What is a knowledge base for AI chat?

How grounded answers work

A general chat model knows a lot but nothing about your return policy. Retrieval-augmented generation (RAG) closes that gap: when a visitor asks a question, the system searches your knowledge base for the most relevant snippets, sends only those snippets and the question to the model, and asks it to answer from them. If nothing relevant is found, a good system says so and hands over to a person instead of guessing.

What makes a good knowledge base

• Short, single-topic entries. One question per FAQ entry retrieves better than one long page covering everything.
• The words visitors use. Write the question as a customer would ask it, not as the internal team names it.
• Current facts. Prices, delivery times and policies change; outdated entries produce confident wrong answers.
• Corrections fed back. When an agent fixes an AI answer, the fix should become an entry.

The knowledge base in TapCub

TapCub's AI assistant first matches a visitor's question against your FAQ entries; an exact FAQ hit is answered directly and does not use AI quota. Otherwise relevant document snippets are retrieved and the model drafts an answer grounded in them. If there is no knowledge match, the model cannot answer, the visitor asks for a person or a turn limit is reached, the chat is handed to a human. AI messages are labelled, and visitors are told they are talking to AI.

Agents can correct an AI answer with "Fix and save": the corrected answer becomes an FAQ entry and is used verbatim the next time. The support report shows the knowledge hit rate, the number of questions and handoffs. You can also use your own DeepSeek or OpenAI-compatible API key, so AI usage is billed to your account rather than the plan's AI limit.