app.py

import os
import io
import sys
import re
import traceback
import subprocess
import gradio as gr
import pandas as pd
from dotenv import load_dotenv
from crewai import Crew, Agent, Task, Process, LLM
from crewai_tools import FileReadTool
from pydantic import BaseModel, Field

# Load environment variables
load_dotenv()

# Get API key from environment variables
OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
if not OPENAI_API_KEY:
    raise ValueError("OPENAI_API_KEY environment variable not set")

llm = LLM(
    model="openai/gpt-4o",
    api_key=OPENAI_API_KEY,
    temperature=0.7
)

# 1) Query parser agent
query_parser_agent = Agent(
    role="Stock Data Analyst",
    goal="Extract stock details and fetch required data from this user query: {query}.",
    backstory="You are a financial analyst specializing in stock market data retrieval.",
    llm=llm,
    verbose=True,
    memory=True,
)

# Need to define QueryAnalysisOutput class here as it's used by the task
class QueryAnalysisOutput(BaseModel):
    """Structured output for the query analysis task."""
    symbols: list[str] = Field(
        ...,
        json_schema_extra={"description": "List of stock ticker symbols (e.g., ['TSLA', 'AAPL'])."}
    )
    timeframe: str = Field(
        ...,
        json_schema_extra={"description": "Time period (e.g., '1d', '1mo', '1y')."}
    )
    action: str = Field(
        ...,
        json_schema_extra={"description": "Action to be performed (e.g., 'fetch', 'plot')."}
    )


query_parsing_task = Task(
    description="Analyze the user query and extract stock details.",
    expected_output="A dictionary with keys: 'symbol', 'timeframe', 'action'.",
    output_pydantic=QueryAnalysisOutput,
    agent=query_parser_agent,
)

# 2) Code writer agent
code_writer_agent = Agent(
    role="Senior Python Developer",
    goal="Write Python code to visualize stock data.",
    backstory="""You are a Senior Python developer specializing in stock market data visualization.
                 You are also a Pandas, Matplotlib and yfinance library expert.
                 You are skilled at writing production-ready Python code.
                 Ensure the code handles potential variations in the DataFrame structure returned by yfinance,
                 especially for different timeframes or delisted stocks.
                 Crucially, ensure the generated script saves any generated plot as 'plot.png' using `plt.savefig('plot.png')` before the script ends.""",
    llm=llm,
    verbose=True,
)

code_writer_task = Task(
    description="""Write Python code to visualize stock data based on the inputs from the stock analyst
                   where you would find stock symbol, timeframe and action.""",
    expected_output="A clean and executable Python script file (.py) for stock visualization.",
    agent=code_writer_agent,
)

# 3) Code output agent (instead of execution agent)
code_output_agent = Agent(
    role="Python Code Presenter",
    goal="Present the generated Python code for stock visualization.",
    backstory="You are an expert in presenting Python code in a clear and readable format.",
    allow_delegation=False,  # This agent just presents the code
    llm=llm,
    verbose=True,
)

code_output_task = Task(
    description="""Receive the Python code for stock visualization from the code writer agent and present it.""",
    expected_output="The complete Python script for stock visualization.",
    agent=code_output_agent,
)

crew = Crew(
    agents=[query_parser_agent, code_writer_agent, code_output_agent],  # Use code_output_agent
    tasks=[query_parsing_task, code_writer_task, code_output_task],  # Use code_output_task
    process=Process.sequential
)


def run_crewai_process(user_query, model, temperature):
    """
    Runs the CrewAI process, captures agent thoughts, gets generated code,
    executes the code, and returns results, including plot.

    Args:
        user_query (str): The user's query for the CrewAI process.
        model (str): The model to use for the LLM.
        temperature (float): The temperature to use for the LLM.

