app.py

import gradio as gr
import requests
import os
import json
from TTS.api import TTS
from langchain.vectorstores import FAISS
from langchain.embeddings import HuggingFaceEmbeddings

try:
    from langchain.vectorstores import FAISS
    print("✅ LangChain loaded successfully.")
except Exception as e:
    print(f"❌ LangChain failed: {str(e)}")



# Load TTS model
tts = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False, gpu=False)

# Memory functions
def save_user_prefs(user_id, prefs):
    with open(f"{user_id}_prefs.json", "w") as f:
        json.dump(prefs, f)

def load_user_prefs(user_id):
    try:
        with open(f"{user_id}_prefs.json", "r") as f:
            return json.load(f)
    except FileNotFoundError:
        return {}

# RAG context
def retrieve_context(query):
    db = FAISS.load_local("rag_vectorstore", HuggingFaceEmbeddings())
    docs = db.similarity_search(query)
    return "\n".join([doc.page_content for doc in docs])

# TTS wrapper
def speak_response(text):
    tts.tts_to_file(text=text, file_path="response.wav")
    return "response.wav"

# Main response logic
def generate_response(message, request: gr.Request):
    token = request.headers.get("Authorization")
    if not token:
        return "⚠️ You must be signed in to use Hermes.", None

    try:
        hermes_model = gr.load(
            "models/NousResearch/Hermes-4-70B",
            provider="nebius",
            token=token
        )
    except Exception as e:
        return f"🚫 Model load failed: {str(e)}", None

    try:
        reply = hermes_model(message)
        audio_path = speak_response(reply)
        return reply, audio_path
    except Exception as e:
        return f"❌ Hermes failed to respond: {str(e)}", None

# File dropdown updater
def update_file_list(file):
    return [os.path.basename(file.name)]

# UI
with gr.Blocks(title="Hermes Chat Interface") as demo:
    gr.Markdown("## 🧠 Hermes Chat Interface\nSpeak to the solar bard. Login required.")

    with gr.Row():
        file_upload = gr.File(
            label="Upload Files",
            file_types=[".txt", ".pdf", ".py", ".doc", ".md", ".png", ".jpg"],
            type="filepath",
            interactive=True
        )
        file_selector = gr.Dropdown(
            label="Select Uploaded File",
            choices=[],
            interactive=True
        )

    file_upload.change(fn=update_file_list, inputs=[file_upload], outputs=[file_selector])

    with gr.Row():
        with gr.Column():
            user_input = gr.Textbox(label="Your question")
            submit_btn = gr.Button("Ask Hermes")
        with gr.Column():
            response_text = gr.Textbox(label="Hermes says")
            response_audio = gr.Audio(label="Voice", type="filepath")

    with gr.Accordion("⚙️ Settings", open=False):
        model_selector = gr.Dropdown(
            label="Choose Model",
            choices=["Hermes", "Qwen", "OSS GPT", "DeepSeek"],
            value="Hermes"
        )
        voice_pitch = gr.Slider(label="Voice Pitch", minimum=0.5, maximum=2.0, value=1.0)
        voice_rate = gr.Slider(label="Voice Speed", minimum=0.5, maximum=2.0, value=1.0)
        voice_engine = gr.Radio(
            label="Voice Engine",
            choices=["Local TTS", "Hugging Face TTS", "Google TTS", "Microsoft TTS"],
            value="Local TTS"
        )
        gemini_key = gr.Textbox(label="Gemini Pro/Flash API Key", type="password")
        rag_toggle = gr.Checkbox(label="Enable RAG (Retrieval-Augmented Generation)", value=False)

    submit_btn.click(
        fn=generate_response,
        inputs=[user_input],
        outputs=[response_text, response_audio]
    )

demo.launch(ssr_mode=False)