What it is

LlamaIndex connects model applications to private or domain data through ingestion, indexing, retrieval, query workflows, and agents.

Data path

• Documents and nodes represent source items and retrievable units.
• An index organizes nodes; a vector index is one option.
• A retriever selects nodes for a query.
• A query engine coordinates retrieval and response synthesis.

Keep stable document identifiers and source metadata through every stage. Without them, citations, deletion, access control, and incremental re-indexing become fragile.

Retrieval checklist

1. Start with a lexical or flat-vector baseline.
2. Tune chunking to document structure.
3. Filter by tenant and permission before retrieval.
4. Evaluate retrieved evidence, not only final-answer fluency.
5. Version the index, embeddings, parser, and prompt.

Further reading: LlamaIndex high-level concepts.