LlamaIndex

LlamaIndex is an open-source (MIT) framework for building LLM apps and agents that answer from your own data: files, databases and APIs. Its core use case is retrieval-augmented generation, where relevant passages are fetched from your data and handed to the model at the moment it answers.

Data connectors load sources into Documents, which are split into Nodes and embedded into an index (in memory by default, or in a vector store you plug in). Query engines, chat engines and retrievers sit on top, and an agent can use an index as one tool among others. Workflows handle multi-step logic: each step receives an event and returns another, written in async Python, with branches and loops as plain code. The main package is Python (llama-index); LlamaIndex.TS covers Node.js, Deno, Bun and Cloudflare Workers.

The high-level API goes from a folder of files to a working query engine in a few lines, and every part can be swapped through the lower-level API. Integration packages cover many models, embedding models and vector stores, and a tutorial runs the whole thing on local models without any hosted API. The framework runs in your own process; the company's paid document-parsing API, LlamaParse (formerly LlamaCloud), is hosted in North America or the EU, or self-hosted on Enterprise plans.

The built-in readers handle clean text, but the docs point to LlamaParse for scanned PDFs, forms, spreadsheets and slides. The company now presents itself as the maker of LlamaParse and calls the framework its original toolkit.