LangGraph & Cyclic State Machines

While standard LangChain chains follow a Directed Acyclic Graph (DAG) flow, real-world autonomous agents require loops, conditional branching, state mutation, and self-correction.

LangGraph Architecture

LangGraph models agent workflows as stateful graphs where Nodes represent python functions (LLMs or Tools) and Edges determine state transitions based on conditional logic.

---

LangGraph Core Components

1. State Schema: A TypedDict or Pydantic model representing the shared memory passed between nodes.
2. Nodes: Python functions that receive the current state, perform computation, and return state updates.
3. Edges: Connections routing execution from one node to another.
4. Conditional Edges: Dynamic functions inspecting state to determine the next destination node (e.g., checking if a tool call was requested).

from typing import Annotated, TypedDict
from langgraph.graph import StateGraph, START, END
from langgraph.graph.message import add_messages

# 1. Define State Schema
class AgentState(TypedDict):
    messages: Annotated[list, add_messages]

# 2. Initialize State Graph
builder = StateGraph(AgentState)

# 3. Define Node Functions
def chatbot_node(state: AgentState):
    return {"messages": [model.invoke(state["messages"])]}

def tool_execution_node(state: AgentState):
    # Execute requested tools
    return {"messages": [...]}

# 4. Add Nodes & Edges
builder.add_node("chatbot", chatbot_node)
builder.add_node("tools", tool_execution_node)

builder.add_edge(START, "chatbot")

# Conditional routing edge
def should_continue(state: AgentState):
    last_message = state["messages"][-1]
    if last_message.tool_calls:
        return "tools"
    return END

builder.add_conditional_edges("chatbot", should_continue)
builder.add_edge("tools", "chatbot") # Loop back after executing tools!

# Compile Graph
graph = builder.compile()

---

Agent Loop Execution Flow

[ START ] ──► [ Chatbot Node ] ──(Has Tool Call?)──► [ Tool Node ] ──► [ Chatbot Node ]
                     │
              (No Tool Call)
                     │
                     ▼
                  [ END ]