    Yields:
        tuple: A tuple containing the agent thoughts (str), the final answer (list of dicts),
               the generated code (str), the execution output (str), and plot file path (str or None).
    """
    # Create a string buffer to capture stdout
    output_buffer = io.StringIO()
    original_stdout = sys.stdout
    sys.stdout = output_buffer
    agent_thoughts = ""
    generated_code = ""
    execution_output = ""
    generated_plot_path = None
    final_answer_chat = [{"role": "user", "content": user_query}]

    try:
        # Initial status update with proper message format
        initial_message = {"role": "assistant", "content": "Starting CrewAI process..."}
        final_answer_chat = [{"role": "user", "content": str(user_query)}, initial_message]
        yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None

        # Run the crew process
        final_result = crew.kickoff(inputs={"query": user_query})

        # Get the captured CrewAI output (agent thoughts)
        agent_thoughts = output_buffer.getvalue()
        
        # Update with processing message
        processing_message = {"role": "assistant", "content": "Processing complete. Generating code..."}
        final_answer_chat = [{"role": "user", "content": str(user_query)}, processing_message]
        yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None

        # The final result is the generated code from the code_output_agent
        generated_code_raw = str(final_result).strip()

        # Use regex to extract the code block
        code_match = re.search(r"```python\n(.*?)\n```", generated_code_raw, re.DOTALL)
        if code_match:
            generated_code = code_match.group(1).strip()
        else:
            # If no code block is found, assume the entire output is code (or handle as error)
            generated_code = generated_code_raw
            if not generated_code.strip():  # Handle cases where output is empty or just whitespace
                execution_output = "CrewAI process completed, but no code was generated."
                final_answer_chat.append({"role": "assistant", "content": execution_output})
                yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path
                return  # Exit the generator

        # Format for Gradio Chatbot (list of dictionaries with 'role' and 'content' keys only)
        code_gen_message = {"role": "assistant", "content": "Code generation complete. See the 'Generated Code' box. Attempting to execute code..."}
        final_answer_chat = [{"role": "user", "content": str(user_query)}, code_gen_message]
        yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None

        # --- Execute the generated code ---
        plot_file_path = 'plot.png'  # Expected plot file name

        if generated_code:
            try:
                # Write the generated code to a temporary file
                temp_script_path = "generated_script.py"
                with open(temp_script_path, "w") as f:
                    f.write(generated_code)

                # Create a debug script that will be executed in a subprocess
                debug_script = """
import os
import sys
import traceback

try:
    print("\n" + "="*80)
    print("STOCK PLOT GENERATION")
    print("="*80)
    
    # Import the stock plot module
    try:
        import stock_plot
        
        # Generate the plot using the module
        plot_path = stock_plot.plot_stock_gain(["META"], "ytd")
        
        print("\n" + "="*80)
        print("PLOT GENERATION COMPLETE")
        print("="*80)
        
        if plot_path and os.path.exists(plot_path):
            file_size = os.path.getsize(plot_path)
            print(f"✓ Plot generated successfully: {os.path.abspath(plot_path)}")
            print(f"✓ File size: {file_size} bytes")
            
            # Also check for plot.png in the root directory
            if os.path.exists('plot.png'):
                print(f"✓ Main plot.png found: {os.path.abspath('plot.png')}")
            else:
                print("ℹ️ plot.png not found in root directory")
        else:
            print("❌ Failed to generate plot or plot file not found")
            
    except ImportError as e:
        print(f"❌ Error importing stock_plot module: {e}")
        print("Make sure the stock_plot.py file exists in the same directory.")
        raise
        
    print("\n" + "="*80)
    print("DIRECTORY CONTENTS")
    print("="*80)
    
    # List all files in the current directory
    for f in sorted(os.listdir('.')):
        try:
            fpath = os.path.join('.', f)
            if os.path.isfile(fpath):
                size = os.path.getsize(fpath)
                print(f"  - {f} ({size} bytes)")
            else:
                print(f"  - {f}/ (dir)")
        except Exception as e:
            print(f"  - {f} (error: {e})")
            
