Jens Fischer

LangGraph

State-graph orchestration for agent workflows.

LangGraph is LangChain's orchestration layer. It models an agent as a graph of nodes (functions) connected by edges (transitions), with explicit shared state. It's much less opinionated than the rest of LangChain.

The model

  • Nodes are functions that read and write a shared state object.
  • Edges are either fixed (always go to node B after A) or conditional (a function decides).
  • State is a typed dict passed between nodes. Usually a TypedDict or Pydantic model.
  • Checkpoints persist state between turns. Good for long-running, resumable workflows.

A small graph

from typing import TypedDict
from langgraph.graph import StateGraph, END

class State(TypedDict):
    query: str
    answer: str | None

def search(state: State) -> State:
    return {**state, "answer": search_index(state["query"])}

def route(state: State) -> str:
    return END if state["answer"] else "search"

graph = StateGraph(State)
graph.add_node("search", search)
graph.set_entry_point("search")
graph.add_conditional_edges("search", route)
app = graph.compile()

Where it shines

  • Human-in-the-loop. Pause at a node, wait for an external input, resume.
  • Branching workflows. Conditional edges express "if the model called tool X, route to handler X" naturally.
  • Durable agents. A worker process can crash and resume from the last checkpoint.
  • Cycles. Tool-using agents that loop until done are easier to express as a graph than a while loop with state.

Pushback

  • Overkill for short loops. A 2-tool, 3-step agent doesn't need a graph. A function will do.
  • Debugging. When the graph gets wide, traces in LangSmith become the only way to follow what happened.
  • Vendor pull. Cleanest with LangSmith for traces and LangChain for chat models. You can use it standalone but it costs you.

Worth knowing

  • LangGraph Platform hosts the graphs with built-in queueing, retries, and persistence. Useful for production but adds lock-in.
  • The MessagesState helper handles chat history append patterns. Read its source before relying on it.
  • LangGraph Studio is a desktop visualizer for graphs. Helpful during design.

Reading