InmemoryStore of langgraph

[farouk09]

Hello everyone,

I'm trying to create an email agent that use this function :

embed_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2", model_kwargs={'device': "cuda"})

store = InMemoryStore(
    index={"embed": embed_model}
)

I'm getting this error when doing langgraph dev, which supposed to run the code locally and not using langgraphcloud api as mentioned in documentation :

The langgraph dev command starts a lightweight development server that requires no Docker installation. This server is ideal for rapid development and testing, with features like:

1. Hot reloading: Changes to your code are automatically detected and reloaded
2. Debugger support: Attach your IDE's debugger for line-by-line debugging
   3. In-memory state with local persistence: Server state is stored in memory for speed but persisted locally between restarts

But i'm getting this error!!!!

[hinthornw]

What is your question? LangGraph dev implements the LangGraph api. It's a dev server for the LangGraph api.

[farouk09]

The question is how to solve the issue i'm having! i know that it's a dev server for the LangGraph ap, but it's supposed to have In-memory state with local persistence as mentioned in the docs.

I'm using InMemoryStore and i'm getting an error saying i need to use postgres database, it doesn't make sense.

[hinthornw]

Hello! The exception is saying that it already has an in-memory store that's injected and managed by the server, so simply omitting that store you're trying to add will solve your problem. That's what that message is saying! It does NOT say that you need to add a postgres store!

When you run langgraph dev, the server includes a version of the InMemoryStore in your graph. If you provide an additional one, that additional one is ignored. The reason this exception is thrown is similar to adding a linter in code during development is to communicate this (it's mentioned a few places in the docs as well, but it's not uncommon to skip the docs and jump right into building. That's great!) It is only thrown for dev and not for other runtime contexts because of course the server shouldn't break production code. It only raises at dev-time for the similar reason people use linters.

The reason this exception exists is that we want to save people time (hopefully). We realized number of people spent time writing custom store or checkpointer implementations that wouldn't fully implement the APIS required to satisfy other expectations of the platform (like TTLs, indexed search, etc.). We want the server to be able to satisfy these configurable aspects of the server and have the types of runtime guarantees we'd expect in production. To ensure this, we provide a a manged and benchmarked version of the persistence layer and do not currently permit custom stores or checkpointers when deploying. Since these implementations would be ignored anyhow, and to help communicate this earlier on, we say so in the docs and raise an exception here.

[farouk09]

Hi! Thank you for the detailed explanation — I understand now.

So if I got it right: there's no need to manually create an InMemoryStore since LangGraph adds it automatically in langgraph dev.

However, I was previously doing something like this:

store = InMemoryStore(index={"embed": embed_model})
email_agent = email_agent.compile(store=store)

The reason I needed this is because I'm building an email agent that uses semantic memory, with these two functions from langmem:

manage_memory_tool = create_manage_memory_tool(
    namespace=("email_assistant", "{langgraph_user_id}", "collection")
)

search_memory_tool = create_search_memory_tool(
    namespace=("email_assistant", "{langgraph_user_id}", "collection")
)

These rely on the store having an embedding model (index={"embed": embed_model}), so I’m wondering:

➡️ How can I specify which embedding model LangGraph should use internally for indexing the store?

And since I used to pass the store explicitly in email_agent.compile(store=store), should I now just call:

email_agent.compile()
Thanks again for the help! 🙏