the Knowledge Base system implement a RAG approach, which means the AI will only retrieve data based on your queries, it can not read your context in full, you can see how the system works here: https://docs.typingmind.com/typingmind-team/branding-and-customizations/how-knowledge-base-works-in-typingmind-team

This approach allows you to upload large files beyond the model context limit

if you want the agent strictly follow all of your instructions, your guide should be put into the custom instructions within the agent or upload as training files, this way, the Agent can inject all of your documents into its initial context to always follow.

This approach is limited by the model context limit.