gpt4_shell

Now using nltk to estimate token counts. We are using a message history class and counting tokens of the contents of the history, we trim it out as need to ensure we are under 4000 tokens, and then send all of that in the query. This is expensive but also puts our historical context in so we can have actual conversations with GPT instead of each prompt being a blank slate

Little python script to clean up logs and pycache

added in message history so it's conversational now, the history is passed in for each new query. This will have practical implications as we run up agains the token limit. Eventually we will need estimate our total token count for our message history, trim it down, and then send. If we get an token error back from the API then we need to trim some more. I think this whole process will be much easier if we have a function that easily converts inputs to token estimates