Kapa is a retrieval-augmented generation system, so it retrieves relevant information from your trusted sources before generating any response, rather than answering from a model's general training data. Every answer is grounded only in your connected sources and cites the material it used, so developers can verify it and click through to the original page. The retrieval pipeline is domain-specific and multi-step, running both embedding-based and sparse search with query decomposition and reranking to maximize recall and relevance. Kapa is also model-agnostic, continuously evaluating and selecting the best-performing models for technical question answering. This focus on grounded, citation-backed responses is the core difference between Kapa and a general-purpose assistant like ChatGPT.