LlamaIndex¶

What it is¶

LlamaIndex is a data framework for LLM applications to ingest, structure, and access private or domain-specific data. It provides tools for building RAG applications, including data connectors, indexes, and query engines.

What problem it solves¶

Simplifies the process of connecting LLMs to private and domain-specific data by providing purpose-built abstractions for data ingestion, indexing, and retrieval. It handles the "glue" code between data sources and LLM prompts.

Where it fits in the stack¶

Data Framework Layer. It sits between your data storage (files, databases, APIs) and your AI agents/applications, providing the context necessary for grounded responses.

Typical use cases¶

RAG Pipelines: Building question-answering systems over private document collections (PDFs, Notion, Slack).
Structured Extraction: Converting unstructured documents like invoices or receipts into Pydantic objects.
Data Agents: Creating agents that can autonomously decide which data source to query to answer a user request.
Knowledge Graphs: Building and querying property graphs for complex relational data.

Strengths¶

Data Centric: Purpose-built for data ingestion and retrieval, making RAG setup straightforward.
LlamaHub: Access to hundreds of data connectors (Google Drive, GitHub, Discord, etc.).
Evaluation Tools: Built-in tools for measuring retrieval quality and response faithfulness.
Advanced Retrieval: Supports complex patterns like sub-question querying and reranking.

Limitations¶

Rapid Evolution: The API changes frequently between versions (e.g., the v0.10.0 refactor was a major breaking change).
Abstractions Overhead: The deep nested abstractions can sometimes make it harder to customize low-level logic compared to building with raw LangChain.

When to use it¶

When the primary goal is building RAG over private or domain-specific data.
When you need a "data-first" approach to LLM application development.
For structured data extraction from complex documents.

When not to use it¶

When building simple chat applications that don't require external data.
For extremely complex multi-agent orchestration where LangGraph might offer more control.

Getting started¶

1. Installation¶

LlamaIndex is now modular. Install the core library and any necessary integrations:

pip install llama-index-core llama-index-readers-file llama-index-llms-openai

2. Basic RAG Example¶

Minimal example to query a directory of documents:

from llama_index.core import VectorStoreIndex, SimpleDirectoryReader

# Load documents from a 'data' directory
documents = SimpleDirectoryReader("./data").load_data()

# Create index and query engine
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()

response = query_engine.query("What is the main topic?")
print(response)

API examples¶

Property Graph Index¶

Property graphs allow for modeling complex relationships between entities extracted from text.

from llama_index.core import PropertyGraphIndex
from llama_index.core.indices.property_graph import SchemaLLMPathExtractor

# Define a schema for extraction
entities = ["Organization", "Person", "Event"]
relations = ["WORKS_AT", "ATTENDED", "FOUNDED"]
validation_schema = {
    "Organization": ["WORKS_AT", "FOUNDED"],
    "Person": ["WORKS_AT", "ATTENDED"],
}

kg_extractor = SchemaLLMPathExtractor(
    kg_schema=validation_schema,
    strict=True
)

index = PropertyGraphIndex.from_documents(
    documents,
    kg_extractors=[kg_extractor]
)

query_engine = index.as_query_engine(include_text=True)
response = query_engine.query("Who founded the organization?")

Customizing the LLM and Embeddings¶

LlamaIndex allows easy switching of backend providers (e.g., using OpenRouter or LocalAI).

from llama_index.core import Settings
from llama_index.llms.openai import OpenAI
from llama_index.embeddings.openai import OpenAIEmbedding

Settings.llm = OpenAI(
    model="gpt-4o",
    api_base="https://openrouter.ai/api/v1",
    api_key="your-key"
)
Settings.embed_model = OpenAIEmbedding(model="text-embedding-3-small")

Sources / references¶

Contribution Metadata¶

Last reviewed: 2026-06-01
Confidence: high