core/agents/langgraph_agent.py

from typing import Dict, Any, List, Optional
from .base import BaseAgent, TaskInput, AgentTool
from langgraph.graph import Graph, StateGraph
from langgraph.prebuilt import ToolExecutor
import operator
from datetime import datetime
import logging

class LangGraphAgent(BaseAgent):
    """LangGraph-based agent implementation"""
    
    async def initialize(self) -> None:
        """Initialize LangGraph agent"""
        await super().initialize()
        
        try:
            # Initialize tool executor
            self.tool_executor = ToolExecutor(
                tools=[
                    self._convert_tool_to_langgraph(tool)
                    for tool in self.agent_config.tools
                ]
            )
            
            # Create agent graph
            self.graph = self._create_agent_graph()
            
        except Exception as e:
            self.logger.error(f"Failed to initialize LangGraph agent: {str(e)}")
            raise
    
    def _convert_tool_to_langgraph(self, tool: AgentTool) -> Dict[str, Any]:
        """Convert Lattice tool to LangGraph format"""
        return {
            "name": tool.name,
            "description": tool.description,
            "func": tool.function,
            "args_schema": {}  # Add appropriate schema
        }
    
    def _create_agent_graph(self) -> StateGraph:
        """Create agent workflow graph"""
        workflow = StateGraph()
        
        # Add nodes
        workflow.add_node("agent", self._agent_node)
        workflow.add_node("tool_executor", self.tool_executor)
        
        # Add edges
        workflow.add_edge("agent", "tool_executor")
        workflow.add_edge("tool_executor", "agent")
        
        # Set entry point
        workflow.set_entry_point("agent")
        
        return workflow.compile()
    
    async def _agent_node(self, state: Dict[str, Any]) -> Dict[str, Any]:
        """Agent node implementation"""
        # Process state and decide next action
        message = state.get("message", "")
        history = state.get("history", [])
        
        # Generate response using LLM
        response = await self._generate_response(message, history)
        
        # Determine if tool execution is needed
        if self._needs_tool(response):
            tool_call = self._extract_tool_call(response)
            return {
                "type": "tool_call",
                "tool": tool_call["name"],
                "args": tool_call["args"]
            }
        
        return {
            "type": "final",
            "response": response
        }
    
    async def _execute_implementation(self, task: TaskInput) -> Dict[str, Any]:
        """Execute task using LangGraph"""
        try:
            # Initialize state
            initial_state = {
                "message": task.description,
                "history": [],
                "context": task.context or {},
                "tools": task.tools or []
            }
            
            # Execute graph
            result = await self.graph.arun(initial_state)
            
            return {
                'output': result.get("response", ""),
                'steps': result.get("history", []),
                'metadata': {
                    'framework': 'langgraph',
                    'tools_used': result.get("tools_used", [])
                }
            }
            
        except Exception as e:
            self.logger.error(f"LangGraph task execution failed: {str(e)}")
            raise

class LangGraphWorkflow:
    """LangGraph workflow implementation"""
    
    def __init__(self, config: Dict[str, Any]):
        self.config = config
        self.logger = logging.getLogger("lattice.agent.langgraph.workflow")
    
    def _create_workflow_graph(
        self,
        agents: List[LangGraphAgent],
        tasks: List[TaskInput]
    ) -> StateGraph:
        """Create workflow graph connecting multiple agents"""
        workflow = StateGraph()
        
        # Add agent nodes
        for i, agent in enumerate(agents):
            workflow.add_node(f"agent_{i}", agent._agent_node)
            workflow.add_node(f"tools_{i}", agent.tool_executor)
        
        # Add task routing node
        workflow.add_node("router", self._create_router(tasks))
        
        # Connect nodes based on task dependencies
        for i in range(len(agents)):
            workflow.add_edge(f"agent_{i}", f"tools_{i}")
            workflow.add_edge(f"tools_{i}", "router")
            workflow.add_edge("router", f"agent_{i}")
        
        # Set entry point
        workflow.set_entry_point("router")
        
        return workflow.compile()
    
    def _create_router(self, tasks: List[TaskInput]):
        """Create routing logic based on tasks"""
        def router(state: Dict[str, Any]) -> str:
            current_task = state.get("current_task", 0)
            if current_task >= len(tasks):
                return "end"
                
            # Route to appropriate agent based on task
            return f"agent_{current_task}"
        
        return router
    
    async def execute_workflow(
        self,
        agents: List[LangGraphAgent],
        tasks: List[TaskInput]
    ) -> Dict[str, Any]:
        """Execute a workflow with multiple agents"""
        try:
            # Create workflow graph
            graph = self._create_workflow_graph(agents, tasks)
            
            # Initialize state
            initial_state = {
                "current_task": 0,
                "tasks": [task.dict() for task in tasks],
                "results": [],
                "history": []
            }
            
            # Execute workflow
            result = await graph.arun(initial_state)
            
            return {
                'results': result.get("results", []),
                'history': result.get("history", []),
                'metadata': {
                    'framework': 'langgraph',
                    'agent_count': len(agents),
                    'task_count': len(tasks)
                },
                'timestamp': datetime.now().isoformat()
            }
            
        except Exception as e:
            self.logger.error(f"LangGraph workflow execution failed: {str(e)}")
            raise

class LangGraphWorkflow