app.py

import os
import gradio as gr
from langchain_core.messages import AnyMessage, HumanMessage, AIMessage
from agent import build_graph

def create_qna_interface():
    """Create the Q&A chatbot interface."""

    # ======================================
    # 2) Helper functions for the Gradio UI
    # ======================================

    def _msg_content_to_str(msg: AnyMessage) -> str:
        """
        Coerce LangChain message content (which might contain tool call structures)
        into displayable text for the Chatbot.
        """
        # Most often, content is a string already
        content = getattr(msg, "content", "")
        if isinstance(content, str):
            return content

        # If it's a list of parts (e.g., tool call traces), join any text parts
        if isinstance(content, list):
            texts = []
            for part in content:
                if isinstance(part, dict) and "text" in part:
                    texts.append(part["text"])
                elif isinstance(part, str):
                    texts.append(part)
            return "\n".join(texts) if texts else str(content)

        # Fallback
        return str(content)

    def startup_state() -> list[AnyMessage]:
        """Start with an empty conversation."""
        return []
    
    # Gradio expects chatbot history as list[tuple[str, str]]
    def submit_user_message(
        user_text: str,
        chat_history: list[tuple[str, str]],
        agent_messages: list[AnyMessage],
    ):
        """
        1) Append HumanMessage to agent state
        2) Run Alfred
        3) Extract last AIMessage and append to chat_history
        """
        if not user_text or user_text.strip() == "":
            return gr.update(), chat_history, agent_messages

        # Step 1: add HumanMessage to state
        agent_messages = list(agent_messages or [])
        agent_messages.append(HumanMessage(content=user_text))

        # get API key
        api_key = os.getenv('HF_TOKEN')

        # Step 2: run the graph
        alfred = build_graph(hf_token=api_key)
        out = alfred.invoke({"messages": agent_messages})

        # The graph returns a new messages list *including* the latest assistant/tool steps.
        # We use the last AIMessage as the displayed reply.
        new_msgs: list[AnyMessage] = out["messages"]
        agent_messages = new_msgs  # keep full state for the next turn

        # Find the last assistant message to show in the UI
        ai_text = ""
        for m in reversed(new_msgs):
            if isinstance(m, AIMessage):
                ai_text = _msg_content_to_str(m)
                break
        if not ai_text:
            # fallback: in rare cases of only tool messages, show a generic note
            ai_text = "I processed your request using my tools."

        chat_history = list(chat_history or [])
        chat_history.append({"role": "user", "content": user_text})
        chat_history.append({"role": "assistant", "content": ai_text})
        return "", chat_history, agent_messages

    def clear_chat():
        """Reset the Gradio UI and agent state."""
        return [], startup_state()
    
    # ========================
    # 3) Gradio App UI layout
    # ========================

    with gr.Blocks(title="Alfred — LangGraph Agent") as demo:
        gr.Markdown(
            """
            # 🎩 Alfred — Your LangGraph Agent
            Ask questions and Alfred will respond, using:
            - a vector search tool over the guest list
            - DuckDuckGo web search
            """
        )

        with gr.Row():
            chatbot = gr.Chatbot(
                label="Conversation",
                type="messages",
                height=500,
                show_copy_button=True,
                avatar_images=(None, None),  # customize if you like
            )

        with gr.Row():
            txt = gr.Textbox(
                label="Your message",
                placeholder="Ask anything…",
                autofocus=True,
                scale=4,
            )
            send_btn = gr.Button("Send", variant="primary", scale=1)
            clear_btn = gr.Button("Clear")

        # Hidden state: the agent’s full message list (LangChain messages)
        agent_state = gr.State(startup_state())

        # Wire up events
        txt.submit(submit_user_message, [txt, chatbot, agent_state], [txt, chatbot, agent_state])
        send_btn.click(submit_user_message, [txt, chatbot, agent_state], [txt, chatbot, agent_state])
        clear_btn.click(clear_chat, outputs=[chatbot, agent_state])

        return demo


# Entry point
if __name__ == "__main__":
    # Create and launch the interface
    demo = create_qna_interface()
    demo.launch()