How to check Langserve Rest API swagger for ConversationalRetrievalChain with memory.

[NageshMashette]

memory = ConversationBufferWindowMemory(k=4, memory_key="chat_history", return_messages=True)
qa = ConversationalRetrievalChain.from_llm(
llm,
db.as_retriever(search_kwargs={"k": 3}),
memory=memory
)

I created RAG QA with ConversationalRetrievalChain with memory, and did

app = FastAPI()
from langserve import add_routes
add_routes(app, qa)

its executed and given api also swagger UI also available how to test and in which endpoint i need to pass user question to get the response.

[eyurtsev]

Hi @NageshMashette,

1. Docs are available here: localhost:800/docs. Check the examples directory to see client and server combinations. I suggest using the python RemoteRunnable client to interact with the server.
2. The chain that you've implementing will NOT work correctly for what I'd consider server use since it has a single global state for the memory that will be shared between all users. We'll be releasing more documentation on how to set this up in the upcoming 1-2 weeks.

[eyurtsev]

Closing as this is not an issue but a question

[NageshMashette]

https://github.com/langchain-ai/langserve/blob/main/examples/conversational_retrieval_chain/client.ipynb
without memory also im facing issue , i fallowed client code provided in example folder.

i used
**from langserve import RemoteRunnable
remote_runnable = RemoteRunnable("http://localhost:8000/")

await remote_runnable.ainvoke({"question": "Hi how are you?", "chat_history": ""})**

HTTPStatusError: Client error '404 Not Found' for url 'http://localhost:8000/invoke'
For more information check: https://httpstatuses.com/404 for {"detail":"Not Found"}

[murali-innerfit]

faced same issue .After updating to latest version of langserve its started working for me.

[NageshMashette]

faced same issue .After updating to latest version of langserve its started working for me.

are you able to execute with memory?

[navjotmakkar]

Has anyone tried to combine conversation memory with langchain chain that is served using langserve?

[eyurtsev]

I don't recommend using the old memory abstractions. Instead take a look at the examples in the readme there are examples that show how to handle conversation history including multi user support.

[NageshMashette]

thanks @eyurtsev ok let me try it.