We use LLM's to determine the users intent. For example: What do you want
to do? We no not send your data to the LLM. There are many reasons for
this, mainly:
Privacy. Sending all your data back to the LLM might be fine and you
might trust your LLM provider but there is a chance you could
accidentally leak sensitive information. Disclaimer: Kwigly does not
shield all your data from the LLM. For example: if you ask an assistant
to create a new task for you the name of the task will need to be sent
to the LLM in order to determine intent.
Consistency. You can ask for the same information 10 times and get 10
different responses. They might answer your question or complete the
task you wanted but it adds cognitive load. You have to read each
response to ensure the task was successful. Instead Kwigly assistants
provide consistent structured responses that stay the same.
Cost. By not sending data to the LLM and not expecting the LLM to
generate output we dramatically reduce costs. The cost of each message
with the assistant is limited to the input.
