Knowledge base (RAG)

What it is

A voice agent answering "do you take my insurance?" or "what are your opening hours?" needs a source. The knowledge base is that source: web pages, PDFs, policy documents, FAQ lists you provide. Retrieval-augmented generation (RAG) is the technique that searches it at the moment of the question, selects the most relevant passages and hands them to the language model with an instruction to answer from those passages only.

The quality of the answer depends on the quality and structure of the documents, on how well the search matches a spoken question to written text, and on how firmly the model is held to the retrieved passage.

Why it matters when buying

The knowledge base is where most day-to-day maintenance happens after go-live, and it is where hallucinations hide. If retrieval finds nothing, a weak agent invents an answer. If two documents disagree, it picks one. Spoken questions are also messier than typed ones ("the thing where you pay monthly" versus "payment plans"), so retrieval tuned on text search can miss. Content freshness is a process question: who updates the price list, and how quickly does the agent reflect it?

What to ask

Ask what the agent says when the knowledge base has no answer; the right response is to say so and offer a message or transfer. Ask how you update content yourself and how long changes take to go live. Ask whether the agent can cite which document it used. In the demo, ask a question your documents do not cover and listen for an invented answer.