LlamaIndex

LlamaIndex is a Python data framework for building LLM and agent applications over your own data. It ingests private data sources, structures them so a model can use them, and answers queries with retrieved context, which makes it a common base for retrieval-augmented generation (RAG).

The framework has four layers of tooling. Data connectors ingest APIs, PDFs, documents, SQL databases, and other formats. Indices and graphs structure that data for model use. A retrieval and query interface takes a prompt and returns retrieved context with a knowledge-augmented answer. Integrations connect it to an outer application framework such as LangChain or Flask. A high-level API ingests and queries data in a few lines, and lower-level APIs let you customize or replace any module, including connectors, indices, retrievers, query engines, and rerankers.

The library is split into a core package and more than 300 integration packages for LLM, embedding, and vector store providers, and imports show which is which (llama_index.core for core, other namespaces for integrations). A starter package bundles core with a selection of integrations. Examples include building a vector store index with OpenAI, or with Llama models served through Ollama and Hugging Face embeddings. Indices stay in memory by default and can be persisted to disk and reloaded.

LlamaIndex the company now concentrates on LlamaParse, a separate document platform with Parse (agentic OCR across 130+ formats), Extract (structured extraction), Index (ingest and RAG), Split, and LlamaAgents for deployed document agents. LlamaParse works with the framework or on its own, and the open-source framework remains available.