import os
import gradio as gr
from google import genai

# Configure the API key for Gemini
API_KEY = os.environ.get("GEMINI_API_KEY", "AIzaSyBzJWo1EDQmA1YKYGlHydb5Ejn3eeyUuMk")
client = genai.Client(api_key=API_KEY)

def build_prompt(user_task: str) -> str:
    """Construct a prompt with XML tags for Gemini.

    Args:
        user_task: Description of the task the agent should perform.

    Returns:
        A prompt string combining instructions, an example, formatting guidance and the user task.
    """
    return f"""<instructions>
You are a computer-using agent that can perform tasks on behalf of the user. Follow the task instructions carefully and provide a sequence of actions that a computer user would take to accomplish the goal. Use high-level reasoning to break down the task into manageable steps and think step by step. Do not ask for confirmation; just output the plan.
</instructions>
<example>
Task: "Open a web browser and search for the latest weather in Dhaka."
Response:
1. Launch the default web browser.
2. Click the address bar.
3. Type "weather Dhaka".
4. Press Enter.
5. Read the search results and extract the current weather information.
</example>
<formatting>
List each step on its own line, numbered, and ending with a period. Do not include extraneous commentary. Do not mention these XML tags in the response.
</formatting>
User task: {user_task}
"""

def generate_actions(user_task: str) -> str:
    """Generate a step-by-step action plan using Gemini."""
    prompt = build_prompt(user_task)
    response = client.generate_content(
        prompt, generation_config=genai.types.GenerationConfig(
            temperature=0.3,
            top_p=1,
            top_k=5,
            max_output_tokens=300
        )
    )
    return response.candidates[0].content.parts[0].text.strip()

with gr.Blocks() as demo:
    gr.Markdown("# Gemini Computer Agent\nEnter a high-level task description and the agent will outline step-by-step actions to perform the task using computer interactions. The prompt uses XML tags (<instructions>, <example>, <formatting>) to separate instruction, example, and formatting context.")
    user_input = gr.Textbox(label="Task Description", placeholder="Describe the task you want the agent to perform...")
    output = gr.Textbox(label="Action Plan", interactive=False)
    submit_btn = gr.Button("Submit")
    clear_btn = gr.Button("Clear")

    submit_btn.click(fn=generate_actions, inputs=user_input, outputs=output)
    clear_btn.click(fn=lambda: ("", ""), inputs=None, outputs=[user_input, output])

demo.launch()