app.py

import os
import gradio as gr
import requests
import pandas as pd
import re
import logging
from agent import initialize_agent # Import the agent initialization function

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

# --- Logging Configuration ---
logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(name)s: %(message)s")
logger = logging.getLogger(__name__)

# --- Global Agent Initialization ---
# The agent is initialized once when the Space starts up.
# This is critical for performance and to avoid reloading the model on every request.
logger.info("🚀 Application starting up! Initializing the GAIA agent...")
AGENT = initialize_agent()
if AGENT is None:
    logger.error("💥 FATAL: Agent initialization failed. The application will not be able to process questions.")
else:
    logger.info("✅ Agent initialized successfully.")

# --- Helper Functions ---

def _fetch_questions(api_url: str) -> list:
    """Fetches evaluation questions from the API."""
    questions_url = f"{api_url}/questions"
    logger.info(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:
            raise ValueError("Fetched questions list is empty or invalid format.")
        logger.info(f"Fetched {len(questions_data)} questions.")
        return questions_data
    except requests.exceptions.RequestException as e:
        raise RuntimeError(f"Error fetching questions: {e}") from e
    except requests.exceptions.JSONDecodeError as e:
        raise RuntimeError(f"Error decoding JSON response from questions endpoint: {e}. Response: {response.text[:500]}") from e
    except Exception as e:
        raise RuntimeError(f"An unexpected error occurred fetching questions: {e}") from e

def _run_agent_on_questions(agent, questions_data: list) -> tuple[list, list]:
    """Runs the agent on each question and collects answers and logs."""
    results_log = []
    answers_payload = []
    logger.info(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:
            logger.warning(f"Skipping item with missing task_id or question: {item}")
            continue
        
        try:
            logger.info(f"Processing task {task_id}: {question_text[:100]}...")
            
            # The agent wrapper returns the final, normalized answer directly.
            submitted_answer = agent(question_text)
            
            logger.info(f"Task {task_id} - Final answer from agent: {submitted_answer}")
            
            answers_payload.append({
                "task_id": task_id, 
                "submitted_answer": submitted_answer
            })
            
            results_log.append({
                "Task ID": task_id, 
                "Question": question_text, 
                "Final Answer": submitted_answer
            })
            
        except Exception as e:
            error_msg = f"AGENT ERROR: {e}"
            logger.error(f"Error running agent on task {task_id}: {e}", exc_info=True)
            
            answers_payload.append({
                "task_id": task_id, 
                "submitted_answer": error_msg
            })
            
            results_log.append({
                "Task ID": task_id, 
                "Question": question_text, 
                "Final Answer": error_msg
            })
    
    return answers_payload, results_log

def _submit_answers(api_url: str, username: str, agent_code_url: str, answers_payload: list) -> dict:
    """Submits the agent's answers to the evaluation API."""
    submit_url = f"{api_url}/submit"
    submission_data = {
        "username": username.strip(), 
        "agent_code": agent_code_url, 
        "answers": answers_payload
    }
    
    logger.info(f"Submitting {len(answers_payload)} answers for user '{username}' to: {submit_url}")
    
    try:
        response = requests.post(submit_url, json=submission_data, timeout=60)
        response.raise_for_status()
        return response.json()
    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]}"
        raise RuntimeError(f"Submission Failed: {error_detail}") from e
    except requests.exceptions.Timeout:
        raise RuntimeError("Submission Failed: The request timed out.") from e
    except requests.exceptions.RequestException as e:
        raise RuntimeError(f"Submission Failed: Network error - {e}") from e
    except Exception as e:
        raise RuntimeError(f"An unexpected error occurred during submission: {e}") from e

# --- Main Gradio Function ---
def run_and_submit_all(profile: gr.OAuthProfile | None):
    """
    Orchestrates the fetching of questions, running the agent, and submitting answers.
    """
    if not profile:
        logger.warning("Attempted to run evaluation without being logged in.")
        return "Please Login to Hugging Face with the button above.", None
    
    username = profile.username
    logger.info(f"User '{username}' initiated evaluation.")

    if AGENT is None:
        return "❌ Error: The agent failed to initialize on startup. Please check the Space logs for details.", None

    space_id = os.getenv("SPACE_ID")
    if not space_id:
        logger.error("SPACE_ID environment variable not found. Cannot determine agent_code URL.")
        return "❌ Error: SPACE_ID not set. This is required for submission.", None
    agent_code_url = f"https://huggingface.co/spaces/{space_id}/tree/main"

    status_message = ""
    results_df = pd.DataFrame()
    results_log = []

    try:
        # 1. Fetch Questions
        questions_data = _fetch_questions(DEFAULT_API_URL)

