app.py

import gradio as gr
from openai import OpenAI
import time, re
import os
from dotenv import load_dotenv
import requests

load_dotenv()

OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
ASSISTANT_ID = os.getenv("ASSISTANT_ID")
BUBBLE_API_TOKEN = os.getenv("BUBBLE_API_TOKEN")
BUBBLE_API_URL = os.getenv("BUBBLE_API_URL")

client = OpenAI()

# Global variable for storing the thread ID
thread_id = None

def get_or_create_thread():
    """Creates a new thread if one doesn't exist."""
    global thread_id
    if thread_id is None:
        thread = client.beta.threads.create()
        thread_id = thread.id
    return thread_id

def stream_response(response_text):
    """
    Reformats the response to insert paragraph breaks and streams it in chunks.
    Inserts two newlines after punctuation and then yields the growing text.
    """
    formatted_response = re.sub(r'([.!?])\s+', r'\1\n\n', response_text)
    words = formatted_response.split(" ")
    current_text = ""
    for word in words:
        current_text += word + " "
        yield current_text.strip()
        time.sleep(0.05)  # Adjust delay for a natural streaming effect
    yield current_text.strip()

def extract_ingredients(response_text):
    """
    Extracts a list of ingredients from the response text.
    Looks for lines that start with a number (followed by a dot or parenthesis), dash, or bullet.
    """
    pattern = r'^\s*(?:\d+[\.\)]|-|\*)\s*(.+)$'
    ingredients = []
    for line in response_text.splitlines():
        match = re.match(pattern, line)
        if match:
            ingredient = match.group(1).strip()
            ingredients.append(ingredient)
    return ingredients

def send_to_bubble_database(ingredients, subject):
    """
    Sends the list of ingredients and the subject (user's message) to your Bubble application's database via an API endpoint.
    """
    bubble_api_url = BUBBLE_API_URL
    payload = {
        "ingredients": ingredients,
        "subject": subject,
        "api_token": BUBBLE_API_TOKEN
    }
    try:
        response = requests.post(bubble_api_url, json=payload)
        if response.status_code == 200:
            print("Ingredients and subject successfully sent to Bubble database.")
        else:
            print(f"Failed to send data to Bubble. Status: {response.status_code}, Response: {response.text}")
    except Exception as e:
        print("Error sending data to Bubble:", str(e))

def chat(message, history):
    """
    Sends the user's message, waits for the assistant's reply, streams the reformatted response,
    and then returns only the new conversation for this call (without appending previous history).
    """
    # Reset history to an empty list so that previous conversation is not included.
    history = []
    
    assistant_id = ASSISTANT_ID
    get_or_create_thread()

    # Get the current messages so we can detect new ones later.
    old_messages = client.beta.threads.messages.list(thread_id=thread_id).data
    old_ids = {msg.id for msg in old_messages}

    # Send the user's message.
    client.beta.threads.messages.create(
        thread_id=thread_id,
        role="user",
        content=message
    )

    # Start a run so the assistant processes the conversation.
    run = client.beta.threads.runs.create(
        thread_id=thread_id,
        assistant_id=assistant_id
    )

    # Wait for the run to complete.
    while True:
        run_status = client.beta.threads.runs.retrieve(thread_id=thread_id, run_id=run.id)
        if run_status.status == "completed":
            break
        time.sleep(1)

    # Fetch updated messages.
    new_messages = client.beta.threads.messages.list(thread_id=thread_id).data

    # Identify the new assistant message.
    assistant_response = ""
    for msg in new_messages:
        if msg.role == "assistant" and msg.id not in old_ids:
            assistant_response = msg.content[0].text.value
            break

    # Fallback: if no new message is found, use the latest assistant message.
    if not assistant_response:
        for msg in reversed(new_messages):
            if msg.role == "assistant":
                assistant_response = msg.content[0].text.value
                break

    # Stream the assistant's response.
    current_assistant_message = ""
    for chunk in stream_response(assistant_response):
        current_assistant_message = chunk
        yield [
            {"role": "user", "content": message},
            {"role": "assistant", "content": current_assistant_message}
        ]

    # Once streaming is complete, extract ingredients.
    ingredients = extract_ingredients(assistant_response)
    if ingredients:
        send_to_bubble_database(ingredients, subject=message)
        info_message = "[INFO] Ingredients list has been sent to your email."
    else:
        info_message = "[INFO] No ingredients list found in the assistant response."

    final_response = [
        {"role": "user", "content": message},
        {"role": "assistant", "content": current_assistant_message},
        {"role": "assistant", "content": info_message}
    ]
    yield final_response

# Optional: Custom CSS for styling the Gradio interface.
custom_css = """
.gradio-container {
    background-color: #186A27 !important;
}
#component-404 {
    background-color: white !important;
}
"""

# Create the Gradio Chat Interface.
chat_interface = gr.ChatInterface(
    fn=chat,
    title="AfrocourierAI",
    description="Let’s talk African food! From jollof rice to suya, I’ve got you covered—what’s on your mind?",
    type="messages",
    css=custom_css
)

chat_interface.launch(show_api=False)