LangGraph Quickstart

Benchmarked against: Anthropic — TypeScript V2 (preview) Line: Line 2 (LangGraph Self-Built) Prerequisites: Python 3.11+, .venv with LangGraph installed

Build a LangGraph agent from scratch that connects to SuperPortia's infrastructure.

---

Step 1: Set up environment

# Create virtual environment
python -m venv .venv
source .venv/bin/activate

# Install LangGraph ecosystem
pip install langgraph langgraph-checkpoint-sqlite langchain-mcp-adapters

---

Step 2: Define the state

from langgraph.graph import StateGraph, MessagesState, START, END
from langchain_core.messages import HumanMessage

# MessagesState tracks conversation history
graph_builder = StateGraph(MessagesState)

---

Step 3: Create the agent node

from langchain_google_genai import ChatGoogleGenerativeAI

# Use Gemini for low-cost operations
llm = ChatGoogleGenerativeAI(model="gemini-2.5-flash")

def agent_node(state: MessagesState):
    """The agent decides what to do next."""
    response = llm.invoke(state["messages"])
    return {"messages": [response]}

graph_builder.add_node("agent", agent_node)

---

Step 4: Add tools

from langchain_core.tools import tool

@tool
def search_knowledge(query: str) -> str:
    """Search the Universal Brain for knowledge."""
    # This would connect to Cloud UB via API
    import requests
    response = requests.post(
        "https://your-worker.workers.dev/brain/search",
        json={"query": query}
    )
    return response.json()

# Bind tools to the LLM
tools = [search_knowledge]
llm_with_tools = llm.bind_tools(tools)

---

Step 5: Add routing logic

from langgraph.prebuilt import ToolNode

tool_node = ToolNode(tools)
graph_builder.add_node("tools", tool_node)

def should_use_tools(state: MessagesState):
    """Route to tools if the agent wants to use them."""
    last_message = state["messages"][-1]
    if hasattr(last_message, "tool_calls") and last_message.tool_calls:
        return "tools"
    return END

graph_builder.add_edge(START, "agent")
graph_builder.add_conditional_edges("agent", should_use_tools)
graph_builder.add_edge("tools", "agent")

---

Step 6: Compile and run

# Add memory for state persistence
from langgraph.checkpoint.sqlite import SqliteSaver

memory = SqliteSaver.from_conn_string(":memory:")
app = graph_builder.compile(checkpointer=memory)

# Run the agent
config = {"configurable": {"thread_id": "session-001"}}
result = app.invoke(
    {"messages": [HumanMessage(content="What do we know about fleet architecture?")]},
    config=config
)

print(result["messages"][-1].content)

---

What you built

Component	What it does
StateGraph	Manages conversation state
agent_node	LLM decides next action
ToolNode	Executes tool calls
SqliteSaver	Persists state across turns
Conditional edges	Routes between agent and tools

---

Next steps

Enhancement	How
Add more tools	MCP adapters for Cloud UB, file ops
Multi-agent	Use supervisor.py for orchestration
Better engine	Swap Gemini for Claude for complex tasks
Persistent checkpoints	Use file-based SQLite instead of in-memory

---

Related pages

Page	Relationship
Python SDK	Full Python SDK reference
Agent SDK Overview	Dual lines context
Engine Overview	Available engines for LangGraph