    # Check for generated_plots directory
    plots_dir = 'generated_plots'
    if os.path.exists(plots_dir) and os.path.isdir(plots_dir):
        print(f"\nContents of {plots_dir}/:")
        try:
            for f in sorted(os.listdir(plots_dir)):
                try:
                    fpath = os.path.join(plots_dir, f)
                    if os.path.isfile(fpath):
                        size = os.path.getsize(fpath)
                        print(f"  - {f} ({size} bytes)")
                except Exception as e:
                    print(f"  - {f} (error: {e})")
        except Exception as e:
            print(f"  Error reading {plots_dir}: {e}")
    
except Exception as e:
    print(f"\n❌ UNEXPECTED ERROR: {str(e)}")
    print("\nTraceback:")
    traceback.print_exc()
    sys.exit(1)
"""
    
    # Create a simple test plot
    try:
        # Create a new figure with a larger size
        fig, ax = plt.subplots(figsize=(10, 6))
        
        # Generate some sample data
        x = [1, 2, 3, 4, 5]
        y = [1, 4, 9, 16, 25]
        
        # Create the plot
        ax.plot(x, y, 'b-', linewidth=2, label='Sample Data')
        
        # Add labels and title
        ax.set_title('Test Plot - Matplotlib Verification', fontsize=14)
        ax.set_xlabel('X Axis', fontsize=12)
        ax.set_ylabel('Y Axis', fontsize=12)
        
        # Add grid and legend
        ax.grid(True, linestyle='--', alpha=0.7)
        ax.legend(fontsize=10)
        
        # Adjust layout to prevent label cutoff
        plt.tight_layout()
        
        # Save the test plot
        test_plot_path = 'test_plot.png'
        fig.savefig(test_plot_path, dpi=120, bbox_inches='tight')
        print(f"✓ Test plot saved to: {os.path.abspath(test_plot_path)}")
        
        # Close the figure to free memory
        plt.close(fig)
        
    except Exception as e:
        print(f"❌ Error creating test plot: {e}")
        print("Traceback:")
        traceback.print_exc()
    
    # Save with different formats and verify
    test_formats = [
        ('test_plot.png', 'PNG'),
        ('test_plot.jpg', 'JPEG'),
        ('test_plot.pdf', 'PDF')
    ]
    
    for filename, fmt in test_formats:
        try:
            test_fig.savefig(filename, bbox_inches='tight', dpi=100)
            abs_path = os.path.abspath(filename)
            file_size = os.path.getsize(filename)
            print(f"✓ Saved {{fmt}} to: {{abs_path}} ({{file_size}} bytes)")
            
            # Verify file is not empty
            if file_size == 0:
                print(f"  ✗ WARNING: {{fmt}} file is empty!")
            elif file_size < 1024:  # Very small file might be corrupted
                print(f"  ⚠ WARNING: {{fmt}} file is unusually small ({{file_size}} bytes)")
                
        except Exception as e:
            print(f"✗ Failed to save {{fmt}}: {{str(e)}}")
    
    # Clean up
    plt.close(test_fig)
    
    # Test 2: Verify file system access
    print("\n" + "="*80)
    print("TEST 2: FILE SYSTEM VERIFICATION")
    print("="*80)
    
    # Check if files were created
    for filename, _ in test_formats:
        if os.path.exists(filename):
            try:
                file_size = os.path.getsize(filename)
                print(f"✓ Found {{filename}} ({{file_size}} bytes)")
                # Try to read the file
                with open(filename, 'rb') as f:
                    header = f.read(4)
                    print(f"  File header: {{header[:20].hex()}}...")
            except Exception as e:
                print(f"✗ Error reading {{filename}}: {{str(e)}}")
        else:
            print(f"✗ File not found: {{filename}}")
    
    # Execute the original script with error handling
    print("\n" + "="*80)
    print("EXECUTING USER SCRIPT")
    print("="*80)
    
