app.py

"""Gradio Space app for the HF Agents Course final assignment.

Mirrors the official template's flow:
  1. Hugging Face OAuth login (identifies your submission).
  2. Fetch the evaluation questions from the scoring API.
  3. Run your agent over every question.
  4. Submit answers + a link to this Space's code for scoring.

Set your HF Inference token as the `HF_TOKEN` Space secret so the agent can
call the model.
"""

from __future__ import annotations

import os

import gradio as gr
import pandas as pd
import requests

try:
    from dotenv import load_dotenv

    load_dotenv()  # load HF_TOKEN etc. from a local .env when running locally
except ImportError:
    pass

from agent import GaiaAgent

# --- Constants --------------------------------------------------------------
DEFAULT_API_URL = os.getenv(
    "GAIA_API_URL", "https://agents-course-unit4-scoring.hf.space"
)
QUESTIONS_URL = f"{DEFAULT_API_URL}/questions"
SUBMIT_URL = f"{DEFAULT_API_URL}/submit"


def run_and_submit_all(profile: gr.OAuthProfile | None):
    """Fetch all questions, run the agent on each, and submit the answers.

    Returns a status string and a results dataframe for display.
    """
    # 1. Resolve the logged-in HF username.
    if profile is None:
        return "Please log in to Hugging Face using the button above.", None
    username = profile.username
    print(f"User logged in: {username}")

    # 2. Derive the public code link for this Space (used for verification).
    space_id = os.getenv("SPACE_ID")
    if space_id:
        agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
    else:
        agent_code = "https://huggingface.co/spaces/local/run/tree/main"
    print(f"Agent code link: {agent_code}")

    # 3. Instantiate the agent.
    try:
        agent = GaiaAgent()
    except Exception as exc:  # noqa: BLE001
        return f"Error initializing agent: {exc}", None

    # 4. Fetch the questions.
    try:
        resp = requests.get(QUESTIONS_URL, timeout=30)
        resp.raise_for_status()
        questions = resp.json()
    except Exception as exc:  # noqa: BLE001
        return f"Error fetching questions: {exc}", None

    if not questions:
        return "Fetched questions list is empty.", None
    print(f"Fetched {len(questions)} questions.")

    # 5. Run the agent over every question.
    results_log = []
    answers_payload = []
    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

        print(f"Running task {task_id}...")
        try:
            submitted = agent(question_text, task_id, item.get("file_name"))
        except Exception as exc:  # noqa: BLE001
            submitted = f"AGENT ERROR: {exc}"

        answers_payload.append(
            {"task_id": task_id, "submitted_answer": submitted}
        )
        results_log.append(
            {
                "Task ID": task_id,
                "Question": question_text,
                "Submitted Answer": submitted,
            }
        )

    if not answers_payload:
        return "Agent produced no answers to submit.", pd.DataFrame(results_log)

    # 6. Submit to the scoring API.
    submission = {
        "username": username.strip(),
        "agent_code": agent_code,
        "answers": answers_payload,
    }
    print(f"Submitting {len(answers_payload)} answers for {username}...")
    try:
        resp = requests.post(SUBMIT_URL, json=submission, timeout=120)
        resp.raise_for_status()
        data = resp.json()
    except requests.exceptions.HTTPError as exc:
        detail = ""
        try:
            detail = exc.response.json().get("detail", "")
        except Exception:  # noqa: BLE001
            detail = exc.response.text[:500] if exc.response is not None else ""
        return f"Submission failed: {detail or exc}", pd.DataFrame(results_log)
    except Exception as exc:  # noqa: BLE001
        return f"Submission error: {exc}", pd.DataFrame(results_log)

    status = (
        "Submission Successful!\n"
        f"User: {data.get('username')}\n"
        f"Overall Score: {data.get('score', 'N/A')}% "
        f"({data.get('correct_count', '?')}/"
        f"{data.get('total_attempted', '?')} correct)\n"
        f"Message: {data.get('message', '')}"
    )
    return status, pd.DataFrame(results_log)


# --- UI ---------------------------------------------------------------------
with gr.Blocks(title="GAIA Final Agent") as demo:
    gr.Markdown("# GAIA Final Assignment — Agent Runner")
    gr.Markdown(
        """
        **How to use**
        1. Log in to your Hugging Face account with the button below.
        2. Make sure the `HF_TOKEN` secret is set in this Space (Settings →
           Variables and secrets) so the agent can call the model.
        3. Click **Run Evaluation & Submit All Answers**.

        The agent fetches all questions, answers each one, and submits the
        results for scoring. Running the full set can take several minutes.
        """
    )

    gr.LoginButton()

    run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
    status_output = gr.Textbox(
        label="Run Status / Submission Result", lines=5, interactive=False
    )
    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("-" * 30 + " App Starting " + "-" * 30)
    space_id = os.getenv("SPACE_ID")
    if space_id:
        print(f"Space ID: {space_id}")
        print(f"Code: https://huggingface.co/spaces/{space_id}/tree/main")
    else:
        print("SPACE_ID not set (running locally).")
    demo.launch(debug=True, share=False)