app.py

import gradio as gr
import os
import json
import uuid
import threading
import time
import re
from dotenv import load_dotenv
from openai import OpenAI
from realtime_transcriber import WebSocketClient, connections, WEBSOCKET_URI, WEBSOCKET_HEADERS

# ------------------ Load Secrets ------------------
load_dotenv()
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
ASSISTANT_ID = os.getenv("ASSISTANT_ID")

if not OPENAI_API_KEY or not ASSISTANT_ID:
    raise ValueError("Missing OPENAI_API_KEY or ASSISTANT_ID")

client = OpenAI(api_key=OPENAI_API_KEY)
session_threads = {}

# ------------------ Session & Chat Logic ------------------
def reset_session():
    session_id = str(uuid.uuid4())
    session_threads[session_id] = client.beta.threads.create().id
    return session_id

def process_chat(message, history, session_id):
    thread_id = session_threads.get(session_id)
    if not thread_id:
        thread_id = client.beta.threads.create().id
        session_threads[session_id] = thread_id

    client.beta.threads.messages.create(thread_id=thread_id, role="user", content=message)
    run = client.beta.threads.runs.create(thread_id=thread_id, assistant_id=ASSISTANT_ID)

    while client.beta.threads.runs.retrieve(thread_id=thread_id, run_id=run.id).status != "completed":
        time.sleep(1)

    messages = client.beta.threads.messages.list(thread_id=thread_id)
    for msg in reversed(messages.data):
        if msg.role == "assistant":
            return msg.content[0].text.value
    return "⚠️ Assistant did not respond."

def extract_image_url(text):
    match = re.search(r'https://raw\.githubusercontent\.com/[^\s"]+\.png', text)
    return match.group(0) if match else None

def handle_chat(message, history, session_id):
    response = process_chat(message, history, session_id)
    history.append((message, response))
    image = extract_image_url(response)
    return history, image

# ------------------ Transcript Actions ------------------
def ask_from_transcript(transcript, history, session_id):
    return handle_chat(transcript, history, session_id)

def clear_all(client_id):
    if client_id in connections:
        connections[client_id].transcript = ""
    return "", []

# ------------------ Real-Time Voice WebSocket ------------------
def create_websocket_client():
    client_id = str(uuid.uuid4())
    connections[client_id] = WebSocketClient(WEBSOCKET_URI, WEBSOCKET_HEADERS, client_id)
    threading.Thread(target=connections[client_id].run, daemon=True).start()
    return client_id

def clear_transcript(client_id):
    if client_id in connections:
        connections[client_id].transcript = ""
    return ""

def send_audio_chunk(audio, client_id):
    if client_id not in connections:
        return "Initializing connection..."
    sr, y = audio
    connections[client_id].enqueue_audio_chunk(sr, y)
    return connections[client_id].transcript

# ------------------ UI ------------------
with gr.Blocks(theme=gr.themes.Soft(), css="""
@media (max-width: 768px) {
  .gr-col { width: 100% !important; }
  .gr-row > div { flex-direction: column !important; }
}
""") as demo:
    gr.Markdown("# 🧠 Document AI + 🎙️ Voice Assistant")

    session_id = gr.State(value=reset_session())
    client_id = gr.State()

    with gr.Row():
        with gr.Column(scale=1):
            image_display = gr.Image(label="📑 Extracted Document Image", show_label=True, height=480)
        with gr.Column(scale=2):
            chatbot = gr.Chatbot(label="💬 Document Assistant", height=480)
            text_input = gr.Textbox(label="Ask about the document", placeholder="e.g. What is clause 3.2?")
            send_btn = gr.Button("Send")

    send_btn.click(handle_chat, inputs=[text_input, chatbot, session_id], outputs=[chatbot, image_display])
    text_input.submit(handle_chat, inputs=[text_input, chatbot, session_id], outputs=[chatbot, image_display])

    with gr.Accordion("🎤 Or Use Voice Instead", open=False):
        with gr.Row():
            transcript_box = gr.Textbox(label="Live Transcript", lines=6, interactive=False, autoscroll=True)
        with gr.Row():
            mic_input = gr.Audio(label="🎙️ Record", streaming=True)
        with gr.Row():
            ask_btn = gr.Button("🤖 Ask from Transcript", size="lg")
            clear_btn = gr.Button("🧹 Clear Chat", size="lg")

        mic_input.stream(fn=send_audio_chunk, inputs=[mic_input, client_id], outputs=transcript_box)
        ask_btn.click(fn=ask_from_transcript, inputs=[transcript_box, chatbot, session_id], outputs=[chatbot, image_display])
        clear_btn.click(fn=clear_all, inputs=[client_id], outputs=[transcript_box, chatbot])
        demo.load(fn=create_websocket_client, outputs=client_id)

demo.launch()