A vector-backed knowledge base can ground a chatbot in useful context, but collecting and maintaining the data for it can be quite a chore. I built Alida Vector after getting tired of manually updating our chatbot’s knowledge base.

Alida Vector collects content from sources such as websites and Jira Service Management. Before an updated index goes live, it can use an LLM to validate the changes. This allows it to run on a schedule and keep the vector database current with little human intervention.

It has been running smoothly in production for a while. Since plenty of other people are integrating LLMs into products, I decided to share it. I hope it can serve as inspiration for your own implementation or be useful to you directly.