Build AI agents with LangGraph

Prerequisites

Step-by-Step

1. 1

   Install LangGraph

   Both TS and Python SDKs are first-class. Pick the one your team already runs.

   pip install langgraph langchain-openai

2. 2

   Define your state shape

   State is the single source of truth between nodes. Keep it small and serializable.

   from typing import TypedDict, list
   
   class AgentState(TypedDict):
       question: str
       plan: list[str]
       findings: list[str]
       answer: str

3. 3

   Build nodes for each step

   Each node is a pure function: state in, partial state out.

   from langchain_openai import ChatOpenAI
   llm = ChatOpenAI(model='gpt-4o')
   
   def plan(state):
       msg = llm.invoke(f'Plan steps to answer: {state["question"]}')
       return { 'plan': msg.content.split('\n') }

4. 4

   Wire the graph

   Add nodes, then add edges. Conditional edges let you branch on state.

   from langgraph.graph import StateGraph, END
   
   g = StateGraph(AgentState)
   g.add_node('plan', plan)
   g.set_entry_point('plan')
   g.add_edge('plan', END)
   app = g.compile()

5. 5

   Add checkpointing

   Checkpointers persist state between turns. Use SqliteSaver locally and Postgres in prod.

   from langgraph.checkpoint.sqlite import SqliteSaver
   memory = SqliteSaver.from_conn_string(':memory:')
   app = g.compile(checkpointer=memory)

6. 6

   Stream and inspect

   Stream events to see exactly which node fired. This is huge when debugging.

   for event in app.stream({ 'question': 'compare X and Y' }, config={'configurable': {'thread_id': '1'}}):
       print(event)

Common Pitfalls

What's Next

• ->Add human-in-the-loop interrupts before destructive tools.
• ->Deploy with LangGraph Cloud or your own FastAPI wrapper.

More Build AI agents