import gradio as gr
from search import search_google
from scraper.py import scrape_url
from rag import VectorStore
from llm import generate_answer

vs = VectorStore()

def ask_agent(question):
    # Step 1: search
    urls = search_google(question, num_results=3)
    # Step 2: scrape
    texts = [scrape_url(url) for url in urls]
    # Step 3: embed + store
    vs.add_texts(texts)
    # Step 4: retrieve
    relevant = vs.retrieve(question, top_k=2)
    context = "\n\n".join(relevant)
    # Step 5: generate answer
    answer = generate_answer(context, question)
    return f"### 🧠 Answer:\n{answer}\n\n\n### 🔗 Sources:\n" + "\n".join(urls)

with gr.Blocks() as demo:
    gr.Markdown("# 🔍 AI Web RAG Agent\nAsk me anything; I'll search, scrape and answer!")
    with gr.Row():
        inp = gr.Textbox(label="Your question")
    out = gr.Markdown()
    btn = gr.Button("Ask")
    btn.click(fn=ask_agent, inputs=inp, outputs=out)

demo.launch()