app.py

import os
import datetime
import requests
import pandas as pd
import gradio as gr
from openai import OpenAI

class ToolEnhancedAgent:
    def __init__(self):
        api_key = os.getenv("OPENAI_API_KEY")
        if not api_key:
            raise ValueError("OPENAI_API_KEY not found in environment variables.")
        self.client = OpenAI(api_key=api_key)
        print("ToolEnhancedAgent initialized.")

    def use_tool(self, tool_name: str, input_text: str) -> str:
        try:
            if tool_name == "calculator":
                # Sangat dasar dan raw eval, hati2 di produksi
                return str(eval(input_text))
            elif tool_name == "date":
                return str(datetime.datetime.now().date())
            elif tool_name == "wikipedia":
                return self.search_wikipedia(input_text)
            else:
                return "[Unknown tool]"
        except Exception as e:
            return f"[Tool error: {e}]"

    def search_wikipedia(self, query: str) -> str:
        try:
            url = f"https://en.wikipedia.org/api/rest_v1/page/summary/{query}"
            res = requests.get(url)
            if res.status_code == 200:
                return res.json().get("extract", "No summary found.")
            else:
                return f"No Wikipedia summary for '{query}'."
        except Exception as e:
            return f"Wikipedia API error: {e}"

    def __call__(self, question: str) -> str:
        # Prompt dengan chain of thought agar GPT berpikir langkah demi langkah
        prompt = (
            "You are an assistant that solves problems step-by-step, and you can use tools like calculator, date, and wikipedia if needed.\n"
            f"Question: {question}\n"
            "Answer (think step-by-step and use tools if helpful):"
        )

        try:
            response = self.client.chat.completions.create(
                model="gpt-4o-mini",  # model yang lebih ringan dan biasanya cukup
                messages=[
                    {"role": "system", "content": "You are a helpful assistant that thinks step-by-step and can use tools."},
                    {"role": "user", "content": prompt}
                ],
                temperature=0.2,
                max_tokens=700,
            )
            answer = response.choices[0].message.content.strip()
            print(f"Generated answer (preview): {answer[:100]}...")
            return answer
        except Exception as e:
            print(f"Agent error: {e}")
            return f"[Agent error: {e}]"


def run_and_submit_all(profile: gr.OAuthProfile | None):
    if profile is None:
        return "Please login to Hugging Face to submit answers.", None

    username = profile.username
    api_url = "https://agents-course-unit4-scoring.hf.space"
    questions_url = f"{api_url}/questions"
    submit_url = f"{api_url}/submit"

    try:
        agent = ToolEnhancedAgent()
    except Exception as e:
        return f"Agent initialization error: {e}", None

    try:
        response = requests.get(questions_url, timeout=15)
        response.raise_for_status()
        questions = response.json()
        if not isinstance(questions, list) or len(questions) == 0:
            return "Failed to fetch questions or empty list.", None
    except Exception as e:
        return f"Error fetching questions: {e}", None

    answers = []
    results_log = []

    for item in questions:
        task_id = item.get("task_id")
        question_text = item.get("question")
        if not task_id or question_text is None:
            continue
        try:
            answer = agent(question_text)
            answers.append({"task_id": task_id, "submitted_answer": answer})
            results_log.append({
                "Task ID": task_id,
                "Question": question_text,
                "Submitted Answer": answer
            })
        except Exception as e:
            results_log.append({
                "Task ID": task_id,
                "Question": question_text,
                "Submitted Answer": f"[Agent error: {e}]"
            })

    if not answers:
        return "Agent did not produce answers.", pd.DataFrame(results_log)

    agent_code_url = f"https://huggingface.co/spaces/{os.getenv('SPACE_ID')}/tree/main"

    submission_payload = {
        "username": username,
        "agent_code": agent_code_url,
        "answers": answers
    }

    try:
        submit_response = requests.post(submit_url, json=submission_payload, timeout=60)
        submit_response.raise_for_status()
        result = submit_response.json()

        final_status = (
            f"Submission Success!\n"
            f"User: {result.get('username')}\n"
            f"Score: {result.get('score', 'N/A')}%\n"
            f"Correct: {result.get('correct_count', '?')}/{result.get('total_attempted', '?')}\n"
            f"Message: {result.get('message', 'No message')}"
        )
        return final_status, pd.DataFrame(results_log)

    except Exception as e:
        return f"Submission error: {e}", pd.DataFrame(results_log)


with gr.Blocks() as demo:
    gr.Markdown("# GAIA Benchmark - Tool Enhanced Agent")

    gr.Markdown(
        """
        1. Login with your Hugging Face account.
        2. Click 'Run Evaluation & Submit All Answers' to run the agent on GAIA tasks.
        3. View your results and submission status.
        """
    )

    login_button = gr.LoginButton()
    run_button = gr.Button("Run Evaluation & Submit All Answers")
    status_box = gr.Textbox(label="Status / Submission Result", lines=7)
    results_table = gr.DataFrame(label="Questions and Agent Answers")

    run_button.click(
        fn=run_and_submit_all,
        inputs=login_button,
        outputs=[status_box, results_table]
    )


if __name__ == "__main__":
    demo.launch()