LlamaIndex — Guide

LlamaIndex — User Guide

LlamaIndex; data + LLM.

Visit website

Freemium

Strengths

Focus on data connections and RAG, going deeper than LangChain in this area
Supports 100+ data sources (PDF, databases, APIs, web pages, etc.)
Advanced RAG technology: HyDE, sentence window, parent-child search, etc.
LlamaCloud provides managed data processing services
Can be used with LangChain

Best for

Enterprise knowledge base question and answer system
Unified retrieval of multiple data sources
Natural language query for structured data (database, Excel)
Document understanding and information extraction
Build an AI research assistant

Quickly build a RAG system

LlamaIndex provides a very simple API, and you can build a RAG system with just a few lines of code.

Scenario

The simplest documentation Q&A

Prompt example

from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.llms.openai import OpenAI
from llama_index.core import Settings

# Configure LLM
Settings.llm = OpenAI(model="gpt-4o", api_key="your-key")

# Load documents (supports PDF, Word, TXT, etc.)
documents = SimpleDirectoryReader("./docs").load_data()

# Build index (automatic chunking, vectorization)
index = VectorStoreIndex.from_documents(documents)

#Create query engine
query_engine = index.as_query_engine()

#Ask a question
response = query_engine.query("What is the company's core product?")
print(response)
print("\nSource:", response.source_nodes[0].text[:200])

Output / what to expect

LlamaIndex autocomplete:

Document loading and parsing
Text chunking (default 1024 tokens)
Vectorized storage
Retrieve the most relevant snippets
Generate answers with sources

Tips

SimpleDirectoryReader supports mixed file types, and PDF, Word, and TXT in one directory can all be loaded.

Scenario

Persistent index (avoid repeated construction)

Prompt example

import os
from llama_index.core import (
    VectorStoreIndex, SimpleDirectoryReader,
    StorageContext, load_index_from_storage
)

PERSIST_DIR = "./storage"

if not os.path.exists(PERSIST_DIR):
    # Build and save for the first time
    documents = SimpleDirectoryReader("./docs").load_data()
    index = VectorStoreIndex.from_documents(documents)
    index.storage_context.persist(persist_dir=PERSIST_DIR)
    print("Index construction completed and saved")
else:
    # Directly load existing indexes
    storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR)
    index = load_index_from_storage(storage_context)
    print("Load index from cache")

query_engine = index.as_query_engine()
response = query_engine.query("your question")

Output / what to expect

The first run will build and save the index, Subsequent runs will load directly. Avoid repeated calls to the Embedding API, Save time and money.

Tips

For large document libraries, persistent indexes are very important and can save a lot of API call costs.

Starter & above

The rest of this guide

Additional scenarios and the full comparison table are included with Starter and above. Sign in with an eligible account to load them.

View plans