Depending on the content uploaded to CHATTERgo and how the users ask their questions, you may encounter situations that the system doesn't trigger knowledge base search, or retrieve too much or too little context for the user questions.
In this case, you may adjust the Knowledge Base setting under Plugins.
You can update the Description to instruct CHATTERgo when to trigger knowledge base search, and also provide a config json to adjust the maximum number of chunk of content and the relevancy threshold of the content retrieved to pass as context to the large language model.
For the config, you can input as below:
Due to the token limitation of the LLM, we suggest match_count to be 2-5. And to provide reasonable result, we suggest to tune the similarity_threshold between 0.6-0.9.