app_modal.py

import modal
import os

# Create Modal app
app = modal.App("alita-chat-app")

# Define the image with all required dependencies
image = (
    modal.Image.debian_slim(python_version="3.11")
    .pip_install([
        "gradio>=4.0.0",
        "llama-index-core",
        "llama-index-llms-anthropic",
        "python-dotenv",
        "openai",
        "llama-index",
        "anthropic",
        "requests",
        "dataclasses",
        "beautifulsoup4",
        "duckduckgo-search",
        "llama-index-tools-duckduckgo"
    ])
    # Main script
    .add_local_file("manager_agent2.py", "/app/manager_agent2.py")
    
    # Models
    .add_local_file("models/__init__.py", "/app/models/__init__.py")
    .add_local_file("models/mcp_tool_spec.py", "/app/models/mcp_tool_spec.py")
    .add_local_file("models/mcp_execution_result.py", "/app/models/mcp_execution_result.py")
    .add_local_file("models/task_prompt.py", "/app/models/task_prompt.py")
    
    # Components
    .add_local_file("components/__init__.py", "/app/components/__init__.py")
    .add_local_file("components/mcp_brainstormer.py", "/app/components/mcp_brainstormer.py")
    .add_local_file("components/web_agent.py", "/app/components/web_agent.py")
    .add_local_file("components/script_generator.py", "/app/components/script_generator.py")
    .add_local_file("components/code_runner.py", "/app/components/code_runner.py")
    .add_local_file("components/mcp_registry.py", "/app/components/mcp_registry.py")
)

# Global variables to store initialized components
llm = None
manager_agent = None

@app.function(
    image=image,
    secrets=[modal.Secret.from_name("anthropic")],
    max_containers=10,
    timeout=300,
    min_containers=1,
    cpu=2,
    memory=2048
)
def initialize_components():
    """Initialize LLM and Manager Agent"""
    global llm, manager_agent
    
    import sys
    sys.path.append("/app")
    
    try:
        # Import required modules
        from llama_index.core import Settings
        from llama_index.llms.anthropic import Anthropic
        from models import TaskPrompt
        from manager_agent import ManagerAgent
        
        # Get API key from environment
        api_key = os.environ.get("ANTHROPIC_API_KEY")
        if not api_key:
            raise ValueError("ANTHROPIC_API_KEY not found in environment variables")
        
        # Initialize LLM
        llm = Anthropic(model="claude-3-5-sonnet-20241022", api_key=api_key)
        Settings.llm = llm
        print("Successfully initialized LlamaIndex with Anthropic model")
        
        # Initialize the ManagerAgent
        manager_agent = ManagerAgent(llm)
        print("✅ ManagerAgent initialized successfully")
        
        return True
        
    except Exception as e:
        print(f"Error initializing components: {e}")
        import traceback
        traceback.print_exc()
        return False

@app.function(
    image=image,
    secrets=[modal.Secret.from_name("anthropic-api-key")],
    max_containers=10,
    timeout=60,
    min_containers=1,
    cpu=2,
    memory=2048
)
def process_message(message: str):
    """Process a single message through the ManagerAgent"""
    import sys
    sys.path.append("/app")
    
    try:
        from models import TaskPrompt
        from manager_agent import ManagerAgent
        from llama_index.core import Settings
        from llama_index.llms.anthropic import Anthropic
        
        # Initialize components if needed
        api_key = os.environ.get("ANTHROPIC_API_KEY")
        if not api_key:
            return "❌ ANTHROPIC_API_KEY not found in environment variables"
        
        llm = Anthropic(model="claude-3-5-sonnet-20241022", api_key=api_key)
        Settings.llm = llm
        manager_agent = ManagerAgent(llm)
        
        # Process the message
        task_prompt = TaskPrompt(text=message)
        
        response = manager_agent.run_task(task_prompt)
        
        return response
        
    except Exception as e:
        import traceback
        error_msg = f"❌ Error processing message: {str(e)}\n{traceback.format_exc()}"
        print(error_msg)
        return error_msg

# FIXED: Simple web server approach
@app.function(
    image=image,
    secrets=[modal.Secret.from_name("anthropic-api-key")],
    max_containers=10,
    timeout=300,
    min_containers=1,
    cpu=2,
    memory=2048
)
@modal.web_server(port=7860, startup_timeout=180)
def gradio_app():
    """Simple Gradio app without complex initialization"""
    import gradio as gr
    import asyncio
    
    async def chat_function(message, history):
        """Simple chat function that calls the Modal function"""
        try:
            # Call the Modal function to process the message
            response = process_message.remote(message)
            
            # Stream the response word by word for better UX
            words = response.split()
            partial_response = ""
            
            for i, word in enumerate(words):
                partial_response += word + " "
                if i % 3 == 0 or i == len(words) - 1:
                    yield partial_response.strip()
                    await asyncio.sleep(0.01)
                    
        except Exception as e:
            yield f"❌ Error: {str(e)}"
    
    # Create simple Gradio interface
    interface = gr.ChatInterface(
        fn=chat_function,
        type="messages",
        title="ALITA",
        description="ALITA: the self learning AI",
        examples=[
            "🔍 search for information about AI",
            "🛠️ Analyse this csv file", 
            "⚡ Generate a script to automate a repetitive task",
            "🌐 Find open source resources for machine learning",
        ],
        theme="soft"
    )
    
    # Launch the interface with Modal-compatible settings
    interface.launch(
        server_name="0.0.0.0",  # Must bind to all interfaces for Modal
        server_port=7840,       # Must match the port in @modal.web_server
        share=False,            # Don't create public links
        quiet=True,             # Reduce logging noise
        show_error=True,
        prevent_thread_lock=True  # Important: prevents blocking Modal's event loop
    )

# For local development and testing
if __name__ == "__main__":
    app.deploy("alita-chat-app")