Skip to main content

Building a Universal Assistant

This tutorial demonstrates how to create a versatile Universal Assistant using CAP and LangGraph that can handle multiple types of tasks and data sources.

Video Tutorial​

Features​

  • Multi-modal input processing
  • Dynamic tool selection
  • Adaptive response generation
  • Context-aware interactions
  • Multiple data source integration
  • MCP-based orchestration for reliable communication

Implementation​

from cap.agents import UniversalAgent
from cap.tools import ToolRegistry
from cap.memory import ConversationMemory
from cap.mcp import MCPClient

class UniversalAssistant(UniversalAgent):
    def __init__(self):
        super().__init__()
        self.memory = ConversationMemory()
        self.tools = ToolRegistry()
        self.mcp = MCPClient()
        
    async def process_input(self, input_data: dict) -> dict:
        # Determine input type and context
        context = await self.analyze_input(input_data)
        
        # Select appropriate tools
        selected_tools = self.tools.select_for_context(context)
        
        # Process with selected tools via MCP
        results = await self.mcp.process_with_tools(input_data, selected_tools)
        
        # Generate appropriate response
        response = await self.generate_response(results, context)
        
        # Update memory
        self.memory.add_interaction(input_data, response)
        
        return response

Deployment​

module "universal_assistant" {
  source = "./modules/lambda"
  
  name = "universal-assistant"
  runtime = "python3.9"
  handler = "handler.lambda_handler"
  
  environment = {
    OPENAI_API_KEY = var.openai_api_key
    MEMORY_STORE = var.memory_store
    TOOL_CONFIG = var.tool_config
    MCP_SERVER_URL = var.mcp_server_url
  }
}

Key Components​

Input Analysis​

  • Content type detection
  • Context extraction
  • Intent recognition
  • Priority assessment

MCP Integration​

  • Reliable tool communication
  • Load balancing
  • Error handling
  • Performance monitoring

Tool Management​

  • Dynamic tool registration
  • Context-based selection
  • Performance monitoring
  • Error handling

Data Sources​

  • Integration with multiple API data sources
  • Create specialized agents for different data sources
  • Combine multiple data sources for comprehensive insights
  • Automated source selection based on context

Try It Out​

  1. Clone the repository
  2. Follow setup in Getting Started
  3. Configure required API keys
  4. Deploy using provided Terraform configuration
  5. Test with various input types

Learn More​