LlamaIndex: RAG Pipelines and Data Agents

After this lesson, you will be able to: Index documents in a vector store, build a RAG pipeline, and create a data agent that queries multiple sources and synthesises answers.

LlamaIndex specialises in data-heavy agents, long documents, structured tables, knowledge graphs. Where LangChain is generalist, LlamaIndex is the RAG specialist. This lesson indexes documents, queries them, and builds a multi-source data agent.

Prerequisites:LangGraph: Stateful Multi-Step Agent Graphs

Install + setup

1
1. `pip install llama-index llama-index-llms-anthropic`
2
2. Set ANTHROPIC_API_KEY.
3
3. Make a folder `./data` and drop in 2–3 PDFs or text files.

Index and query documents

5 lines from PDFs to RAG.

python

from llama_index.core import SimpleDirectoryReader, VectorStoreIndex, Settings
from llama_index.llms.anthropic import Anthropic

Settings.llm = Anthropic(model='claude-sonnet-4-6')
documents = SimpleDirectoryReader('./data').load_data()
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()

print(query_engine.query('What is the main argument across these documents?'))

Multi-document data agent

Two indexes (e.g., HR docs + product docs), one agent routes between them.

python

from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.core.tools import QueryEngineTool, ToolMetadata
from llama_index.core.agent import ReActAgent
from llama_index.llms.anthropic import Anthropic

llm = Anthropic(model='claude-sonnet-4-6')

hr_index = VectorStoreIndex.from_documents(SimpleDirectoryReader('./hr_docs').load_data())
prod_index = VectorStoreIndex.from_documents(SimpleDirectoryReader('./product_docs').load_data())

tools = [
    QueryEngineTool(query_engine=hr_index.as_query_engine(),
        metadata=ToolMetadata(name='hr_docs', description='HR policies, benefits, time off.')),
    QueryEngineTool(query_engine=prod_index.as_query_engine(),
        metadata=ToolMetadata(name='product_docs', description='Product specs, features, roadmap.')),
]
agent = ReActAgent.from_tools(tools, llm=llm, verbose=True)

print(agent.chat('How many vacation days do I get and which features ship next quarter?'))

Persistence

By default, the index lives in memory. For production: use Chroma, Qdrant, Weaviate, or Pinecone storage. `index.storage_context.persist('./store')` saves to disk; `load_index_from_storage('./store')` reloads. Saves you from re-embedding every run.

←LangGraph: Stateful Multi-Step Agent Graphs

Back to Code-First Agents

CrewAI: Multi-Agent Teams with Roles and Tasks→