app.py

import gradio as gr
import os
from http.cookies import SimpleCookie
from dotenv import load_dotenv
from llama_index.core import StorageContext, load_index_from_storage, VectorStoreIndex, SimpleDirectoryReader, ChatPromptTemplate, Settings
import random
import datetime

# Load environment variables
load_dotenv()

# Configure the Llama index settings
Settings.llm = HuggingFaceInferenceAPI(
    model_name="meta-llama/Meta-Llama-3-8B-Instruct",
    tokenizer_name="meta-llama/Meta-Llama-3-8B-Instruct",
    context_window=3000,
    token=os.getenv("HF_TOKEN"),
    max_new_tokens=512,
    generate_kwargs={"temperature": 0.1},
)
Settings.embed_model = HuggingFaceEmbedding(
    model_name="BAAI/bge-small-en-v1.5"
)

# Define the directory for persistent storage and data
PERSIST_DIR = "db"
PDF_DIRECTORY = 'data'

# Ensure directories exist
os.makedirs(PDF_DIRECTORY, exist_ok=True)
os.makedirs(PERSIST_DIR, exist_ok=True)

# Function to save chat history to cookies
def save_chat_history_to_cookies(chat_id, query, response, cookies):
    if cookies is None:
        cookies = {}
    history = cookies.get('chat_history', '[]')
    history_list = eval(history)
    history_list.append({
        "chat_id": chat_id,
        "query": query,
        "response": response,
        "timestamp": str(datetime.datetime.now())
    })
    cookies['chat_history'] = str(history_list)

def handle_query(query, cookies=None):
    chat_text_qa_msgs = [
        (
            "user",
            """
           You are the Lily Redfernstech chatbot. Your goal is to provide accurate, professional, and helpful answers to user queries based on the company's data. Always ensure your responses are clear and concise. Give response within 10-15 words only       
           {context_str}
            Question:
            {query_str}
            """
        )
    ]
    text_qa_template = ChatPromptTemplate.from_messages(chat_text_qa_msgs)

    # Load index from storage
    storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR)
    index = load_index_from_storage(storage_context)

    # Use chat history to enhance response
    context_str = ""
    if cookies:
        history = cookies.get('chat_history', '[]')
        history_list = eval(history)
        for entry in reversed(history_list):
            if entry["query"].strip():
                context_str += f"User asked: '{entry['query']}'\nBot answered: '{entry['response']}'\n"

    query_engine = index.as_query_engine(text_qa_template=text_qa_template, context_str=context_str)
    answer = query_engine.query(query)

    if hasattr(answer, 'response'):
        response = answer.response
    elif isinstance(answer, dict) and 'response' in answer:
        response = answer['response']
    else:
        response = "Sorry, I couldn't find an answer."

    # Update current chat history dictionary (use unique ID as key)
    chat_id = str(datetime.datetime.now().timestamp())
    save_chat_history_to_cookies(chat_id, query, response, cookies)

    return response

# Function to detect iframe and block chat history access
def detect_iframe():
    iframe_script = '''
        <script>
            if (window != window.top) {
                alert("Chat history access is disabled in iframes.");
                document.getElementById('chat_history').style.display = 'none';
            }
        </script>
    '''
    return iframe_script

# Define your Gradio chat interface function
def chat_interface(message, history):
    cookies = {}  # You might need to get cookies from the request in a real implementation
    try:
        # Process the user message and generate a response
        response = handle_query(message, cookies)

        # Return the bot response
        return response
    except Exception as e:
        return str(e)

# Custom CSS for styling
css = '''
  .circle-logo {
  display: inline-block;
  width: 40px;
  height: 40px;
  border-radius: 50%;
  overflow: hidden;
  margin-right: 10px;
  vertical-align: middle;
}
.circle-logo img {
  width: 100%;
  height: 100%;
  object-fit: cover;
}
.response-with-logo {
  display: flex;
  align-items: center;
  margin-bottom: 10px;
}
footer {
    display: none !important;
    background-color: #F8D7DA;
  }
label.svelte-1b6s6s {display: none}
div.svelte-rk35yg {display: none;}
div.svelte-1rjryqp{display: none;}
div.progress-text.svelte-z7cif2.meta-text {display: none;}
'''

# Use Gradio Blocks to wrap components and add iframe detection
with gr.Blocks() as demo:
    gr.HTML(detect_iframe())
    chat = gr.ChatInterface(chat_interface, css=css, clear_btn=None, undo_btn=None, retry_btn=None)

# Launch the Gradio interface
demo.launch()