The assistant answers only from your content. When your content has no answer it does not guess: it says so, and hands the conversation to your team if you want it to.
A general-purpose model writes the most likely sentence it remembers from training. If it does not know your return window it will state the one common in your industry — it sounds right and it is wrong. Telling a customer something false costs more than telling them nothing.
SemAgent searches your knowledge base before it answers and looks at how relevant the passages it found actually are. Below a confidence threshold no answer is produced. This is not advice to the model, it is a gate in the code: at that point the model never gets to speak.
The assistant says it does not know, records the question, and — depending on your setting — connects the visitor to a person or asks for an email address. The recorded question lands in the unanswered questions list, where you feed your knowledge base so the same question is not missed twice.
Give your URL and your assistant starts answering from your knowledge base.
Start free