pass config values to agent tool function

[shaojun]

Say I have a chat application with user id and conversation id under config, and need to support tools calling.
The core code would be:

def func1(product_name: str):
    # how to get user id and conversation id which are necessary to get a user based vector store
    pass
def func2(product_name: str):
    # how to get user id and conversation id which are necessary to get a user based vector store
    pass
tools = [func1,func2]
llm_with_tools = llm.bind(
    tools=[convert_to_openai_tool(tool) for tool in tools])
agent = (
    {
        "input": lambda x: x["human_input"],
        "agent_scratchpad": lambda x: format_to_openai_tool_messages(
            x["intermediate_steps"]
        ),
        "chat_history": lambda x: x["chat_history"],
    }
    | prompt
    | llm_with_tools
    | OpenAIToolsAgentOutputParser()
)
agent_executor = AgentExecutor(
    agent=agent, tools=tools, verbose=True)
chain_with_history = RunnableWithMessageHistory(
    agent_executor,
    chat_message_history_factory,
    input_messages_key="human_input",
    history_messages_key="chat_history",
    history_factory_config=[
        ConfigurableFieldSpec(
            id="user_id",
            annotation=str,
            name="User ID",
            description="Unique identifier for the user.",
            default="",
            is_shared=True,
        ),
        ConfigurableFieldSpec(
            id="conversation_id",
            annotation=str,
            name="Conversation ID",
            description="Unique identifier for the conversation.",
            default="",
            is_shared=True,
        ),
    ],
).with_types(input_type=InputChat)

How to pass the user_id and conversation_id from config to that 2 tool functions?

[eyurtsev]

You need to either generate the tools at run time and bind to them information from the config, or you can inherit from the structuredtool class (which is a runnable) and set it up with configurable tools.

The main issue is the agent executor in langchain does not propagate configuration information (I suspect that the langgraph agents do not have that limitation so you could try those as well!), so you'll need to wrap it as a custom runnable; e.g., https://github.com/langchain-ai/langserve/tree/main/examples/configurable_agent_executor