Knowledge bases and RAG: answers grounded in your documents

Retrieval-augmented generation (RAG) gives a model your documents at answer time. The question is embedded, the most relevant chunks of your content are retrieved, and they are placed in the prompt so the answer is grounded in your data. It is how an agent answers from a product manual, a policy or an internal wiki without retraining anything.

Good RAG is mostly good plumbing: sensible chunks, a retrieval strategy that catches exact terms as well as meaning, a token budget so the prompt stays bounded, and a record of what was retrieved so a wrong answer can be traced to a wrong chunk.

Create a knowledge base under Knowledge, add documents, then create an index. The index chunks the documents (chunk size from 200 to 8,000 characters, with overlap) and embeds each chunk with the embedding provider you choose. Re-index after editing documents.

Test retrieval before attaching anything: the retrieval test runs a query against the knowledge base and shows the matching chunks with scores, so you can tune chunk size and thresholds with real questions. Then attach the knowledge base in the agent's Knowledge tab. The default workflow already contains a knowledge retrieval node.

The knowledge retrieval node embeds the query (by default the latest user message), runs keyword and vector search, fuses the results with reciprocal rank fusion, optionally reranks them, drops results under the minimum score, and keeps the top k within the per-chunk and total token caps.

Selected chunks enter the prompt as relevant knowledge with their scores and, when enabled, citations. Every retrieval is logged with the candidates, the selected chunks, what was trimmed and why, and whether a source chunk has changed since the run.