app.py

import gradio as gr
from my_memory_logic import run_with_session_memory

def chat_interface_fn(message, history, session_id):
    """
    Multi-turn chat function for Gradio's ChatInterface.
    'session_id' is used to store conversation across turns.
    Deduplicates consecutive repeated Q&A pairs to avoid repetition.
    """
    # Ensure history is a list of dictionaries
    if history and isinstance(history[0], tuple):
        print("DEBUG: Converting history from tuple format to dictionary format.")
        history = [
            msg for h in history 
            for msg in [
                {"role": "user", "content": h[0]},
                {"role": "assistant", "content": h[1]}
            ]
        ]
    
    # 1) Get answer from the session-based memory pipeline
    answer = run_with_session_memory(message, session_id)
    
    # 2) Deduplicate consecutive identical exchanges
    if not history or history[-1]["content"] != answer:
        history.append({"role": "user", "content": message})
        history.append({"role": "assistant", "content": answer})
    
    # 3) Convert history to message dictionaries for display
    message_dicts = []
    for msg in history:
        message_dicts.append(msg)
    
    # Return the message dicts and updated history
    return message_dicts, history

# Custom CSS for chat interface
my_chat_css = """
.gradio-container {
    margin: auto;
}
.user .wrap {
    text-align: right !important;
}
.assistant .wrap {
    text-align: left !important;
}
"""

# Set up Gradio interface
with gr.Blocks(css=my_chat_css) as demo:
    gr.Markdown("### DailyWellnessAI (User on right, Assistant on left)")
    session_id_box = gr.Textbox(label="Session ID", value="abc123", interactive=True)
    
    chat_interface = gr.ChatInterface(
        fn=lambda msg, hist: chat_interface_fn(msg, hist, session_id_box.value),
        title="DailyWellnessAI (Session-based Memory)",
        description="Ask your questions. The session_id determines your stored memory."
    )

# Launch the Gradio interface
demo.launch()