Grounding means making sure a model's output is based on real, checkable information instead of only what it absorbed in training. A grounded agent answers from your documents, a database query or a fresh tool result, and can point to where the information came from.
The most common method is Retrieval-Augmented generation: fetch relevant passages from a knowledge base and put them in the prompt. Live tool calling is another way, such as checking an order status through an API instead of guessing it.
Grounding reduces LLM hallucination but does not remove it. The model can still misread a source, so good systems also show citations, tell the agent to say when the sources do not contain the answer, and test for that case.
Related: AI Agent, AI Agent Context, Semantic Search.