app.py

import streamlit as st
from multi_agent import init_system, ask_multi_agent

st.set_page_config(
    page_title="Multi-Agent AI Assistant",
    page_icon="🤖",
    layout="centered"
)

import os
from ingest import run_ingestion

# Session state for tracking the database path
if "db_path" not in st.session_state:
    st.session_state.db_path = "data/demo.sqlite"

# Use st.cache_resource so we only load the heavy ML models and DB connections once!
@st.cache_resource(show_spinner="Booting up AI Agents...")
def load_agents(db_path):
    if not os.path.exists("faiss_index") or not os.path.exists("bm25_index.pkl"):
        print("Indices missing! Running automatic ingestion...")
        run_ingestion()
    return init_system(db_path)

agents_loaded = False
try:
    sql_agent, rag_agent, router = load_agents(st.session_state.db_path)
    agents_loaded = True
except Exception as e:
    st.warning(f"Waiting for data: {e} Please use the sidebar to upload a Document and a Database!")

st.title("Multi-Agent AI Assistant 🤖")
st.markdown("Ask a question about our structured database (SQL) or unstructured documents (RAG). The Master Router will automatically decide which expert agent to use!")

with st.sidebar:
    st.header("Upload Data")
    st.caption("Upload your own files to replace the dummy data!")
    
    # Document uploader
    doc_file = st.file_uploader("Upload Company Policy (.md, .txt, .pdf)", type=["md", "txt", "pdf"])
    if doc_file is not None:
        if st.button("Process Document"):
            os.makedirs("documents", exist_ok=True)
            file_path = os.path.join("documents", doc_file.name)
            with open(file_path, "wb") as f:
                f.write(doc_file.getbuffer())
            with st.spinner("Indexing document into Vector Database..."):
                run_ingestion()
                st.cache_resource.clear()
            st.success(f"Successfully indexed {doc_file.name}!")
            st.rerun()
            
    st.divider()
    
    # SQL uploader
    db_file = st.file_uploader("Upload SQLite Database (.sqlite, .db)", type=["sqlite", "db"])
    if db_file is not None:
        if st.button("Connect Database"):
            os.makedirs("data", exist_ok=True)
            file_path = os.path.join("data", db_file.name)
            with open(file_path, "wb") as f:
                f.write(db_file.getbuffer())
            st.session_state.db_path = file_path
            st.cache_resource.clear()
            st.success(f"Successfully connected to {db_file.name}!")
            st.rerun()

# Initialize chat history in Streamlit session state
if "messages" not in st.session_state:
    st.session_state.messages = []

# Display chat history on app rerun
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])
        if "route" in message:
            st.caption(f"🧠 *Routed via {message['route'].upper()} Agent*")

if agents_loaded:
    # React to user input
    if prompt := st.chat_input("What is your question?"):
        # Display user message in chat message container
        with st.chat_message("user"):
            st.markdown(prompt)
        # Add user message to chat history
        st.session_state.messages.append({"role": "user", "content": prompt})

        with st.spinner("Agent is thinking..."):
            # Format the last 6 messages as chat history so the agent understands follow-ups
            history_str = ""
            recent_messages = st.session_state.messages[-7:-1] # Exclude the prompt we just added
            for m in recent_messages:
                role = "User" if m["role"] == "user" else "Assistant"
                history_str += f"{role}: {m['content']}\n"
                
            # The magic happens here!
            route, response = ask_multi_agent(prompt, sql_agent, rag_agent, router, history_str)
            
        # Display assistant response in chat message container
        with st.chat_message("assistant"):
            st.markdown(response)
            st.caption(f"🧠 *Routed via {route.upper()} Agent*")
            
        # Add assistant response to chat history
        st.session_state.messages.append({
            "role": "assistant", 
            "content": response, 
            "route": route
        })