        # 2. Run Agent on Questions (using the pre-initialized global agent)
        answers_payload, results_log = _run_agent_on_questions(AGENT, questions_data)
        if not answers_payload:
            status_message = "Agent did not produce any answers to submit."
            return status_message, pd.DataFrame(results_log)

        # 3. Submit Answers
        submission_result = _submit_answers(DEFAULT_API_URL, username, agent_code_url, answers_payload)

        final_status = (
            f"🎉 Submission Successful!\n"
            f"👤 User: {submission_result.get('username')}\n"
            f"📊 Overall Score: {submission_result.get('score', 'N/A')}% "
            f"({submission_result.get('correct_count', '?')}/{submission_result.get('total_attempted', '?')} correct)\n"
            f"💬 Message: {submission_result.get('message', 'No message received.')}\n"
            f"🔗 Agent Code: {agent_code_url}"
        )
        status_message = final_status
        results_df = pd.DataFrame(results_log)

    except RuntimeError as e:
        status_message = f"❌ Operation Failed: {e}"
        logger.error(status_message)
        results_df = pd.DataFrame(results_log) if results_log else pd.DataFrame([{"Status": "Error", "Details": str(e)}])
    except Exception as e:
        status_message = f"💥 Critical Error: {e}"
        logger.error(status_message, exc_info=True)
        results_df = pd.DataFrame([{"Status": "Critical Error", "Details": str(e)}])

    return status_message, results_df

# --- Gradio Interface Definition ---
with gr.Blocks(title="GAIA Benchmark Agent", theme=gr.themes.Soft()) as demo:
    gr.Markdown("""
    # 🧠 GAIA Benchmark Evaluation Agent
    
    **An advanced agent designed to tackle the General AI Assistant (GAIA) benchmark.**
    """)
    
    gr.Markdown("""
    ## 📋 Instructions:
    
    1.  **Add Secrets**: If you have cloned this Space, go to the **Settings** tab and add your API keys as **Secrets**.
        *   `TOGETHER_API_KEY`: Your key from Together AI.
        *   `SERPAPI_API_KEY`: Your key from SerpApi for Google Search (optional but recommended).
    
    2.  **Login**: Use the button below to log in with your Hugging Face account. Your username is required for submission.
    
    3.  **Run**: Click 'Run Evaluation & Submit' to start the process. The agent will fetch all questions, solve them, and submit the answers automatically.
    
    4.  **Wait**: The process can take several minutes. You can monitor the progress in the status box and see detailed results in the table below.
    
    ---
    
    ### 🎯 GAIA Answer Formatting
    The agent is designed to automatically format answers according to GAIA's strict requirements (e.g., no commas in numbers, no articles in strings).
    """)

    with gr.Row():
        gr.LoginButton(scale=1)
        run_button = gr.Button("🚀 Run Evaluation & Submit All Answers", variant="primary", scale=2)

    status_output = gr.Textbox(
        label="📊 Evaluation Status & Results", 
        lines=8, 
        interactive=False,
        placeholder="Click 'Run Evaluation' to start the process..."
    )
    
    results_table = gr.DataFrame(
        label="📝 Detailed Question Results", 
        wrap=True,
        interactive=False,
        column_widths=["10%", "60%", "30%"]
    )

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

if __name__ == "__main__":
    print("\n" + "="*70)
    print("🚀 GAIA BENCHMARK AGENT STARTING UP")
    print("="*70)
    
    # Check environment variables loaded from HF Secrets
    space_id = os.getenv("SPACE_ID")
    together_key = os.getenv("TOGETHER_API_KEY")
    serpapi_key = os.getenv("SERPAPI_API_KEY")

    if space_id:
        print(f"✅ SPACE_ID: {space_id}")
        print(f"   - Submission URL will be: https://huggingface.co/spaces/{space_id}")
    else:
        print("⚠️  SPACE_ID not found - submissions will fail. This is normal for local dev.")

    print(f"🔑 API Keys Status (from Secrets):")
    print(f"   - Together AI: {'✅ Set' if together_key else '❌ Missing - Agent will fail to initialize!'}")
    print(f"   - SerpAPI:     {'✅ Set' if serpapi_key else '⚠️ Missing - Google Search tool will be disabled.'}")

    if not together_key:
        print("\n‼️ CRITICAL: TOGETHER_API_KEY is not set in the Space Secrets.")
        print("   Please add it in the 'Settings' tab of your Space.")

    print("="*70)
    print("🎯 Launching Gradio Interface...")
    print("="*70 + "\n")

    demo.launch(debug=False, share=False)