Grounding

Without grounding, LLM-based applications suffer from two problems: knowledge ages (the model knows nothing new after its training cutoff) and the model invents plausible but wrong content (hallucinations). Grounding solves both by giving the model runtime access to current, verified data.

Technically, grounding is usually implemented through Retrieval Augmented Generation (RAG). On a request, relevant data is pulled from a knowledge base and supplied to the model as context. The model then answers based on that data, not just from its training memory. The quality of the knowledge base determines the quality of the answers.

Pimcore's Data Spine is the natural grounding layer for AI applications with product context. Structured product data, master data, assets, and content are consolidated in a governed layer and available as a factual basis for AI agents. The Agent SDK orchestrates retrieval and generation, so AI answers always build on current, verified data, not on model memory.