How to beginner · 3 min read

How to add checkpointing to LangGraph

Q: How to add checkpointing to LangGraph

Add checkpointing to LangGraph by using a checkpointer like MemorySaver when compiling the graph with graph.compile(checkpointer=MemorySaver()). This enables state persistence across invocations, allowing your AI agent to resume from saved states.

Quick answer

Add checkpointing to LangGraph by using a checkpointer like MemorySaver when compiling the graph with graph.compile(checkpointer=MemorySaver()). This enables state persistence across invocations, allowing your AI agent to resume from saved states.

PREREQUISITES

Python 3.8+
pip install langgraph
Basic knowledge of Python typing and functions

Setup

Install the langgraph package via pip and import necessary classes. Ensure you have Python 3.8 or newer.

bash

pip install langgraph

output

Collecting langgraph
  Downloading langgraph-0.1.0-py3-none-any.whl (10 kB)
Installing collected packages: langgraph
Successfully installed langgraph-0.1.0

Step by step

Create a StateGraph with a typed state, define nodes as functions, add edges, and compile the graph with a MemorySaver checkpointer to enable checkpointing. Invoke the compiled app multiple times to see state persistence.

python

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

class State(TypedDict):
    messages: list[str]

# Define a node function that appends a response

def my_node(state: State) -> State:
    new_messages = state["messages"] + ["response"]
    return {"messages": new_messages}

# Create the graph and add nodes and edges
graph = StateGraph(State)
graph.add_node("my_node", my_node)
graph.set_entry_point("my_node")
graph.add_edge("my_node", END)

# Compile with MemorySaver for checkpointing
app = graph.compile(checkpointer=MemorySaver())

# Initial invocation
result1 = app.invoke({"messages": ["Hello"]})
print("First call messages:", result1["messages"])

# Subsequent invocation resumes from checkpoint
result2 = app.invoke({})
print("Second call messages:", result2["messages"])

output

First call messages: ['Hello', 'response']
Second call messages: ['Hello', 'response', 'response']

Common variations

You can use other checkpointers or implement custom ones by subclassing the Checkpointer interface. Async invocation is supported by calling await app.invoke_async() if your nodes are async functions. Checkpointing works with complex state types as long as they are serializable.

python

import asyncio

async def async_node(state: State) -> State:
    new_messages = state["messages"] + ["async response"]
    return {"messages": new_messages}

graph.add_node("async_node", async_node)
graph.add_edge("my_node", "async_node")
graph.add_edge("async_node", END)

app_async = graph.compile(checkpointer=MemorySaver())

async def run_async():
    result = await app_async.invoke_async({"messages": ["Start"]})
    print("Async call messages:", result["messages"])

asyncio.run(run_async())

output

Async call messages: ['Start', 'response', 'async response']

Troubleshooting

If checkpointing does not persist state, ensure you pass checkpointer=MemorySaver() when compiling the graph.
For custom state types, verify they are JSON serializable.
If state resets unexpectedly, confirm your environment supports file writes or use a persistent checkpointer.

✅

Key Takeaways

Use MemorySaver as a checkpointer to enable checkpointing in LangGraph.
Pass the checkpointer to graph.compile() to persist and resume state.
Checkpointing supports both synchronous and asynchronous node functions.
Ensure your state data is serializable for checkpointing to work correctly.

Verified 2026-04

Verify ↗