Grounded knowledge, often built with retrieval-augmented generation (RAG), means an AI's answers are limited to a specific, approved set of content instead of whatever the model picked up in general training. The AI first pulls the relevant passages from your documentation, then writes its answer based only on those sources.
A general-purpose language model knows a little about everything and nothing specific about your equipment. Grounding helps address that. Before answering, the system pulls the right sections from your manuals, service bulletins, and workflows and tells the model to answer from those alone.
Grounding is the main defense against hallucinations. It also means the AI reflects your procedures, your part numbers, and your safety notes, not a generic version from the internet.
Administrators upload content, attach it to specific equipment models, and publish it to the knowledge base. The AI Assistant and Training agent are designed to answer only from that published content, and anything not yet published stays invisible to the AI.