LlamaIndex
The data framework for LLM applications.
LlamaIndex specializes in connecting LLMs to your private data. It provides loaders, indexes, retrievers, and query engines optimized for retrieval-augmented generation, plus a workflow system for agentic pipelines.
Install
pip install llama-indexnpm install llamaindexQuickstart
A minimal example to verify your setup.
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
# Load every file under ./data
documents = SimpleDirectoryReader("data").load_data()
# Build an in-memory vector index
index = VectorStoreIndex.from_documents(documents)
# Query it
query_engine = index.as_query_engine()
print(query_engine.query("Summarize the onboarding guide."))Core concepts
Data connectors
300+ loaders via LlamaHub — PDFs, Notion, Slack, S3, SQL, Confluence — normalized into a common Document representation.
Indexes
Vector, summary, knowledge-graph, and composable indexes. Choose the structure that matches your query patterns.
Query engines
Retrieval, post-processing, and response synthesis are pluggable stages. Add re-rankers, citations, and structured outputs.
Workflows & agents
Event-driven workflows compose tools, retrievers, and LLM calls into multi-step agents with first-class streaming.
Common use cases
- ›Enterprise knowledge assistants
- ›Document Q&A over private corpora
- ›Structured extraction from unstructured data
- ›Multi-modal retrieval (text, tables, images)