trying to set up a LangServe application with RunnableWithMessageHistory and a chat playground about langserve HOT 4 CLOSED

This issue is in RunnableWithMessageHistory get_output_schema, I was able to work around it by defining my chain as follow:

from langchain_core.output_parsers.string import StrOutputParser

json_parser = StrOutputParser()

chain_with_parser = (
    chain_with_history |
    json_parser
)

then passing chain_with_parser to add_routes:

# Edit this to add the chain you want to add
add_routes(app, chain_with_parser)

not sure exactly how to solve the issue in RunnableWithMessageHistory tho.

from langserve.

I'm having this same error. Code is very similar —

agent_executor = create_agent_executor(database=DATABASE)

def get_message_history(jwt):
    return RedisChatMessageHistory(url=REDIS_URL, ttl=600, session_id=jwt)


with_message_history = RunnableWithMessageHistory(
    agent_executor,
    get_message_history,
    input_messages_key="input",
    history_messages_key="history",
    history_factory_config=[
        ConfigurableFieldSpec(
            id="jwt",
            annotation=str,
            name="User JWT",
            description="Unique session identifier for the user.",
            default="",
            is_shared=True,
        ),
    ],
)

add_routes(
    app,
    with_message_history,
    per_req_config_modifier=_per_req_config_modifier,
    path="/agent",
)

I suspect something with pydantic versions. I'm using pydantic = "^2.8.2" in my pyproject.toml.

I'm using an app generated with the langchain-cli last week.

from langserve.

I'm trying to integrate both RunnableWithMessageHistory as chat with persistence playground_type='chat'

app = FastAPI()

llm_model_path = "model/llama-3-8b.gguf"

llm = ChatLlamaCpp(model_path=llm_model_path,
                   n_gpu_layers=-1,
                   n_ctx=512,
                   chat_format='chatml')

# Declare a chain
# prompt = ChatPromptTemplate.from_messages(
#     [
#         ("system", "You are a helpful AI assistant. Please provide a concise list of course with university that suits users question."),
#         MessagesPlaceholder(variable_name="history"),
#         ("human", "{human_input}"),
#     ]
# )

prompt = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful AI assistant that recommends courses along with universities to the students. Please provide a concise list of course with university that suits users question."),
        MessagesPlaceholder(variable_name="messages"),
    ]
)

chain = prompt | llm | StrOutputParser()

def get_session_history():
    return InMemoryChatMessageHistory()

class InputChat(BaseModel):
    """Input for the chat endpoint."""

    messages: List[Union[HumanMessage, AIMessage, SystemMessage]] = Field(
        ...,
        description="The chat messages representing the current conversation.",
    )

chain_with_history = RunnableWithMessageHistory(
    chain,
    get_session_history,
    input_messages_key="input",
    history_messages_key="chat_history"
).with_types(input_type=InputChat)

add_routes(
    app,
    chain_with_history,
    path="/chat",
    playground_type="chat"
)

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)

TypeError: issubclass() arg 1 must be a class

But RunnableWithMessageHistory is a class right?

Is there a option to do this?

My Configs are:
pydantic-2.8.2
pydantic-core- 2.20.1

from langserve.

This issue is in RunnableWithMessageHistory get_output_schema, I was able to work around it by defining my chain as follow:

from langchain_core.output_parsers.string import StrOutputParser

json_parser = StrOutputParser()

chain_with_parser = (
    chain_with_history |
    json_parser
)

then passing chain_with_parser to add_routes:

# Edit this to add the chain you want to add
add_routes(app, chain_with_parser)

not sure exactly how to solve the issue in RunnableWithMessageHistory tho.

Yes, It's solved!!

Still, The chat playground is only supported for chains that take one of the following as input:

• a dict with a single key containing a list of messages
• a dict with two keys: one a string input, one an list of messages

and which return either an AIMessage or a string.

You can test this chain in the default LangServe playground instead.

To use the default playground, set playground_type="default" when adding the route in your backend.

from langserve.