Knowledge base

A knowledge base is the collection of business-approved information that an AI agent consults when it answers questions. Opening hours, address, services, return and cancellation rules and frequently asked questions are typical content. To a large extent, it decides what the agent knows and what it does not. A good knowledge base is concise, current and consistent with itself.

How it works

Content is added as text, documents or question-and-answer pairs and split into searchable chunks. When a question comes in, the system first finds the relevant chunks in the knowledge base (retrieval), and the language model then builds its answer on those chunks. That way the model uses the business's own information instead of guessing from general knowledge.

Why it matters

A knowledge base is the most basic safeguard against hallucination: when the agent relies on an approved source, it is far less likely to make things up. But if the source is out of date, the agent repeats outdated information, and if two documents contradict each other, its answers become inconsistent. Every piece of content therefore needs an owner and a regular review schedule. Temporary information, such as holiday opening hours or a campaign that has ended, is the easiest to forget.