    # Create a safe execution environment
    try:
        # Execute the user's code
        exec_globals = {{'plt': plt, 'pd': __import__('pandas')}}
        exec_globals.update({{'__builtins__': __builtins__}})
        
        # Execute in a separate namespace to avoid polluting globals
        user_namespace = {{}}
        user_code = compile(
            {generated_code!r}.lstrip('\n').lstrip(' '),
            '<user_code>', 'exec',
            dont_inherit=True,
            optimize=2
        )
        exec(user_code, user_namespace, user_namespace)
        
        # Ensure any pending plots are drawn
        plt.ioff()
        
        # Save any open figures
        print("\n" + "="*80)
        print("SAVING PLOTS")
        print("="*80)
        
        # Get list of all figure numbers
        fig_nums = plt.get_fignums()
        print(f"Found {{len(fig_nums)}} open figures")
        
        if fig_nums:
            # Create plots directory if it doesn't exist
            plots_dir = 'generated_plots'
            os.makedirs(plots_dir, exist_ok=True)
            
            # Save all figures with unique names
            saved_plots = []
            for i, num in enumerate(fig_nums, 1):
                try:
                    fig = plt.figure(num)
                    plot_name = f'plot_{{i}}.png'
                    plot_path = os.path.abspath(os.path.join(plots_dir, plot_name))
                    
                    # Save with high DPI and tight layout
                    fig.savefig(
                        plot_path,
                        bbox_inches='tight',
                        dpi=150,
                        facecolor=fig.get_facecolor(),
                        edgecolor='none',
                        transparent=False
                    )
                    
                    file_size = os.path.getsize(plot_path)
                    print(f"✓ Saved plot {{i}} to: {{plot_path}} ({{file_size}} bytes)")
                    saved_plots.append(plot_path)
                    
                    # If this is the last figure, also save as plot.png in root
                    if i == len(fig_nums):
                        root_plot_path = os.path.abspath('plot.png')
                        fig.savefig(root_plot_path, bbox_inches='tight', dpi=150)
                        print(f"✓ Saved main plot to: {{root_plot_path}}")
                        saved_plots.append(root_plot_path)
                    
                except Exception as e:
                    print(f"✗ Error saving figure {{num}}: {{str(e)}}")
            
            if saved_plots:
                generated_plot_path = saved_plots[-1]  # Use the last saved plot as the main plot
                print(f"\nSuccessfully saved {{len(saved_plots)}} plot(s)")
            
        else:
            print("ℹ️ No figures were created by the user script")
            
    except Exception as e:
        print(f"\n❌ Error executing user script: {{str(e)}}")
        print("\nTraceback:")
        traceback.print_exc()
        raise
    
    # Print final directory contents
    print("\n" + "="*80)
    print("FINAL DIRECTORY CONTENTS")
    print("="*80)
    for f in os.listdir('.'):
        fpath = os.path.join('.', f)
        if os.path.isfile(fpath):
            print(f"  - {{f}} ({{os.path.getsize(fpath)}} bytes)")
        else:
            print(f"  - {{f}}/ (dir)")
    
    print("\n" + "="*80)
    print("SCRIPT EXECUTION COMPLETE")
    print("="*80)
    
except Exception as e:
    print("\n" + "!"*80)
    print("ERROR DURING EXECUTION")
    print("!"*80)
    print(f"Error type: {{type(e).__name__}}")
    print(f"Error message: {{str(e)}}")
    print("\nTraceback:")
    traceback.print_exc()
    print("\n" + "!"*80 + "\n")
    raise

finally:
    # Always close all figures to free memory
    plt.close('all')
"""
                    # Write the debug script to a temporary file
                    with open(temp_script_path, "w") as f:
                        f.write(debug_script)

                    # Execute the script
                    process = subprocess.run(
                        ["python3", temp_script_path],
                        capture_output=True,
                        text=True,
                        check=False
                    )
                    
                    # Capture both stdout and stderr
                    execution_output = process.stdout
                    if process.stderr:
                        execution_output += "\n\n[ERROR] Script execution errors:\n" + process.stderr
                    
