Embeddings

An embedding is a list of numbers that captures the meaning of a piece of content so that similar items sit near each other in vector space. Embeddings turn unstructured text, images, or code into a form that can be compared mathematically, which is what makes semantic search and retrieval-augmented generation possible.

Embeddings are usually stored in a vector database and queried at runtime to find the passages most relevant to a prompt. That retrieval step pulls external content into the model's context, so it is both a powerful grounding mechanism and a path for sensitive data or injected instructions to enter a request - which is why the retrieval boundary is a point worth governing.