app.py

import re
import os
os.system(f"pip install open-interpreter")
os.system(f"pip install matplotlib")
import json
import gradio as gr
from interpreter import interpreter
import time
import matplotlib
matplotlib.use('Agg') 


#get openaiapikey from environment variable
openaiapikey = os.environ.get('OPENAI_API_KEY')

#install open-interpreter package
os.system(f"pip install open-interpreter")

interpreter.auto_run = True
interpreter.llm.model = "gpt-4-turbo"
interpreter.custom_instructions = "First ask the user what they want to do. Based on the input, describe the next steps. If user agrees, proceed; if not, ask what they want next.If it is anything to display , always at the end open up the file."

def json_to_markdown(json_data):
    full_message = ""
    images = []
    for item in json_data:
        if item['role'] == 'assistant':
            if item['type'] == 'message':
                content = item.get('content', " ")
                full_message += content + " "
            elif item['type'] == 'image' and 'path' in item:
                images.append(item['path'])
    return full_message.strip(), images

    
def list_png_files(directory):
    """List .png files in a given directory, sorted by modification time."""
    png_files = [os.path.join(directory, f) for f in os.listdir(directory) if f.endswith('.png')]
    png_files.sort(key=lambda x: os.path.getmtime(x), reverse=True)  # Newest first
    return png_files

def update_images(image_component, directory="/home/user/app/"):
    """Update the image component with the latest image from the directory."""
    png_files = list_png_files(directory)
    if png_files:
        # Load the most recent image file
        return png_files[0]
    return "No images available"

def create_chat_widget():
    with gr.Blocks() as chatblock:
        # Adding a row for the New Chat button at the top
        with gr.Row():
            new_chat_button = gr.Button("New Chat")
            new_chat_button.click(
                lambda: ([], ""),  
                inputs=[], 
                outputs=[chatbot] )

        # Main chat interface
        with gr.Row():
            chatbot = gr.Chatbot(
                [],
                elem_id="gpt4",
                elem_classes="gpt4",
                layout="llm",
                bubble_full_width=False,
                height=600,
            )
        with gr.Row():
            txt = gr.Textbox(
                placeholder="Enter text and press enter to chat with the bot.",
                show_label=False,
            )
            send_button = gr.Button("Send")
            txt.submit(update_bot, inputs=[txt, chatbot], outputs=[chatbot, txt])
            send_button.click(update_bot, inputs=[txt, chatbot], outputs=[chatbot, txt])

        # Image display row
        with gr.Row():
            image_display = gr.Image()
            update_image_button = gr.Button("Update Image")
            update_image_button.click(update_images, inputs=[], outputs=image_display)

        return chatblock


def update_bot(text, chatbot):
    if text:  # Append user input to the chat with "User:" prefix
        chatbot.append(("User", f"User: {text}"))
    response_json = interpreter.chat(text, stream=True, display=False)
    formatted_response, images = json_to_markdown(response_json)
    # Append text response with "Assistant:" prefix if it exists
    if formatted_response.strip():
        chatbot.append(("Assistant", f"Assistant: {formatted_response}"))
    # Append images if any
    for img_path in images:
        if os.path.exists(img_path) and img_path.endswith('.png'):
            chatbot.append(("Assistant", img_path))  # Assuming the path is accessible
        else:
            chatbot.append(("Assistant", "Image not found or path is invalid."))
    return chatbot, ""


with gr.Blocks() as demo:
    with gr.Tab("HEXON Chatbot Assignment"):
        chat_interface = create_chat_widget()
demo.launch()