                    # Check for common issues in the output
                    if "KeyError" in execution_output:
                        execution_output += "\n\n[HELP] The script encountered a KeyError. This typically happens when trying to access a column that doesn't exist in the stock data.\n"
                        execution_output += "Common causes:\n"
                        execution_output += "1. The stock symbol might not be recognized by yfinance\n"
                        execution_output += "2. The requested time period might not have data (e.g., weekends, holidays)\n"
                        execution_output += "3. The data column names might be different than expected\n\n"
                        execution_output += "Please try a different stock symbol or time period."
                        
                    if "No data" in execution_output or "not found" in execution_output.lower():
                        execution_output += "\n\n[HELP] No data was returned for the specified stock symbol or time period.\n"
                        execution_output += "Please check the stock symbol and try a different time period."
                        
                    if "Figure(" in execution_output and "plot.png" not in os.listdir():
                        execution_output += "\n\n[HELP] A plot was created but not saved. Adding save command...\n"
                        # Try to save the plot if it wasn't saved
                        try:
                            import matplotlib.pyplot as plt
                            if plt.get_fignums():
                                plt.savefig('plot.png')
                                execution_output += "Successfully saved plot to plot.png"
                                generated_plot_path = 'plot.png'
                            plt.close('all')
                        except Exception as e:
                            execution_output += f"Failed to save plot: {str(e)}"
                except Exception as e:
                    execution_output = f"Error during script execution: {str(e)}\n\n"
                    execution_output += "Please check the generated code for issues or try a different query."

                # Enhanced plot file checking with more detailed debugging
                plot_debug_info = []
                plot_found = False
                
                # Check in current directory first
                current_dir = os.getcwd()
                plot_abs_path = os.path.abspath(plot_file_path)
                
                # Log directory contents for debugging
                plot_debug_info.append(f"Current directory: {current_dir}")
                plot_debug_info.append("Directory contents:" + "\n- " + "\n- ".join(os.listdir('.')))
                
                # Check if plot exists in current directory
                if os.path.exists(plot_file_path):
                    plot_found = True
                    plot_debug_info.append(f"✅ Plot file found at: {plot_abs_path}")
                    generated_plot_path = plot_file_path
                else:
                    # Try to find the plot file in subdirectories
                    for root, _, files in os.walk('.'):
                        if plot_file_path in files:
                            found_path = os.path.join(root, plot_file_path)
                            plot_found = True
                            plot_debug_info.append(f"✅ Plot file found at: {os.path.abspath(found_path)}")
                            generated_plot_path = found_path
                            break
                    
                    if not plot_found:
                        plot_debug_info.append(f"❌ Plot file not found at: {plot_abs_path}")
                        plot_debug_info.append("Troubleshooting tips:")
                        plot_debug_info.append("1. Ensure the script calls plt.savefig('plot.png')")
                        plot_debug_info.append("2. Check for any errors in the execution output")
                        plot_debug_info.append("3. Verify the script has write permissions in the current directory")
                
                # Add debug info to execution output
                execution_output += "\n\n[PLOT DEBUG] " + "\n[PLOT DEBUG] ".join(plot_debug_info)
                
                if not plot_found:
                    execution_output += f"\n\n[ERROR] Plot file '{plot_file_path}' was not generated. Check the debug information above for details."

            except Exception as e:
                traceback_str = traceback.format_exc()
                execution_output = f"An error occurred during code execution: {e}\n{traceback_str}"

            finally:
                # Clean up the temporary script file
                if os.path.exists(temp_script_path):
                    os.remove(temp_script_path)

        else:
            execution_output = "No code was generated to execute."

