LlamaIndex started as a RAG-focused library, built specifically to solve the problem of getting an LLM to answer questions using data it wasn't trained on: documents, databases, and internal tools. Through LlamaHub, it offers more than 150 connectors to sources like Notion, Slack, and SQL databases, handling the ingestion, chunking, and indexing work that RAG pipelines need before a single query ever reaches a model.
Workflows, its event-driven orchestration layer, extended the project well beyond pure retrieval into general agent building, letting developers define multi-step processes with explicit events rather than a fixed chain. That shift mirrors what happened across the whole ecosystem: tools that started as RAG libraries kept absorbing agent-framework features as the line between answering questions over your data and taking actions on your behalf blurred.
Newer products built on the same foundation, like LlamaParse for document understanding, show the company pushing beyond the open-source framework into a broader toolkit for building production RAG and agent systems, while keeping the core library free and open source.