agent.py

import os
import time
from typing import TypedDict, Annotated, Sequence
import operator
from langgraph.graph import StateGraph, END
from langgraph.prebuilt import ToolNode
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_community.tools import DuckDuckGoSearchRun
from langchain_community.document_loaders import WikipediaLoader
from langchain_core.messages import BaseMessage, SystemMessage, AIMessage
from langchain.agents import Tool
from langchain_core.tools import tool

# Load environment variables
google_api_key = os.getenv("GOOGLE_API_KEY") or os.environ.get("GOOGLE_API_KEY")
if not google_api_key:
    raise ValueError("Missing GOOGLE_API_KEY environment variable")

# Load system prompt
with open("System_Prompt.txt", "r", encoding="utf-8") as f:
    system_prompt = f.read()
sys_msg = SystemMessage(content=system_prompt)

# Tool Definitions
@tool
def multiply(a: int, b: int) -> int:
    """Multiply two integers together."""
    return a * b

@tool
def add(a: int, b: int) -> int:
    """Add two integers together."""
    return a + b

@tool
def subtract(a: int, b: int) -> int:
    """Subtract b from a."""
    return a - b

@tool
def divide(a: int, b: int) -> float:
    """Divide a by b. Returns float. Raises error if b is zero."""
    if b == 0:
        raise ValueError("Cannot divide by zero.")
    return a / b

@tool
def wiki_search(query: str) -> str:
    """Search Wikipedia and return up to 2 relevant documents."""
    docs = WikipediaLoader(query=query, load_max_docs=2).load()
    if not docs:
        return "No Wikipedia results found."
    return "\n\n".join([d.page_content[:1000] for d in docs])

# Tool inventory with valid names
tools = [
    Tool(name="Math_Multiply", func=multiply, description="Multiplies two integers"),
    Tool(name="Math_Add", func=add, description="Adds two integers"),
    Tool(name="Math_Subtract", func=subtract, description="Subtracts two integers"),
    Tool(name="Math_Divide", func=divide, description="Divides two numbers"),
    Tool(name="Search_Wikipedia", func=wiki_search, description="Searches Wikipedia"),
    Tool(
        name="Search_Web", 
        func=DuckDuckGoSearchRun().run,
        description="Searches the web using DuckDuckGo"
    )
]

# Graph Definition
class AgentState(TypedDict):
    """State definition for the agent workflow"""
    messages: Annotated[Sequence[BaseMessage], operator.add]

def build_graph():
    """Constructs and compiles the LangGraph workflow"""
    
    # Initialize LLM with Gemini 2.0 Flash and rate limiting
    llm = ChatGoogleGenerativeAI(
        model="gemini-2.0-flash-exp",
        temperature=0.3,
        google_api_key=google_api_key,
        max_retries=3,  # 添加内置重试
        request_timeout=30  # 设置超时
    )
    llm_with_tools = llm.bind_tools(tools)

    # Custom tool condition to avoid '__end__' issues
    def custom_tools_condition(state: AgentState):
        """Determines whether to use tools or end the workflow"""
        last_message = state["messages"][-1]
        
        # Check if it's an AI message with tool calls
        if hasattr(last_message, "tool_calls") and last_message.tool_calls:
            return "use_tools"
        
        # Check if it's an error message
        if isinstance(last_message, AIMessage) and "ERROR" in last_message.content:
            return "end"
            
        # Check if it's a final answer
        if "FINAL ANSWER" in last_message.content:
            return "end"
            
        return "use_tools"

    # Node definitions with error handling
    def agent_node(state: AgentState):
        """Main agent node that processes messages with retry logic"""
        try:
            # Add rate limiting
            time.sleep(1)  # 1 second delay between requests
            
            response = llm_with_tools.invoke(state["messages"])
            return {"messages": [response]}
        
        except Exception as e:
            # Handle specific errors
            error_type = "UNKNOWN"
            if "429" in str(e):
                error_type = "QUOTA_EXCEEDED"
            elif "400" in str(e):
                error_type = "INVALID_REQUEST"
                
            error_msg = f"AGENT ERROR ({error_type}): {str(e)[:200]}"
            return {"messages": [AIMessage(content=error_msg)]}

    # Graph construction
    workflow = StateGraph(AgentState)
    
    # Add nodes to the workflow
    workflow.add_node("agent", agent_node)
    workflow.add_node("tools", ToolNode(tools))
    
    # Configure graph flow with clear endpoints
    workflow.set_entry_point("agent")
    
    # Add conditional edges with custom condition
    workflow.add_conditional_edges(
        "agent",
        custom_tools_condition,
        {
            "use_tools": "tools",
            "end": END  # Directly to END
        }
    )
    
    # Add edge from tools back to agent
    workflow.add_edge("tools", "agent")
    
    # Ensure END has no incoming edges except through condition
    return workflow.compile()

# Initialize the agent graph
agent_graph = build_graph()