        # Update final answer chat to reflect execution attempt
        execution_complete_msg = "Code execution finished. See 'Execution Output'."
        if generated_plot_path:
            plot_msg = "Plot generated successfully. See 'Generated Plot'."
            final_answer_chat = [
                {"role": "user", "content": str(user_query)},
                {"role": "assistant", "content": execution_complete_msg},
                {"role": "assistant", "content": plot_msg}
            ]
        else:
            no_plot_msg = "No plot was generated. Check the execution output for details."
            final_answer_chat = [
                {"role": "user", "content": str(user_query)},
                {"role": "assistant", "content": execution_complete_msg},
                {"role": "assistant", "content": no_plot_msg}
            ]

        yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path

    except Exception as e:
        # If an error occurs during CrewAI process, return the error message
        traceback_str = traceback.format_exc()
        agent_thoughts += f"\nAn error occurred during CrewAI process: {e}\n{traceback_str}"
        error_message = f"An error occurred during CrewAI process: {e}"
        final_answer_chat = [
            {"role": "user", "content": str(user_query)},
            {"role": "assistant", "content": error_message}
        ]
        yield final_answer_chat, agent_thoughts, generated_code, execution_output, None, None

    finally:
        # Restore original stdout
        sys.stdout = original_stdout


def create_interface():
    """Create and return the Gradio interface."""
    with gr.Blocks(title="Financial Analytics Agent", theme=gr.themes.Soft()) as interface:
        gr.Markdown("# 📊 Financial Analytics Agent")
        gr.Markdown("Enter your financial query to analyze stock data and generate visualizations.")
        
        with gr.Row():
            with gr.Column(scale=2):
                user_query_input = gr.Textbox(
                    label="Enter your financial query",
                    placeholder="e.g., Show me the stock performance of AAPL and MSFT for the last year",
                    lines=3
                )
                submit_btn = gr.Button("Analyze", variant="primary")
                
                with gr.Accordion("Advanced Options", open=False):
                    gr.Markdown("### Model Settings")
                    model_dropdown = gr.Dropdown(
                        ["gpt-4o", "gpt-4-turbo", "gpt-3.5-turbo"],
                        value="gpt-4o",
                        label="Model"
                    )
                    temperature = gr.Slider(
                        minimum=0.1,
                        maximum=1.0,
                        value=0.7,
                        step=0.1,
                        label="Creativity (Temperature)"
                    )
            
            with gr.Column(scale=3):
                with gr.Tabs():
                    with gr.TabItem("Analysis"):
                        final_answer_chat = gr.Chatbot(
                            label="Analysis Results",
                            height=300,
                            show_copy_button=True,
                            type="messages"  # Explicitly set to use OpenAI-style message format
                        )
                    
                    with gr.TabItem("Agent Thoughts"):
                        agent_thoughts = gr.Textbox(
                            label="Agent Thinking Process",
                            interactive=False,
                            lines=15,
                            max_lines=30,
                            show_copy_button=True
                        )
                    
                    with gr.TabItem("Generated Code"):
                        generated_code = gr.Code(
                            label="Generated Python Code",
                            language="python",
                            interactive=False,
                            lines=15
                        )
                    
                    with gr.TabItem("Execution Output"):
                        execution_output = gr.Textbox(
                            label="Code Execution Output",
                            interactive=False,
                            lines=10,
                            show_copy_button=True
                        )
                
                with gr.Row():
                    with gr.Column():
                        plot_output = gr.Plot(
                            label="Generated Visualization",
                            visible=False
                        )
                        image_output = gr.Image(
                            label="Generated Plot",
                            type="filepath",
                            visible=False
                        )
        
        # Handle form submission
        inputs = [user_query_input, model_dropdown, temperature]
        outputs = [
            final_answer_chat,
            agent_thoughts,
            generated_code,
            execution_output,
            plot_output,
            image_output
        ]
        
        submit_btn.click(
            fn=run_crewai_process,
            inputs=inputs,
            outputs=outputs,
            api_name="analyze"
        )
    
    return interface


def main():
    """Run the Gradio interface."""
    interface = create_interface()
    interface.launch(share=False, server_name="0.0.0.0", server_port=7860)


if __name__ == "__main__":
    main()