Basic LangGraph Agent

Learn how to trace basic LangGraph agent workflows with Noveum Trace

This guide shows you how to trace basic LangGraph agent workflows using Noveum Trace. You'll learn how to monitor agent decision-making, tool usage, and state management.

🎯 Use Case

Research Assistant Agent: A simple agent that can search for information and provide answers. We'll trace the agent's decision-making process, tool usage, and state transitions.

🚀 Complete Working Example

Here's a complete, working example you can copy and run:

import os
from typing import Annotated, Literal, TypedDict
from dotenv import load_dotenv
import noveum_trace
from noveum_trace import NoveumTraceCallbackHandler
from langchain_core.messages import AIMessage, HumanMessage
from langchain_core.tools import tool
from langchain_openai import ChatOpenAI
from langgraph.graph import END, StateGraph
 
load_dotenv()
 
# Initialize Noveum Trace
noveum_trace.init(
    api_key=os.getenv("NOVEUM_API_KEY"),
    project="customer-support-bot",
    environment="development"
)
 
# Define the agent state
class AgentState(TypedDict):
    messages: Annotated[list, "The messages in the conversation"]
    research_complete: bool
 
# Define tools
@tool
def search_web(query: str) -> str:
    """Search the web for information about a query."""
    # Simulate web search
    return f"Search results for: {query}"
 
@tool
def analyze_information(info: str) -> str:
    """Analyze and summarize information."""
    return f"Analysis: {info} is a comprehensive topic with many aspects."
 
def research_node(state: AgentState):
    """Node that performs research using tools."""
    print("🔍 Researching...")
    
    # Get the last human message
    last_message = state["messages"][-1].content
    
    # Search for information
    search_results = search_web(f"research about {last_message}")
    
    # Analyze the results
    analysis = analyze_information(search_results)
    
    # Add the research results to messages
    state["messages"].append(AIMessage(content=f"Research completed: {analysis}"))
    state["research_complete"] = True
    
    return state
 
def should_continue(state: AgentState) -> Literal["research", "end"]:
    """Decide whether to continue researching or end."""
    if state["research_complete"]:
        return "end"
    return "research"
 
def create_research_agent():
    """Create a basic research agent with tracing."""
    # Initialize callback handler
    callback_handler = NoveumTraceCallbackHandler()
    
    # Create LLM with callback
    llm = ChatOpenAI(
        model="gpt-4",
        temperature=0.7,
        callbacks=[callback_handler]
    )
    
    # Create the graph
    graph = StateGraph(AgentState)
    
    # Add nodes
    graph.add_node("research", research_node)
    
    # Add edges
    graph.add_edge("research", "decision")
    graph.add_conditional_edges(
        "decision",
        should_continue,
        {
            "research": "research",
            "end": END
        }
    )
    
    # Set entry point
    graph.set_entry_point("research")
    
    return graph.compile()
 
def run_research_agent():
    """Run the research agent with tracing."""
    print("=== Basic LangGraph Agent Tracing ===")
    
    # Create the agent
    agent = create_research_agent()
    
    # Run the agent
    result = agent.invoke({
        "messages": [HumanMessage(content="Tell me about artificial intelligence")],
        "research_complete": False
    })
    
    print(f"Final result: {result['messages'][-1].content}")
    return result
 
if __name__ == "__main__":
    run_research_agent()

📋 Prerequisites

pip install noveum-trace langchain-openai langgraph python-dotenv

Set your environment variables:

export NOVEUM_API_KEY="your-noveum-api-key"
export OPENAI_API_KEY="your-openai-api-key"

🔧 How It Works

1. State Management

The AgentState TypedDict defines the state structure:

messages: Conversation history
research_complete: Boolean flag for completion

2. Node Tracing

Each node execution is automatically traced:

Input state
Processing logic
Output state changes
Tool calls and results

3. Conditional Routing

The should_continue function determines the next step:

Traced as a decision point
Shows routing logic in the dashboard

4. Tool Integration

Tools are automatically traced:

Input parameters
Execution time
Output results
Error handling

🎨 Advanced Examples

Multi-Step Agent

def create_multi_step_agent():
    """Create an agent with multiple processing steps."""
    callback_handler = NoveumTraceCallbackHandler()
    llm = ChatOpenAI(callbacks=[callback_handler])
    
    def planning_node(state: AgentState):
        """Plan the research approach."""
        print("📋 Planning research approach...")
        # Planning logic here
        return state
    
    def execution_node(state: AgentState):
        """Execute the research plan."""
        print("⚡ Executing research...")
        # Execution logic here
        return state
    
    def review_node(state: AgentState):
        """Review and finalize results."""
        print("📝 Reviewing results...")
        # Review logic here
        return state
    
    graph = StateGraph(AgentState)
    graph.add_node("planning", planning_node)
    graph.add_node("execution", execution_node)
    graph.add_node("review", review_node)
    
    # Linear flow
    graph.add_edge("planning", "execution")
    graph.add_edge("execution", "review")
    graph.add_edge("review", END)
    
    graph.set_entry_point("planning")
    return graph.compile()

Agent with LLM Integration

def create_llm_agent():
    """Create an agent that uses LLM for decision making."""
    callback_handler = NoveumTraceCallbackHandler()
    llm = ChatOpenAI(callbacks=[callback_handler])
    
    def llm_decision_node(state: AgentState):
        """Use LLM to make decisions."""
        print("🤖 LLM making decision...")
        
        # Use LLM to decide next action
        response = llm.invoke([
            HumanMessage(content=f"Based on this context: {state['messages'][-1].content}, what should I do next?")
        ])
        
        # Add LLM response to state
        state["messages"].append(response)
        return state
    
    graph = StateGraph(AgentState)
    graph.add_node("llm_decision", llm_decision_node)
    graph.add_edge("llm_decision", END)
    graph.set_entry_point("llm_decision")
    
    return graph.compile()