#!/usr/bin/env python3
import os
import gradio as gr
from crewai import Agent, Task, Crew, Process, LLM

# ========== LLM WRAPPER SETUP ==========
# Configure your OpenAI API key via environment variable
openai_api_key = os.getenv("OPEN_API_KEY")
if not openai_api_key:
    raise RuntimeError("Please set the OPENAI_API_KEY environment variable.")

# Instantiate the LLM wrapper
llm = LLM(model="openai/gpt-4o", api_key=openai_api_key)

# ========== AGENT DEFINITIONS ==========
parser_agent = Agent(
    role="Parser Agent",
    goal="Understand and structure unstructured user thoughts",
    backstory=(
        "You're great at breaking down user inputs into project insights, "
        "actionable tasks, and mood notes."
    ),
    llm=llm,
    verbose=False
)

memory_agent = Agent(
    role="Memory Checker Agent",
    goal="Identify recurring topics in past logs",
    backstory=(
        "You're responsible for detecting topics mentioned multiple times."
    ),
    llm=llm,
    verbose=False
)

executor_agent = Agent(
    role="Execution Agent",
    goal="Confirm actions and simulate execution based on insights",
    backstory=(
        "You're the executor: propose confirmations and mock-execute tasks."
    ),
    llm=llm,
    verbose=False
)

# ========== BUSINESS LOGIC ==========
def process_thoughts(thought_dump: str) -> str:
    """
    Run the CrewAI workflow: parse, check memory, and execute.
    Returns the final summary string.
    """
    # Define the Task
    main_task = Task(
        description=(
            """You will coordinate the understanding of the following user brain-dump:

            --- START OF USER THOUGHTS ---
            {thought_dump}
            --- END ---

            Step 1: Ask the Parser Agent to extract:
            - Any project-related insights
            - Actionable tasks
            - Health or mood notes

            Step 2: Ask the Memory Agent to check for repeating patterns in the thoughts (like repeated mentioned words)

            Step 3: Based on the parsed output and memory findings, ask the Execution Agent to:
            - Propose a confirmation message to the user
            - If the user agrees (simulate yes), execute the corresponding actions (print logs as mock)

            Finally, return the final actions taken.
            """        
        ),
        expected_output="A summary of actions executed like tasks created, insights logged, and health notes recorded.",
        agent=executor_agent
    )
    # Build and run the Crew
    crew = Crew(
        agents=[parser_agent, memory_agent, executor_agent],
        tasks=[main_task],
        process=Process.sequential,
        verbose=False
    )
    result = crew.kickoff({"thought_dump":thought_dump})
    return result

# ========== Gradio UI ==========
with gr.Blocks(title="ThoughtBridge AI") as demo:
    gr.Markdown("# ThoughtBridge AI: Bridge your mind’s streams into actionable insights")
    inp = gr.Textbox(
        lines=8,
        placeholder="Enter your brain-dump here...",
        label="Your Thoughts"
    )
    out = gr.Textbox(
        lines=10,
        label="Results"
    )
    run_btn = gr.Button("Process")

    run_btn.click(
        fn=process_thoughts,
        inputs=inp,
        outputs=out
    )

# Launch the app
if __name__ == "__main__":
    demo.launch()