The RAG Glossary
One term per page. Precise definitions for a field full of overloaded words.
Read the blog →One term per page.
Retrieval-augmented generation is a field full of overloaded words. Document means four things depending on who’s speaking. Index is a data structure, a service, and a line on an invoice. Chunk has no agreed definition at all.
This is a reference work for those words. Each entry defines exactly one term, says what it is routinely confused with, and notes where usage in the wild is inconsistent.
How an entry is built
Every page follows the same shape:
- A one-sentence definition on the first line, complete on its own.
- What it actually is — a few paragraphs, plain before precise.
- In practice — how the thing shows up in a real system.
- Commonly confused with — the near neighbours, and the actual distinction.
- Usage notes — where the word is contested, loose, or vendor-specific.
No tutorials. No recommendations. No opinions about which tool to buy. A glossary describes; it doesn’t advise.
Terms whose meaning depends on the model, the library, or the vendor are marked as such rather than flattened into a single answer.
Latest posts
-
Embedding
A fixed-length list of numbers representing a piece of text, produced so that texts with similar meanings land near one another in the space.
-
Chunk
A unit of text stored and retrieved as a single item in a retrieval system — the smallest thing a search can return. No standard size exists.
-
Vector Index
A data structure that stores embeddings and finds the ones nearest to a query vector without comparing against every stored vector.
-
Reranking
A second scoring pass that reorders an already-retrieved candidate set using a slower, more accurate model that reads query and document together.
-
Context Window
The maximum amount of text, measured in tokens, that a language model can process in a single request — input and output together.