Give GPT (with own knowledge base) an instruction on how to behave before user prompt

Question

I have given GPT some information in CSV format to learn and now I would like to transmit an instruction on how to behave before the user prompt.

def chatbot(input_text):
    index = GPTSimpleVectorIndex.load_from_disk('index.json')
    message_history.append({"role": "user", "content": f"{input_text}"})
    response = index.query(input_text+message_history, response_mode="compact")
    return response.response

"message_history" looks like this:

message_history = [{"role": "user", "content": f"You are a Pirate! Answer every question in pirate-slang!"}, {"role": "assistant", "content": f"OK"}]

I got the following error:

"TypeError: can only concatenate str (not "list") to str"

I remember that I have to convert this into tuples but everything I try only causes more chaos...

Here's the whole code:

from gpt_index import SimpleDirectoryReader, GPTListIndex, GPTSimpleVectorIndex, LLMPredictor, PromptHelper
from langchain import OpenAI
import gradio as gr
import sys
import os

os.environ["OPENAI_API_KEY"] = 'INSERT_KEY_HERE'

message_history = [{"role": "user", "content": f"You are a Pirate! Answer every question in pirate-slang!"},
               {"role": "assistant", "content": f"OK"}]


def construct_index(directory_path):
    # Index is made of CSV, TXT and PDF Files
    max_input_size = 4096
    num_outputs = 512
    max_chunk_overlap = 20
    chunk_size_limit = 600

prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)

llm_predictor = LLMPredictor(llm=OpenAI(temperature=0.5, model_name="gpt-3.5-turbo", max_tokens=num_outputs))

documents = SimpleDirectoryReader(directory_path).load_data()

index = GPTSimpleVectorIndex.from_documents(documents)

index.save_to_disk('index.json')

return index

def chatbot(input_text):
    index = GPTSimpleVectorIndex.load_from_disk('index.json')
    message_history.append({"role": "user", "content": f"{input_text}"})
    response = index.query(input_text+message_history, response_mode="compact")
    return response.response

 iface = gr.Interface(fn=chatbot,
                 inputs=gr.inputs.Textbox(lines=7, label="Enter something here..."),
                 outputs="text",
                 title="ChatBot")

index = construct_index("docs")
iface.launch(share=True)