Knowledge bases are just as they sound, a way to build context around a subject, whether that's a client, project, product, industry or anything else that it would be beneficial to have a shared understanding of across the team.
Context can be added as written text or (as of October 2023) you can upload PDFs to build up your knowledge base.
Knowledge bases within Hoist are supported by a "vector database". This means that when a user adds context to a Knowledge Base, Hoist extracts all of the data and splits it into "vector chunks".
Storing the knowledge bases in this way means that when a user later interacts with an App that has access to that Knowledge Base, the most relevant chunks of data are retrieved and sent to the LLM to help it answer the user's question. By default, Hoist sends 5 chunks of data to the LLM, but on the App settings the user can modify this between 1 and 20 chunks More chunks can improve responses but consume more tokens, increasing costs. Less chunks may lead to less accurate responses but consume fewer tokens (so are cheaper)
If your knowledge base has recently been saved, it might take a few minutes before you see the effects flow through to your apps.