Building a knowledge-base AI assistant in Botpress

The goal of the exercise

Botpress is a visual development environment for building AI agents and chatbots. In this exercise you build an information assistant for the visitors of the fictional GreenDesk Hub coworking space in Budapest, answering questions about opening hours, prices, meeting rooms and house rules. The goal is not for the bot to answer everything: a good assistant works from the given knowledge base, does not invent missing information, recognises questions outside its remit, and resists instructions that try to override its rules. You first try the bot with no knowledge base (the V1 baseline), then build the Rich text knowledge source and the detailed agent instructions, and measure with 12 recorded tests how much more reliable it became (V2). GreenDesk Hub is entirely fictional: the addresses, prices and rules exist only for this exercise.

In this exercise you learn: how a Botpress agent, Workflow, Autonomous Node and Knowledge Base fit together; what RAG means, and why the model does not permanently learn the pasted text; how to write agent instructions that cover role, remit, handling of missing knowledge, operational limits, privacy and security; how to design a test set that measures invented facts and rule violations separately; how to use Inspect to diagnose whether the knowledge search actually ran; how to recognise and try out the risk of prompt injection; and when a chatbot must hand over to a human.

What will you need?

• A computer or laptop with a stable internet connection
• A modern desktop browser, preferably Google Chrome or Microsoft Edge