LlamaIndex: AI agents connected to your own data

Built to connect AI to your documents

LlamaIndex launched in November 2022 under the name GPT Index, with a clear mission: letting a language model answer from your own data. That is the idea behind RAG, retrieval-augmented generation. You split documents into pieces, store them in an index (often a vector database), then retrieve the relevant passages when a question comes in.

The framework has since expanded into agents. The official docs now describe it as a Python toolkit for building agents that work over your data, built around two ideas:

• agents, which use tools to complete a task, such as querying several documents;
• workflows, event-driven multi-step processes with branching, retries and human review, close to an agentic workflow.

A change of direction worth knowing about

The official README is upfront about it: the company now puts most of its effort into LlamaParse, its commercial platform for document parsing and extraction (OCR, structured extraction, indexing, document agents). The framework remains open source and keeps evolving, but it is no longer the flagship product. The TypeScript version has been archived. For a long-term project, keep that context in mind.

Getting started

Install the llama-index package, which bundles the core framework with a selection of integrations, or start from llama-index-core and add only the integrations you need. Then follow the official starter tutorial, using either OpenAI or local models. A typical first project: load a folder of documents, build an index, then ask questions to an agent that relies on it.

Strengths and weaknesses

Strengths: deep expertise in document ingestion and indexing, many connectors for data sources, models and vector stores, MCP support through a dedicated integration, and a modular design.

Weaknesses: the company’s commercial focus has shifted to LlamaParse; the TypeScript ecosystem is no longer maintained; and for highly autonomous agents or agent teams, the framework is less central than tools built specifically for orchestration.

Alternatives

LangChain also covers RAG, with a very broad integration ecosystem. Haystack is another open source framework widely used for document search. Our agent frameworks comparison can help you decide.