LangGraph

1.0.x · active · verified Sat Feb 28

LangGraph is a low-level agent orchestration framework providing graph-based execution with durable state, built-in persistence, and human-in-the-loop patterns. As of v1.0 it is the foundational runtime for LangChain agents. Use LangChain for high-level agent abstractions, LangGraph for fine-grained control.

Warnings

deprecated langgraph.prebuilt module deprecated in 1.0. All prebuilt functionality moved to langchain.agents.
Fix: Replace from langgraph.prebuilt import create_react_agent with from langchain import create_agent
breaking langgraph-prebuilt==1.0.2 introduced a breaking signature change to ToolNode.afunc adding a required runtime parameter. langgraph==1.0.1 does not pin this dependency.
Fix: Pin langgraph-prebuilt==1.0.1 explicitly or upgrade to latest langgraph
gotcha MemorySaver is in-memory only. State is lost on process restart. Not for production.
Fix: Use langgraph-checkpoint-postgres or langgraph-checkpoint-sqlite for persistence
gotcha LangGraph requires a compiled graph before invocation. Always call graph.compile() before graph.invoke().
Fix: app = graph.compile() then app.invoke(...)
gotcha State TypedDict must be defined before StateGraph instantiation. Missing keys raise KeyError at runtime not at definition time.
Fix: Define all state keys in your TypedDict schema upfront

Install

pip install langgraph Python
npm install @langchain/langgraph Node.js

Imports

StateGraph
```
from langgraph.graph import StateGraph
```
Core graph primitive. Unchanged in 1.0.
create_react_agent
```
from langchain import create_agent
```
langgraph.prebuilt deprecated in 1.0. create_react_agent moved to langchain.agents. Use create_agent from langchain instead.
MemorySaver
```
from langgraph.checkpoint.memory import MemorySaver
```
In-memory checkpointer for dev/testing. Not for production.

Quickstart

Minimal single-node StateGraph in LangGraph 1.0.

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

class State(TypedDict):
    messages: list

def call_model(state: State):
    # your LLM call here
    return {"messages": state["messages"]}

graph = StateGraph(State)
graph.add_node("agent", call_model)
graph.set_entry_point("agent")
graph.add_edge("agent", END)

app = graph.compile()
result = app.invoke({"messages": [{"role": "user", "content": "hello"}]})

view raw JSON →