How to intermediate · 3 min read

How to resume LangGraph from checkpoint

Q: How to resume LangGraph from checkpoint

To resume LangGraph from a checkpoint, compile your StateGraph with a checkpointer instance like MemorySaver() or a custom saver, then invoke the compiled app with the saved state. This restores the graph's state and allows continuation from the last checkpoint.

Quick answer

To resume LangGraph from a checkpoint, compile your StateGraph with a checkpointer instance like MemorySaver() or a custom saver, then invoke the compiled app with the saved state. This restores the graph's state and allows continuation from the last checkpoint.

PREREQUISITES

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

Setup

Install the langgraph package via pip if you haven't already:

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

This example shows how to create a StateGraph, add nodes and edges, compile it with a MemorySaver checkpointer, save state, and then resume from that checkpoint.

python

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

# Define the state type
class State(TypedDict):
    messages: list[str]

# Define a node function

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

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

# Create a checkpointer
checkpointer = MemorySaver()

# Compile the graph with the checkpointer
app = graph.compile(checkpointer=checkpointer)

# Initial state
initial_state = {"messages": ["Hello"]}

# Invoke the graph to run and save checkpoint
result = app.invoke(initial_state)
print("First run messages:", result["messages"])

# Save checkpoint key
checkpoint_key = checkpointer.save_checkpoint(result)

# Later: resume from checkpoint
restored_state = checkpointer.load_checkpoint(checkpoint_key)
print("Restored state messages:", restored_state["messages"])

# Continue running from restored state
result2 = app.invoke(restored_state)
print("Continued run messages:", result2["messages"])

output

First run messages: ['Hello', 'response from my_node']
Restored state messages: ['Hello', 'response from my_node']
Continued run messages: ['Hello', 'response from my_node', 'response from my_node']

Common variations

You can implement custom checkpointers by subclassing Checkpointer for persistent storage (e.g., files or databases).
Use async invocation with await app.invoke_async(state) if your nodes are async functions.
Multiple checkpoints can be managed by storing keys externally and loading the desired checkpoint to resume.

Troubleshooting

If you see KeyError when loading a checkpoint, verify the checkpoint key is correct and the checkpoint data exists.
If state is not restored properly, ensure your State TypedDict matches the saved state structure exactly.
For persistent checkpoints, confirm your custom checkpointer correctly serializes and deserializes state.

✅

Key Takeaways

Use MemorySaver or a custom checkpointer to save and load LangGraph state checkpoints.
Compile your StateGraph with the checkpointer parameter to enable checkpointing.
Invoke the compiled app with saved state to resume execution from the checkpoint.
Ensure your state schema matches exactly to avoid deserialization errors.
Async nodes require async invocation methods to resume properly.

Verified 2026-04

Verify ↗