Local deployment server with dockercompose, how does it work with checkpointer?

[laelhalawani]

If I use langgraph cli to create dockercompose to set up the server locally, how does that work with memory checkpointer configured with the graph?
When using cli to compose we're setting up redis and postgres containers.
https://langchain-ai.github.io/langgraph/how-tos/deploy-self-hosted/#using-docker-compose
Does that mean checkpointer settings for graphs are overwritten and the database will be postgres and the graph will automatically connect, or do I still need to set up a persistent checkpointer for the graph to retain the information in case of server restart?

[BoyPit]

Hi,

Any update ?
Same question here, if i use :

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

Does the checkpoint store in the pg container ?

Or do i need to run :

graph = workflow.compile()

Thank you !

[hinthornw]

Got it—here’s a cleaner, more direct version in the same tone as your last one:

---

Hi @laelhalawani, thanks for the question. All deployments of LangGraph Server (including local ones via Docker Compose) come with a built-in Postgres checkpointer. That means if you pass a custom saver like MemorySaver(), it will be ignored at runtime. You can just drop that argument and write:

graph = workflow.compile()

The server will automatically persist checkpoints to the Postgres container defined in the docker-compose.yml. If you’d like to connect to your own Postgres instance instead, just set the POSTGRES_URI environment variable—details here:
https://langchain-ai.github.io/langgraph/cloud/reference/env_var/#postgres_uri_custom

MemorySaver is an in-memory-only implementation meant for testing and debugging; if you use it in a deployment context, your state won’t survive a restart.