LlamaIndex Framework Documentation

What LlamaIndex is

LlamaIndex describes its framework as a Python toolkit for building LLM-powered agents over your data. Where some frameworks start from the agent loop, LlamaIndex starts from the data: how to load it, index it, retrieve the right parts and give them to a model.

The main pieces

• Data connectors: loaders for files, APIs, databases and many SaaS tools.
• Indexes and vector stores: ways to organise data for search, with integrations for most vector databases.
• Query and chat engines: ready-made RAG patterns for question answering and multi-turn chat over your data.
• Agents and tools: models that use tools, including query engines, to complete tasks.
• Workflows: event-driven, multi-step processes with branching and human review.

Its sister product, LlamaParse, handles hard document parsing (scans, forms, spreadsheets, slides) and feeds clean Markdown into these indexes, but the framework works with any parser, including open-source ones.

When to choose it

LlamaIndex fits best when your project is mostly about your own documents: a RAG assistant over policies, a research tool over papers, a support bot over a help centre. It gives you many retrieval patterns to compare quickly. For a small feature, plain SDK code plus a vector store may still be simpler; for complex general agents, compare it with LangGraph or Haystack before committing.

Whatever you pick, keep a golden test set of real questions so you can tell whether a retrieval change actually helped.

Prerequisites and time

Comfortable Python and a basic understanding of embeddings, vector search and RAG. The core guides take a weekend; building and evaluating a real RAG app takes longer.

Free or paid?

The framework is free and open source under the MIT licence. LlamaParse and the hosted services are separate, with a free allowance and paid plans.

Common questions

LlamaIndex or LangChain? Both can build RAG and agents. LlamaIndex starts from data loading and retrieval; LangChain starts from models, tools and agents. Pick the one whose core matches the hardest part of your project, or neither for a small feature.

Best for: RAG and document-heavy projects where loading, indexing and retrieving data is most of the work.