app.py

import os, re
from flask import Flask, request, jsonify, make_response
from flask_cors import CORS
from time import sleep
from openai import OpenAI

# Set environment variables and OpenAI configurations
print("OpenAI:\t\t"+os.environ['OPENAI_API_KEY'])
# Connect to the assistant

openai_client = OpenAI(api_key=os.environ['OPENAI_API_KEY'])
openai_assistant = openai_client.beta.assistants.retrieve(assistant_id=os.environ['OPENAI_ASSISTANT_ID'])
openai_assistant_id=openai_assistant.id
openai_thread_id = ""
openai_additional_instruction = os.environ['OPENAI_ADDITIONAL_INSTRUCTION']
max_response_length = os.environ['MAX_RESPONSE_LENGTH']

app = Flask(__name__)
CORS(app)

# Function to create a thread

def create_thread():

    openai_thread = openai_client.beta.threads.create()
    return(openai_thread.id)

# Function to create a message in a given thread

def create_message(thread_id,user_message):
    thread_message = openai_client.beta.threads.messages.create(
    thread_id,
    role='user',
    content=user_message,
    )
    return thread_message

# Function to retrieve a message from a given thread

def retrieve_message(thread_id,message_id):
    message = openai_client.beta.threads.messages.retrieve(
    message_id=message_id,
    thread_id=thread_id,
    )
    return message

# Function to run the thread

def run_thread(thread_id,assistant_id):
    run = openai_client.beta.threads.runs.create_and_poll(
    thread_id=thread_id,
    additional_instructions=openai_additional_instruction,
    assistant_id=assistant_id,
    )

    return run

# Function to check the status of the run

def run_status(run):
    return run.status

# Function to clear a thread

def delete_thread(thread_id):
    return openai_client.beta.threads.delete(thread_id)

def delete_message(message_id,thread_id):
    deleted_message = openai_client.beta.threads.messages.delete(
    message_id=message_id,
    thread_id=thread_id,
    )
    return deleted_message.id

#This handles general Q&A to the LLM
def process_query(query,thread_id):

    retval = {"answer":""}
    query = "Answer the user's question in " + str(max_response_length) + " words or less: '" + query + "'"
    new_message = create_message(thread_id,query)
    run =run_thread(thread_id,openai_assistant_id)

    messages = openai_client.beta.threads.messages.list(thread_id=thread_id)
    for message in messages:
        
        if message.run_id == run.id:
            if message.role=='assistant':
                # gets the answer from the assistant
                answer = str(message.content[0].text.value)
                # kills the source reference in the response, if there
                regex_pattern = r"【.*?】"
                scrubbed_answer = re.sub(regex_pattern, '', answer)
                retval = {"answer":scrubbed_answer}

    return retval

#POST request to this service
@app.route('/query', methods=['POST'])
def handle_query():
    print(request)
    print(request.json)
    data = request.json
    query=data['prompt']
    openai_thread_id=data['thread']
    print("Assistant:\t\t "+openai_assistant.id)
    print("Thread:\t\t"+openai_thread_id)
    answer = ''

    # need to grab a thread id or create a new thread

    if openai_thread_id == "":
        print("Creating a new thread ")
        # create the thread
        openai_thread_id=create_thread()

        print("New thread:\t"+openai_thread_id)

    result = process_query(query,openai_thread_id)
    answer = result['answer']

    serialized_result = {
        "answer": answer,
        "matchedContext": "",
        "conversationPayload": "",
        "thread": openai_thread_id
    }
    print(serialized_result['answer'])
    return jsonify(serialized_result), 200

@app.route('/')
def hello():
    version = os.environ.get("CODE_VERSION")
    return jsonify({"status": "Healthy", "version": version}), 200

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=15002)