Grounding
Give your AI agents current information they can cite, from your own enterprise data, the web, or both.
What is grounding?
Grounding connects an AI model to trusted, current sources of information so its answers rest on evidence it can cite, instead of relying only on training data. A grounded agent retrieves evidence first, reasons over it, then answers with its sources attached.
Two approaches cover most production work. Enterprise data grounding draws on your own documents, wikis, and databases. Web Search draws on current public information from the web.
Benefits
What changes when agents work from sources instead of training data.
Grounded responses carry the sources they drew on, so a reviewer can check a claim instead of taking it on trust.
The model reads its sources at query time, so new documents and recent developments are picked up as your sources change, with no fine-tuning cycle.
Some questions need both: how a change in the world affects something only your own records know. An agent can draw on enterprise sources and the web in the same turn.
Enterprise grounding retrieves only from sources you own, inside your own boundary. Web grounding is served from an Amazon-operated index with zero data egress by default, so your data never leaves your AWS environment.
Choose your grounding approach
Two managed paths, depending on where your answers need to come from.
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages