app.py

import os
import gradio as gr
import requests
import inspect
import pandas as pd
from langgraph.prebuilt import ToolNode


# from typing import Any, Dict
# from typing import TypedDict, Annotated

from langchain_openai import ChatOpenAI
from langgraph.graph import StateGraph, START, END
from langgraph.graph.message import add_messages
from langchain.schema import HumanMessage, SystemMessage, AIMessage
# Create a ToolNode that knows about your web_search function
import json
from state import AgentState

# --- Constants ---
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"

from tools import ocr_image_tool, parse_excel_tool, web_search_tool, run_tools, audio_transcriber_tool, wikipedia_search_tool

llm = ChatOpenAI(model_name="gpt-4o-mini")

# ─── 1) plan_node ───
def plan_node(state: AgentState) -> AgentState:
    """
    Step 1: Ask GPT to draft a concise direct answer (INTERIM_ANSWER),
            then decide if it’s confident enough to stop or if it needs one tool.
    If confident: return {"final_answer": "<answer>"}
    Otherwise:   return exactly one of 
                 {"wiki_query": "..."},
                 {"web_search_query": "..."},
                 {"ocr_path": "..."},
                 {"excel_path": "...", "excel_sheet_name": "..."},
                 {"audio_path": "..."}
    """
    prior_msgs = state.get("messages", [])
    user_input = ""
    for msg in reversed(prior_msgs):
        if isinstance(msg, HumanMessage):
            user_input = msg.content
            break

    system_msg = SystemMessage(
        content=(
            "You are an agent that must do two things in one JSON output:\n\n"
            "  1) Provide a concise, direct answer to the user’s question (no explanation).\n"
            "  2) Judge whether that answer is reliable:\n"
            "     • If you are fully confident, return exactly:\n"
            "         {\"final_answer\":\"<your concise answer>\"}\n"
            "       and nothing else.\n"
            "     • Otherwise, return exactly one of:\n"
            "         {\"wiki_query\":\"<Wikipedia search>\"}\n"
            "         {\"web_search_query\":\"<search terms>\"}\n"
            "         {\"ocr_path\":\"<image path or task_id>\"}\n"
            "         {\"excel_path\":\"<xlsx path>\", \"excel_sheet_name\":\"<sheet name>\"}\n"
            "         {\"audio_path\":\"<audio path or task_id>\"}\n"
            "       and nothing else.\n"
            "Do NOT wrap in markdown—output only a single JSON object.\n"
            f"User’s question: \"{user_input}\"\n"
        )
    )
    human_msg = HumanMessage(content=user_input)
    llm_response = llm([system_msg, human_msg])
    llm_out = llm_response.content.strip()

    ai_msg = AIMessage(content=llm_out)
    new_msgs = prior_msgs.copy() + [ai_msg]

    try:
        parsed = json.loads(llm_out)
        if isinstance(parsed, dict):
            partial: AgentState = {"messages": new_msgs}
            allowed_keys = {
                "final_answer",
                "wiki_query",
                "web_search_query",
                "ocr_path",
                "excel_path",
                "excel_sheet_name",
                "audio_path"
            }
            for k, v in parsed.items():
                if k in allowed_keys:
                    partial[k] = v
            return partial
    except json.JSONDecodeError:
        pass

    return {
        "messages": new_msgs,
        "final_answer": "Sorry, I could not parse your intent."
    }


# ─── 2) store_prev_state ───
def store_prev_state(state: AgentState) -> AgentState:
    return {**state, "prev_state": state.copy()}


# ─── 3) tools_node ───
def tool_node(state: AgentState) -> AgentState:
    """
    Dispatch exactly one tool based on which key was set:
      - wiki_query → wikipedia_search_tool
      - web_search_query → web_search_tool
      - ocr_path → ocr_image_tool
      - excel_path → parse_excel_tool
      - audio_path → audio_transcriber_tool
    """
    if state.get("wiki_query"):
        return wikipedia_search_tool(state)
    if state.get("web_search_query"):
        return web_search_tool(state)
    if state.get("ocr_path"):
        return ocr_image_tool(state)
    if state.get("excel_path"):
        return parse_excel_tool(state)
    if state.get("audio_path"):
        return audio_transcriber_tool(state)
    return {}


# ─── 4) merge_tool_output ───
def merge_tool_output(state: AgentState) -> AgentState:
    prev_state = state.get("prev_state", {})
    merged = {**prev_state, **state}
    merged.pop("prev_state", None)

    # Detect which tool key was used in prev_state (it’s exactly one of these)
    for tool_key in ("wiki_query", "web_search_query", "ocr_path", "excel_path", "audio_path"):
        if prev_state.get(tool_key) is not None:
            # Increment the count of tool calls
            merged["tool_calls"] = merged.get("tool_calls", 0) + 1
            # Record that we have used this tool_key
            used = merged.get("used_tools", []).copy()
            if tool_key not in used:
                used.append(tool_key)
            merged["used_tools"] = used
            break

    return merged


# ─── 5) inspect_node ───
def inspect_node(state: AgentState) -> AgentState:
    messages_for_llm = []

    # 1) Original question
    question = ""
    for msg in reversed(state.get("messages", [])):
        if isinstance(msg, HumanMessage):
            question = msg.content
            break
    messages_for_llm.append(SystemMessage(content=f"USER_QUESTION: {question}"))

    # 2) Any tool results so far
    if sr := state.get("web_search_result"):
        messages_for_llm.append(SystemMessage(content=f"WEB_SEARCH_RESULT: {sr}"))
    if orc := state.get("ocr_result"):
        messages_for_llm.append(SystemMessage(content=f"OCR_RESULT: {orc}"))
    if exr := state.get("excel_result"):
        messages_for_llm.append(SystemMessage(content=f"EXCEL_RESULT: {exr}"))
    if tr := state.get("transcript"):
        messages_for_llm.append(SystemMessage(content=f"AUDIO_TRANSCRIPT: {tr}"))
    if wr := state.get("wiki_result"):
        messages_for_llm.append(SystemMessage(content=f"WIKIPEDIA_RESULT: {wr}"))

    # 3) Interim answer
    if ia := state.get("final_answer"):
        messages_for_llm.append(SystemMessage(content=f"INTERIM_ANSWER: {ia}"))

    # 4) How many times have we called a tool?
    used_tools = state.get("used_tools", [])
    tool_calls = state.get("tool_calls", 0)

    # If we've already tried all five tools once, or exceeded a small limit (e.g. 5),
    # force a final answer now. We append a dummy instruction so the LLM knows:
    if tool_calls >= 5 or len(used_tools) >= 5:
        # The user’s interim answer and all tool results exist;
        # we instruct GPT to treat it as final.
        prompt = (
            "We have already used every available tool or reached the maximum number of attempts.\n"
            "Therefore, return exactly {\"final_answer\":\"<your best final answer>\"} and nothing else.\n"
        )
        messages_for_llm.append(SystemMessage(content=prompt))
        llm_response = llm(messages_for_llm)
        raw = llm_response.content.strip()
        new_msgs = state["messages"] + [AIMessage(content=raw)]
        try:
            parsed = json.loads(raw)
            if isinstance(parsed, dict) and "final_answer" in parsed:
                return {"messages": new_msgs, "final_answer": parsed["final_answer"],
                        "used_tools": used_tools, "tool_calls": tool_calls}
        except json.JSONDecodeError:
            pass
        # Fallback
        return {
            "messages": new_msgs,
            "final_answer": "ERROR: inspect forced final but parsing failed.",
            "used_tools": used_tools,
            "tool_calls": tool_calls
        }

    # 5) Otherwise, ask GPT if it wants another tool
    prompt = (
        "You have a current draft answer (INTERIM_ANSWER) and possibly some tool results above.\n"
        "If you are confident it’s now correct, return exactly:\n"
        "  {\"final_answer\":\"<your final answer>\"}\n"
        "and nothing else.\n"
        "Otherwise, return exactly one of these JSON literals to fetch another tool, "
        "but DO NOT return a tool you have already used:\n"
        "  {\"wiki_query\":\"<query for Wikipedia>\"}\n"
        "  {\"web_search_query\":\"<search terms>\"}\n"
        "  {\"ocr_path\":\"<image path or task_id>\"}\n"
        "  {\"excel_path\":\"<xls path>\", \"excel_sheet_name\":\"<sheet name>\"}\n"
        "  {\"audio_path\":\"<audio path or task_id>\"}\n"
        "Do NOT wrap in markdown—return only the JSON object.\n"
        f"Already used tools: {used_tools}\n"
    )
    messages_for_llm.append(SystemMessage(content=prompt))
    llm_response = llm(messages_for_llm)
    raw = llm_response.content.strip()

    new_msgs = state["messages"] + [AIMessage(content=raw)]
    try:
        parsed = json.loads(raw)
        if isinstance(parsed, dict):
            # If GPT asks for a tool that’s already in used_tools, override and force finalize.
            for key in parsed:
                if key in ("wiki_query", "web_search_query", "ocr_path", "excel_path", "audio_path"):
                    if key in used_tools:
                        # GPT tried to reuse a tool → force a final answer instead
                        return {
                            "messages": new_msgs,
                            "final_answer": state.get("final_answer", ""),
                            "used_tools": used_tools,
                            "tool_calls": tool_calls
                        }
            # Otherwise, it’s either final_answer or a brand‐new tool request
            partial: AgentState = {
                "messages": new_msgs,
                "used_tools": used_tools,
                "tool_calls": tool_calls
            }
            for k, v in parsed.items():
                partial[k] = v
            return partial
    except json.JSONDecodeError:
        pass

    return {
        "messages": new_msgs,
        "final_answer": "ERROR: could not parse inspect decision.",
        "used_tools": used_tools,
        "tool_calls": tool_calls
    }


# ─── 6) finalize_node ───
def finalize_node(state: AgentState) -> AgentState:
    """
    If state already has "final_answer", return it. Otherwise, gather all tool outputs
    and ask GPT for a final answer. But in our cyclic design, finalize_node is only called
    after plan_node or inspect_node returned "final_answer".
    """
    if fa := state.get("final_answer"):
        return {"final_answer": fa}
    # (In practice, we never reach here because we always pick finalize only when "final_answer" exists.)
    return {"final_answer": "ERROR: finalize called without a final_answer."}


# ─── 7) Build the graph and wire edges ───
graph = StateGraph(AgentState)

# Register nodes
graph.add_node("plan", plan_node)
graph.add_node("store_prev_state", store_prev_state)
graph.add_node("tools", tool_node)
graph.add_node("merge_tool_output", merge_tool_output)
graph.add_node("inspect", inspect_node)
graph.add_node("finalize", finalize_node)

# START → plan
graph.add_edge(START, "plan")

# plan → either finalize (if plan set final_answer) or store_prev_state (if plan wants a tool)
def route_plan(plan_out: AgentState) -> str:
    if plan_out.get("final_answer") is not None:
        return "finalize"
    return "store_prev_state"

graph.add_conditional_edges(
    "plan",
    route_plan,
    {"store_prev_state": "store_prev_state", "finalize": "finalize"}
)

# store_prev_state → tools
graph.add_edge("store_prev_state", "tools")

# tools → merge_tool_output
graph.add_edge("tools", "merge_tool_output")

# merge_tool_output → inspect
graph.add_edge("merge_tool_output", "inspect")

# inspect → either finalize (if inspect set final_answer) or store_prev_state (if inspect wants another tool)
def route_inspect(inspect_out: AgentState) -> str:
    if inspect_out.get("final_answer") is not None:
        return "finalize"
    return "store_prev_state"

graph.add_conditional_edges(
    "inspect",
    route_inspect,
    {"store_prev_state": "store_prev_state", "finalize": "finalize"}
)

# finalize → END
graph.add_edge("finalize", END)

compiled_graph = graph.compile()


# ─── 8) respond_to_input ───
def respond_to_input(user_input: str, task_id: str) -> str:
    """
    Seed state['messages'] with a SystemMessage + HumanMessage(user_input),
    include the current task_id so that OCR/Audio tools can fetch files,
    then invoke the cyclic graph. Return the final_answer from the resulting state.
    """
    system_msg = SystemMessage(
        content=(
            "You are an agent orchestrator. Decide whether to use a tool or answer directly.\n"
            "Tools available:\n"
            "  • Wikipedia: set {\"wiki_query\":\"<search terms>\"}\n"
            "  • Web search: set {\"web_search_query\":\"<search terms>\"}\n"
            "  • OCR: set {\"ocr_path\":\"<image path or task_id>\"}\n"
            "  • Excel: set {\"excel_path\":\"<xlsx path>\", \"excel_sheet_name\":\"<sheet>\"}\n"
            "  • Audio transcription: set {\"audio_path\":\"<audio path or task_id>\"}\n"
            "If you can answer immediately, set {\"final_answer\":\"<answer>\"}.\n"
            "Respond with only one JSON object and no extra formatting."
        )
    )
    human_msg = HumanMessage(content=user_input)

    initial_state: AgentState = {
        "messages": [system_msg, human_msg],
        "task_id": task_id,
        "used_tools": [],    # track which tools have been requested
        "tool_calls": 0      # count of how many tool invocations so far
    }
    final_state = compiled_graph.invoke(initial_state)
    return final_state.get("final_answer", "Error: No final answer generated.")



class BasicAgent:
    def __init__(self):
        print("BasicAgent initialized.")
    def __call__(self, question: str, task_id) -> str:
        # print(f"Agent received question (first 50 chars): {question[:50]}...")
        # fixed_answer = "This is a default answer."
        # print(f"Agent returning fixed answer: {fixed_answer}")
        print()
        print()
        print()
        print()
        
        
        print(f"Agent received question: {question}")
        print()
        return respond_to_input(question, task_id)
        # return fixed_answer






def run_and_submit_all( profile: gr.OAuthProfile | None):
    """
    Fetches all questions, runs the BasicAgent on them, submits all answers,
    and displays the results.
    """
    # --- Determine HF Space Runtime URL and Repo URL ---
    space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code

    if profile:
        username= f"{profile.username}"
        print(f"User logged in: {username}")
    else:
        print("User not logged in.")
        return "Please Login to Hugging Face with the button.", None

    api_url = DEFAULT_API_URL
    questions_url = f"{api_url}/questions"
    submit_url = f"{api_url}/submit"

    # 1. Instantiate Agent ( modify this part to create your agent)
    try:
        agent = BasicAgent()
    except Exception as e:
        print(f"Error instantiating agent: {e}")
        return f"Error initializing agent: {e}", None
    # In the case of an app running as a hugging Face space, this link points toward your codebase ( usefull for others so please keep it public)
    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
    print(agent_code)

    # 2. Fetch Questions
    print(f"Fetching questions from: {questions_url}")
    try:
        response = requests.get(questions_url, timeout=15)
        response.raise_for_status()
        questions_data = response.json()
        if not questions_data:
             print("Fetched questions list is empty.")
             return "Fetched questions list is empty or invalid format.", None
        print(f"Fetched {len(questions_data)} questions.")
    except requests.exceptions.RequestException as e:
        print(f"Error fetching questions: {e}")
        return f"Error fetching questions: {e}", None
    except requests.exceptions.JSONDecodeError as e:
         print(f"Error decoding JSON response from questions endpoint: {e}")
         print(f"Response text: {response.text[:500]}")
         return f"Error decoding server response for questions: {e}", None
    except Exception as e:
        print(f"An unexpected error occurred fetching questions: {e}")
        return f"An unexpected error occurred fetching questions: {e}", None

    # 3. Run your Agent
    
    results_log = []
    answers_payload = []
    print(f"Running agent on {len(questions_data)} questions...")
    for item in questions_data:
        task_id = item.get("task_id")
        question_text = item.get("question")
        if not task_id or question_text is None:
            print(f"Skipping item with missing task_id or question: {item}")
            continue
        try:
            submitted_answer = agent(question_text, task_id)
            answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
            results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
        except Exception as e:
             print(f"Error running agent on task {task_id}: {e}")
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})

    if not answers_payload:
        print("Agent did not produce any answers to submit.")
        return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)

    # 4. Prepare Submission 
    submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
    status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
    print(status_update)

    # 5. Submit
    print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
    try:
        response = requests.post(submit_url, json=submission_data, timeout=60)
        response.raise_for_status()
        result_data = response.json()
        final_status = (
            f"Submission Successful!\n"
            f"User: {result_data.get('username')}\n"
            f"Overall Score: {result_data.get('score', 'N/A')}% "
            f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
            f"Message: {result_data.get('message', 'No message received.')}"
        )
        print("Submission successful.")
        results_df = pd.DataFrame(results_log)
        return final_status, results_df
    except requests.exceptions.HTTPError as e:
        error_detail = f"Server responded with status {e.response.status_code}."
        try:
            error_json = e.response.json()
            error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
        except requests.exceptions.JSONDecodeError:
            error_detail += f" Response: {e.response.text[:500]}"
        status_message = f"Submission Failed: {error_detail}"
        print(status_message)
        results_df = pd.DataFrame(results_log)
        return status_message, results_df
    except requests.exceptions.Timeout:
        status_message = "Submission Failed: The request timed out."
        print(status_message)
        results_df = pd.DataFrame(results_log)
        return status_message, results_df
    except requests.exceptions.RequestException as e:
        status_message = f"Submission Failed: Network error - {e}"
        print(status_message)
        results_df = pd.DataFrame(results_log)
        return status_message, results_df
    except Exception as e:
        status_message = f"An unexpected error occurred during submission: {e}"
        print(status_message)
        results_df = pd.DataFrame(results_log)
        return status_message, results_df


# --- Build Gradio Interface using Blocks ---
with gr.Blocks() as demo:
    gr.Markdown("# Basic Agent Evaluation Runner")
    gr.Markdown(
        """
        **Instructions:**

        1.  Please clone this space, then modify the code to define your agent's logic, the tools, the necessary packages, etc ...
        2.  Log in to your Hugging Face account using the button below. This uses your HF username for submission.
        3.  Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.

        ---
        **Disclaimers:**
        Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
        This space provides a basic setup and is intentionally sub-optimal to encourage you to develop your own, more robust solution. For instance for the delay process of the submit button, a solution could be to cache the answers and submit in a seperate action or even to answer the questions in async.
        """
    )

    gr.LoginButton()

    run_button = gr.Button("Run Evaluation & Submit All Answers")

    status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
    # Removed max_rows=10 from DataFrame constructor
    results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)

    run_button.click(
        fn=run_and_submit_all,
        outputs=[status_output, results_table]
    )

if __name__ == "__main__":
    # print("LangGraph version:", langgraph.__version__) 
    print("\n" + "-"*30 + " App Starting " + "-"*30)
    # Check for SPACE_HOST and SPACE_ID at startup for information
    space_host_startup = os.getenv("SPACE_HOST")
    space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
    # import langgraph
    # print("▶︎ LangGraph version:", langgraph.__version__)
    if space_host_startup:
        print(f"✅ SPACE_HOST found: {space_host_startup}")
        print(f"   Runtime URL should be: https://{space_host_startup}.hf.space")
    else:
        print("ℹ️  SPACE_HOST environment variable not found (running locally?).")

    if space_id_startup: # Print repo URLs if SPACE_ID is found
        print(f"✅ SPACE_ID found: {space_id_startup}")
        print(f"   Repo URL: https://huggingface.co/spaces/{space_id_startup}")
        print(f"   Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
    else:
        print("ℹ️  SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")

    print("-"*(60 + len(" App Starting ")) + "\n")

    print("Launching Gradio Interface for Basic Agent Evaluation...")
    demo.launch(debug=True, share=False)