app.py

import streamlit as st
import os
from langchain_groq import ChatGroq
from langgraph.graph import MessagesState, StateGraph, START
from langgraph.checkpoint.memory import MemorySaver
from langchain_core.messages import HumanMessage, AIMessage

# Load API key from Hugging Face Secrets
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
if not GROQ_API_KEY:
    st.error("GROQ_API_KEY not found. Please add it as a Secret in Hugging Face Spaces.")
    st.stop()

# Initialize the Groq LLM
llm = ChatGroq(groq_api_key=GROQ_API_KEY, model_name="llama3-8b-8192")

# Helper functions
def extract_display_text(message):
    if "Rewritten text in" in message:
        return message.split(": ", 1)[1]
    elif "Text approved:" in message:
        return message
    return message

# Assistant function
def assistant(state: MessagesState):
    messages = state["messages"]
    latest_msg = messages[-1].content

    if "Rewrite this text in" in latest_msg:
        tone = latest_msg.split(" in ")[1].split(" tone:")[0]
        text = latest_msg.split("tone: ")[1]
        response = llm.invoke(f"Rewrite the following legal text in a {tone} tone: {text}")
        return {"messages": [AIMessage(content=f"Rewritten text in {tone} tone: {response.content}")]}

    elif latest_msg == "regenerate":
        for msg in messages:
            if isinstance(msg, HumanMessage) and "Rewrite this text in" in msg.content:
                original_text = msg.content.split("tone: ")[1]
                tone = msg.content.split(" in ")[1].split(" tone:")[0]
                response = llm.invoke(f"Rewrite the following legal text in a {tone} tone: {original_text}")
                return {"messages": [AIMessage(content=f"Rewritten text in {tone} tone: {response.content}")]}

    elif latest_msg.startswith("feedback:"):
        feedback = latest_msg.split("feedback: ")[1].strip()
        for msg in reversed(messages):
            if isinstance(msg, AIMessage) and "Rewritten text in" in msg.content:
                text = msg.content.split(": ", 1)[1]
                tone = msg.content.split(" in ")[1].split(" tone")[0]
                response = llm.invoke(f"Refine the following text based on this feedback: '{feedback}'. Maintain the {tone} tone. Text: {text}")
                return {"messages": [AIMessage(content=f"Rewritten text in {tone} tone: {response.content}")]}

    elif latest_msg.lower() == "approve":
        for msg in reversed(messages):
            if isinstance(msg, AIMessage) and "Rewritten text in" in msg.content:
                return {"messages": [AIMessage(content=f"Text approved: {msg.content.split(': ', 1)[1]}")]}

    return {"messages": [AIMessage(content="Invalid command.")]}

# Human feedback node (stop condition)
def human_feedback(state: MessagesState):
    return state

# Build the LangGraph pipeline
builder = StateGraph(MessagesState)
builder.add_node("assistant", assistant)
builder.add_node("human_feedback", human_feedback)

# Define edges with a stopping condition
builder.add_edge(START, "assistant")
builder.add_edge("assistant", "human_feedback")  # Stops after assistant
builder.add_edge("human_feedback", "assistant")  # Only runs if more input is provided

memory = MemorySaver()
graph = builder.compile(interrupt_before=["human_feedback"], checkpointer=memory)

# Streamlit UI
st.title("📜 Legal Text Rewriter")
st.markdown("Rewrite legal text into different tones using AI.")

# Store session state
if "state" not in st.session_state:
    st.session_state.state = {"configurable": {"thread_id": "1"}}

# User input section
legal_text = st.text_area("Enter Legal Text:", "")
initial_tone = st.selectbox("Select Tone:", ["Formal", "Empathetic", "Neutral", "Strength-Based"])
submit_btn = st.button("Submit")

if submit_btn and legal_text.strip():
    command = f"Rewrite this text in {initial_tone} tone: {legal_text}"
    graph.update_state(st.session_state.state, {"messages": [HumanMessage(content=command)]})
    for event in graph.stream(None, st.session_state.state, stream_mode="values"):
        if "messages" in event:
            st.session_state.rewritten_text = extract_display_text(event["messages"][-1].content)

# Display rewritten text
if "rewritten_text" in st.session_state:
    st.subheader("Rewritten Text")
    st.write(st.session_state.rewritten_text)

# Buttons
col1, col2, col3 = st.columns([1, 2, 1])

with col1:
    if st.button("🔄 Regenerate"):
        graph.update_state(st.session_state.state, {"messages": [HumanMessage(content="regenerate")]})
        for event in graph.stream(None, st.session_state.state, stream_mode="values"):
            if "messages" in event:
                st.session_state.rewritten_text = extract_display_text(event["messages"][-1].content)

with col2:
    feedback_text = st.text_input("Enter Feedback:")
    if st.button("💬 Submit Feedback"):
        graph.update_state(st.session_state.state, {"messages": [HumanMessage(content=f'feedback: {feedback_text}')]})
        for event in graph.stream(None, st.session_state.state, stream_mode="values"):
            if "messages" in event:
                st.session_state.rewritten_text = extract_display_text(event["messages"][-1].content)

with col3:
    if st.button("✅ Approve"):
        graph.update_state(st.session_state.state, {"messages": [HumanMessage(content="approve")]})
        for event in graph.stream(None, st.session_state.state, stream_mode="values"):
            if "messages" in event:
                st.session_state.rewritten_text = extract_display_text(event["messages"][-1].content)