app.py

# Streamlit is an open-source framework for building interactive web apps in Python.
# Run this file with the SAME interpreter the notebook uses (has the matching
# LangChain 1.x versions):
#   d:\2026\July\Agentic-RAG-with-LangGraph-and-Ollama\.venv\Scripts\python.exe -m streamlit run app.py
# (NOT `python app.py` — that skips Streamlit's runtime and breaks session state.)

import os
import time
import logging

import streamlit as st
from dotenv import load_dotenv
from langchain_openai import ChatOpenAI
from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace

# ---- Logging setup ---------------------------------------------------------
# Logs print to the terminal where you ran `streamlit run app.py`.
logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s | %(levelname)-7s | %(message)s",
    datefmt="%H:%M:%S",
)
log = logging.getLogger("langchain-demo")

# Load OPENAI_API_KEY and HUGGINGFACEHUB_API_TOKEN from the .env file
log.info("STEP 0: Loading environment variables from .env")
load_dotenv()
log.info("        OPENAI_API_KEY set:            %s", bool(os.getenv("OPENAI_API_KEY")))
log.info("        HUGGINGFACEHUB_API_TOKEN set:  %s", bool(os.getenv("HUGGINGFACEHUB_API_TOKEN")))


# ---- Answer function -------------------------------------------------------
# Same code shape as LLM_Intro.ipynb, with step-by-step logging to the terminal.
def load_answer(provider, question):
    t0 = time.perf_counter()
    log.info("STEP 1: Provider selected → %s", provider)
    log.info("STEP 2: Question → %r", question)

    if provider == "OpenAI":
        log.info("STEP 3: Building ChatOpenAI(model='gpt-4o')")
        llm = ChatOpenAI(model="gpt-4o")
        log.info("STEP 4: Calling llm.invoke(...) — waiting for OpenAI")
        answer = llm.invoke(question).content
    else:
        log.info("STEP 3: Building HuggingFaceEndpoint(repo_id='Qwen/Qwen2.5-72B-Instruct', "
                 "task='conversational', provider='novita')")
        llm = HuggingFaceEndpoint(
            repo_id="Qwen/Qwen2.5-72B-Instruct",
            task="conversational",
            provider="novita",
            huggingfacehub_api_token=os.getenv("HUGGINGFACEHUB_API_TOKEN"),
        )
        log.info("STEP 4: Wrapping in ChatHuggingFace and calling chat.invoke(...) — waiting for HF")
        chat = ChatHuggingFace(llm=llm)
        answer = chat.invoke(question).content

    elapsed = time.perf_counter() - t0
    log.info("STEP 5: Got response (%d chars) in %.2fs", len(answer), elapsed)
    return answer


# ---- App UI ----------------------------------------------------------------
st.set_page_config(page_title="LangChain Demo", page_icon=":robot:")
st.header("HuggingFace Demo")

# Sidebar: let the user choose the provider.
with st.sidebar:
    st.subheader("Model settings")
    provider = st.selectbox("Provider", ["OpenAI", "HuggingFace"])


# Gets the user input
def get_text():
    input_text = st.text_input("You: ", key="input")
    return input_text


user_input = get_text()

submit = st.button("Generate")

# Only call the LLM when the button is clicked and there is input
if submit and user_input:
    with st.spinner(f"Asking {provider}…"):
        try:
            response = load_answer(provider, user_input)
            st.subheader("Answer:")
            st.write(response)
        except Exception as e:
            log.exception("ERROR during load_answer")
            st.error(f"{type(e).__name__}: {e}")
elif submit:
    st.warning("Please enter a question first.")