Grounding is the practice of anchoring an agent's outputs in verifiable external sources — retrieved documents, database query results, API responses, or tool outputs — rather than relying entirely on the model's parametric memory. Grounded agents produce more factual, up-to-date, and trustworthy responses because each claim can be traced back to a source. Grounding is closely related to retrieval-augmented generation but extends beyond it to include any mechanism that ties agent output to external evidence.