Adding configurable options to a chain for langserve

[jambudipa]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

def get_dynamic_human_prompt(config):
    personality_name = config.get("configurable", {}).get("personality_name", "study-buddy")
    return personality_cache.find_personality_by_name(personality_name)['llm_prompt']


def create_dynamic_rag_prompt(inputs, config):
    human_prompt = get_dynamic_human_prompt(config)
    return ChatPromptTemplate.from_messages(
        [
            ("system", rag_system_prompt),
            MessagesPlaceholder(variable_name="history"),
            ("human", human_prompt)
        ]
    )


retriever_chain = RunnablePassthrough.assign(
    context=question_chain
            | books_parent_retriever
            | (lambda docs: combine_metadata_and_content(docs))
)

rag_chain = (
    retriever_chain
    | (lambda inputs, config: create_dynamic_rag_prompt(inputs, config))
    | llm_gpt4o
    | parse_output
)

with_message_history = RunnableWithMessageHistory(
    rag_chain,
    get_session_history,
    input_messages_key="question",
    history_messages_key="history",
)

Description

I have been at this for about five hours. It seems fairly basic to me, but I cannot work it out!

TLDR; how do I indicate to langserve the requirement for a particular configurable property on my with_message_history runnable?

Currently, the reported FastAPI schema for my chain looks like this:

{
  "input": {
    "question": "string"
  },
  "config": {
    "configurable": {
      "session_id": ""
    }
  },
  "kwargs": {}
}

I am trying to introduce a personality_name property that is used to create a prompt, rather than have a fixed prompt.

In my code above, I have no idea how to indicate to langserve that there is now a prompt_name configurable, so that the API would now look like this:

{
  "input": {
    "question": "string"
  },
  "config": {
    "configurable": {
      "session_id": "",
      "personality_name": ""
    }
  },
  "kwargs": {}
}

I have tried at least a half dozen ways to enforce this, but none of them have worked.

history_factory_config looked promising, but then this passes all of those properties to the history retriever, which has no interest in them and I believe should not be modified. I suppose I want a sort of "passthrough" behaviour.

I did encounter this explanation of configuring runnables, but the warning at the top made me hesitate, since I do not believe this to be an advanced requirement.

So I thought I would check with the community before implementing my own Runnable.

System Info

System Information

OS: Darwin
OS Version: Darwin Kernel Version 24.3.0: Mon Dec 9 19:48:56 PST 2024; root:xnu-11215.80.25.501.2~1/RELEASE_ARM64_T6000
Python Version: 3.12.8 (main, Dec 3 2024, 18:42:41) [Clang 16.0.0 (clang-1600.0.26.4)]

Package Information

langchain_core: 0.3.28
langchain: 0.3.13
langsmith: 0.2.7
langchain_cli: 0.0.35
langchain_mongodb: 0.3.0
langchain_openai: 0.2.14
langchain_text_splitters: 0.3.4
langserve: 0.3.1

Other Dependencies

aiohttp: 3.11.11
async-timeout: Installed. No version info available.
fastapi: 0.115.6
gitpython: 3.1.43
gritql: 0.1.5
httpx: 0.28.1
jsonpatch: 1.33
langserve[all]: Installed. No version info available.
langsmith-pyo3: Installed. No version info available.
motor: 3.6.0
numpy: 1.26.4
openai: 1.58.1
orjson: 3.10.13
packaging: 24.2
pydantic: 2.10.4
pymongo: 4.9.2
PyYAML: 6.0.2
requests: 2.32.3
requests-toolbelt: 1.0.0
SQLAlchemy: 2.0.36
sse-starlette: 1.8.2
tenacity: 9.0.0
tiktoken: 0.8.0
tomlkit: 0.13.2
typer[all]: Installed. No version info available.
typing-extensions: 4.12.2
uvicorn: 0.23.2
zstandard: Installed. No version info available.

[jambudipa]

Was easy enough it turns out

class InputChat(BaseModel):
    question: str
    prompt_name: str

add_routes(
    app,
    with_message_history.with_types(input_type=InputChat),
    path="/kadampa_gpt"
)