app.py

import streamlit as st
from typing import TypedDict, Annotated
from langgraph.graph import StateGraph
from langgraph.checkpoint.memory import MemorySaver
from langgraph.graph.message import add_messages
from langchain_openai import ChatOpenAI
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_core.messages import HumanMessage, ToolMessage, AIMessage
from langgraph.prebuilt import ToolNode, tools_condition
import os

# Streamlit UI Header
st.title("Checkpoints and Breakpoints")
st.caption("Demonstrating workflow execution with checkpoints and tool invocation.")

# Fetch API Keys
openai_api_key = os.getenv("OPENAI_API_KEY")
tavily_api_key = os.getenv("TAVILY_API_KEY")

if openai_api_key and tavily_api_key:
    os.environ["OPENAI_API_KEY"] = openai_api_key
    os.environ["TAVILY_API_KEY"] = tavily_api_key

    # Define State Class
    class State(TypedDict):
        messages: Annotated[list, add_messages]

    # Initialize LLM and Tools
    llm = ChatOpenAI(model="gpt-4o-mini")
    tool = TavilySearchResults(max_results=2)
    llm_with_tools = llm.bind_tools([tool])

    # Agent Function
    def Agent(state: State):
        st.sidebar.write("Agent received state:", state["messages"])
        response = llm_with_tools.invoke(state["messages"])
        st.sidebar.write("Agent Response:", response)
        return {"messages": [response]}

    # Tool Execution Logic
    def ToolExecutor(state: State):
        last_message = state["messages"][-1]
        if "tool_calls" in last_message.additional_kwargs:
            tool_call = last_message.additional_kwargs["tool_calls"][0]
            tool_name = tool_call["function"]["name"]
            tool_args = tool_call["function"]["arguments"]

            st.sidebar.write("Tool Call Detected:", tool_name, tool_args)

            # Execute the tool with provided arguments
            if tool_name == "tavily_search_results_json":
                query = eval(tool_args)["query"]  # Convert stringified arguments to dict
                tool_response = tool.invoke({"query": query})
                st.sidebar.write("Tool Response:", tool_response)

                # Return the tool response in a new ToolMessage
                return {"messages": [ToolMessage(content=str(tool_response), tool_call_id=tool_call["id"])]}

        return state  # Fallback

    # Memory Checkpoint
    memory = MemorySaver()

    # Build the Graph
    graph = StateGraph(State)
    graph.add_node("Agent", Agent)
    graph.add_node("ToolExecutor", ToolExecutor)

    graph.add_conditional_edges("Agent", tools_condition, {"True": "ToolExecutor"})
    graph.add_edge("ToolExecutor", "Agent")
    graph.set_entry_point("Agent")

    # Compile Graph
    app = graph.compile(checkpointer=memory, interrupt_before=["ToolExecutor"])

    # Display Graph Visualization
    st.subheader("Graph Workflow")
    st.image(app.get_graph().draw_mermaid_png(), caption="Graph Visualization", use_container_width=True)

    # Input Section
    st.subheader("Run the Workflow")
    user_input = st.text_input("Enter a message to start the graph:", "Search for the weather in Uttar Pradesh")
    thread_id = st.text_input("Thread ID", "1")

    if st.button("Execute Workflow"):
        thread = {"configurable": {"thread_id": thread_id}}
        input_message = {"messages": [HumanMessage(content=user_input)]}

        st.write("### Execution Outputs")
        outputs = []
        for event in app.stream(input_message, thread, stream_mode="values"):
            st.code(event["messages"][-1].content)
            outputs.append(event["messages"][-1].content)
            st.sidebar.write("Intermediate State:", event["messages"])

        if outputs:
            st.subheader("Intermediate Outputs")
            for idx, output in enumerate(outputs, start=1):
                st.write(f"**Step {idx}:**")
                st.code(output)
        else:
            st.warning("No outputs generated. Adjust your input to trigger tools.")

        # Display Snapshot of State
        st.subheader("Current State Snapshot")
        snapshot = app.get_state(thread)
        current_message = snapshot.values["messages"][-1]
        st.code(current_message.pretty_print())

        # Manual Update Section
        if hasattr(current_message, "tool_calls") and current_message.tool_calls:
            tool_call_id = current_message.tool_calls[0]["id"]
            manual_response = st.text_area("Manual Tool Response", "Enter your response to continue execution...")
            if st.button("Update State"):
                new_messages = [
                    ToolMessage(content=manual_response, tool_call_id=tool_call_id),
                    AIMessage(content=manual_response),
                ]
                app.update_state(thread, {"messages": new_messages})
                st.success("State updated successfully!")
                st.code(app.get_state(thread).values["messages"][-1].pretty_print())
        else:
            st.warning("No tool calls available for manual updates.")
else:
    st.error("API keys are missing! Please set `OPENAI_API_KEY` and `TAVILY_API_KEY` in Hugging Face Spaces Secrets.")