testing:

mkdir ai-tools-test
cd ai-tools-test

2. Create and activate a virtual environment (uv preferred for speed, or use venv/pip if uv not available):

Using uv (recommended if installed):

uv venv
source .venv/bin/activate

Or using python venv:

python -m venv .venv
source .venv/bin/activate

3. Create a requirements.txt:

openai
langchain
llama-index

📝 Note: Adjust packages to match your local or cloud environment versions.

4. Install requirements:

Using uv:

uv pip install -r requirements.txt

Or using pip:

pip install -r requirements.txt

💻 Sample Code: LangChain

from langchain import OpenAI, LLMChain
from langchain.prompts import PromptTemplate

# Initialize LLM (set your OPENAI_API_KEY in env)
llm = OpenAI()

prompt = PromptTemplate(
    input_variables=['name'],
    template='Hello {name}, how can I assist you today?'
)

chain = LLMChain(llm=llm, prompt=prompt)

print(chain.run('Alice'))

✅ Explanation:

• Uses a prompt template with an LLM chain to generate output.

💻 Sample Code: LlamaIndex (Updated)

from llama_index.core import SimpleDirectoryReader, VectorStoreIndex
from llama_index.core.settings import Settings
from llama_index.llms.ollama import Ollama
from llama_index.embeddings.ollama import OllamaEmbedding

# ✅ Updated usage with working settings
Settings.llm = Ollama(model="llama3:latest", request_timeout=9999.0)
Settings.embed_model = OllamaEmbedding(model_name="mxbai-embed-large:latest", request_timeout=9999.0)

reader = SimpleDirectoryReader(input_dir="./data")
documents = reader.load_data()

index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()
response = query_engine.query("What is this dataset about?")
print(response)

✅ Explanation:

• Uses updated llama_index.core import paths.
• Configures Ollama as LLM with embedding model.
• Loads data from ./data, builds the index, and queries context.

🗂️ Folder Structure Example

ai-tools-test/
├── .venv/
├── requirements.txt
├── langchain_test.py
├── llamaindex_test.py
└── data/
    └── file1.txt
    └── file2.txt

• langchain_test.py: Contains LangChain example code
• llamaindex_test.py: Contains LlamaIndex example code
• data/sample.txt: Test data file for indexing

🔄 TL;DR

If your AI app needs:

• Complex workflows with agents and tool use → LangChain is ideal.
• Knowledge retrieval with RAG-style context integration → LlamaIndex shines.

Both can integrate with each other too – it’s not always a strict choice.