app.py

import os
import re
import streamlit as st
import openai
from dotenv import load_dotenv
from langchain.agents.openai_assistant import OpenAIAssistantRunnable

# Load environment variables
load_dotenv()
api_key = os.getenv("OPENAI_API_KEY")
extractor_agent = os.getenv("ASSISTANT_ID_SOLUTION_SPECIFIER_A")

# Create the assistant
extractor_llm = OpenAIAssistantRunnable(
    assistant_id=extractor_agent,
    api_key=api_key,
    as_agent=True
)

def remove_citation(text: str) -> str:
    pattern = r"【\d+†\w+】"
    return re.sub(pattern, "📚", text)

# Initialize session state for messages and thread_id
if "messages" not in st.session_state:
    st.session_state["messages"] = []
if "thread_id" not in st.session_state:
    st.session_state["thread_id"] = None

st.title("Solution Specifier A")

def predict(user_input: str) -> str:
    """
    This function calls our OpenAIAssistantRunnable to get a response.
    We either create a new thread (no thread_id yet) or continue the existing thread.
    If the server complains that a run is still active, we reset the thread_id and retry once.
    """
    try:
        if st.session_state["thread_id"] is None:
            # Start a new thread
            response = extractor_llm.invoke({"content": user_input})
            st.session_state["thread_id"] = response.thread_id
        else:
            # Continue the existing thread
            response = extractor_llm.invoke(
                {"content": user_input, "thread_id": st.session_state["thread_id"]}
            )
        output = response.return_values["output"]
        return remove_citation(output)

    except openai.error.BadRequestError as e:
        # If the error says a run is still active, reset to a new thread and re-invoke once
        err_msg = str(e)
        if "while a run" in err_msg:
            st.session_state["thread_id"] = None
            # Re-invoke once to get a fresh thread
            try:
                response = extractor_llm.invoke({"content": user_input})
                st.session_state["thread_id"] = response.thread_id
                output = response.return_values["output"]
                return remove_citation(output)
            except Exception as e2:
                st.error(f"Error after resetting thread: {e2}")
                return ""
        else:
            # Some other bad request
            st.error(err_msg)
            return ""
    except Exception as e:
        # Catch-all for any other error
        st.error(str(e))
        return ""

# Display any existing messages
for msg in st.session_state["messages"]:
    if msg["role"] == "user":
        with st.chat_message("user"):
            st.write(msg["content"])
    else:
        with st.chat_message("assistant"):
            st.write(msg["content"])

# Create the chat input widget at the bottom of the page
user_input = st.chat_input("Type your message here...")

# When the user hits ENTER on st.chat_input
if user_input:
    # Add the user message to session state
    st.session_state["messages"].append({"role": "user", "content": user_input})

    # Display the user's message
    with st.chat_message("user"):
        st.write(user_input)

    # Get the assistant's response
    response_text = predict(user_input)

    # Add the assistant response to session state
    st.session_state["messages"].append({"role": "assistant", "content": response_text})

    # Display the assistant's reply
    with st.chat_message("assistant"):
        st.write(response_text)