A language model only knows what it saw during training. It has never read your company policies, your product documentation, or last week's support tickets, and it will sometimes guess instead of admitting that. Retrieval-augmented generation, or RAG, fixes this by searching your own documents first and handing the most relevant passages to the model along with the question, so the answer is grounded in real sources.
To search by meaning rather than exact keywords, text is turned into embeddings, lists of numbers that place similar ideas close together, and stored in a vector database. Long documents are cut into smaller chunks before this step, and how you cut them has a large effect on whether the right passage is found. Retrieval quality sets the ceiling for the whole system.
Once these pieces make sense, you will see how they fit into a workflow: one flow loads and indexes new documents, and an agent looks things up in them to answer questions. Watch the lessons in order, then continue on to Building AI Agents.
IBM's Marina Danilevsky explains why language models give out-of-date or unsupported answers, and how adding a content store the model searches first makes answers current and traceable to a source.
IBM Technology on how text and other data become vector embeddings, why similar items sit close together, and how indexing makes similarity search fast enough for millions of records.
Greg Kamradt walks through ways to cut documents into chunks, from fixed sizes to recursive, semantic and agent-based splitting. It uses Python, so treat it as a look at the idea behind choosing a chunking strategy.
Cole Medin builds a RAG agent in n8n without code, with Supabase as the vector database and chat memory and Google Drive as the document source, including how to avoid duplicate knowledge when a document is updated.