Passing config to tools contained inside agent executor

[tyadav7]

I have an agent executor like

config = {"user_id" : 123 , "conversation_id" : 1234}

AgentExecutor(
agent=agent,
llm=llm,
tools=tools,
verbose=True,
handle_parsing_errors=True,
callbacks=[FileCallbackHandler("logs/genai.log")],
).withConfig(config=config)

and I have a tool

@tool()
def get_some_sauce(input: str, config: RunnableConfig):

add_route(

I want the agent executor when invoked to pass down the runnable config to the tool. How to achieve this?

[dosubot[bot]]

To pass a runnable config to a tool within an AgentExecutor, you need to ensure that your tool is set up to accept a RunnableConfig parameter. LangChain will automatically populate this parameter with the correct config value when the tool is invoked. Here's how you can achieve this:

1. Define your tool with a RunnableConfig parameter:

from langchain_core.runnables import RunnableConfig
from langchain_core.tools import tool

@tool
def get_some_sauce(input: str, config: RunnableConfig):
    # Your tool logic here
    pass

2. When you configure your AgentExecutor, use the withConfig method to pass the configuration:

config = {"user_id": 123, "conversation_id": 1234}

agent_executor = AgentExecutor(
    agent=agent,
    llm=llm,
    tools=tools,
    verbose=True,
    handle_parsing_errors=True,
    callbacks=[FileCallbackHandler("logs/genai.log")]
).withConfig(config=config)

This setup ensures that when the AgentExecutor is invoked, the RunnableConfig is passed to the get_some_sauce tool, allowing it to access the configuration values like user_id and conversation_id ^[1].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}

[tyadav7]

I have already done that but the runnableConfig is not passed to the Tool

[tyadav7]

@dosui have done that but this does not propagate my config in the runnable tool.Sent from my iPhoneOn 13 Dec 2024, at 1:40 AM, dosubot[bot] ***@***.***> wrote: To pass a runnable config to a tool within an AgentExecutor, you need to ensure that your tool is set up to accept a RunnableConfig parameter. LangChain will automatically populate this parameter with the correct config value when the tool is invoked. Here's how you can achieve this: Define your tool with a RunnableConfig parameter: from langchain_core.runnables import RunnableConfig from langchain_core.tools import tool @tool def get_some_sauce(input: str, config: RunnableConfig): # Your tool logic here pass When you configure your AgentExecutor, use the withConfig method to pass the configuration: config = {"user_id": 123, "conversation_id": 1234} agent_executor = AgentExecutor( agent=agent, llm=llm, tools=tools, verbose=True, handle_parsing_errors=True, callbacks=[FileCallbackHandler("logs/genai.log")] ).withConfig(config=config) This setup ensures that when the AgentExecutor is invoked, the RunnableConfig is passed to the get_some_sauce tool, allowing it to access the configuration values like user_id and conversation_id [1]. To continue talking to Dosu, mention @dosu. Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>

[dosubot[bot]]

To debug why the RunnableConfig is not being passed to your tool, you can use LangChain's debugging and logging features to trace the configuration passing process. Here are some steps you can take:

1. Enable Debugging: Use the set_debug(True) function to enable detailed logging of inputs and outputs for all LangChain components. This will help you see if the configuration is being passed correctly.

2. Use ConsoleCallbackHandler: Add a ConsoleCallbackHandler to your AgentExecutor to log events to the console. This can provide insights into the execution flow and where the configuration might be failing to propagate.

3. Check Tool Signature: Ensure that your tool's function signature correctly includes the RunnableConfig parameter and that it is being used within the tool.

4. Verify Configuration Structure: Double-check that the configuration structure you are passing matches what your tool expects. Any mismatch could prevent the config from being used correctly.

Here's an example of how you might set up debugging:

from langchain.globals import set_debug
from langchain_core.tracers import ConsoleCallbackHandler

set_debug(True)

agent_executor = AgentExecutor(
    agent=agent,
    llm=llm,
    tools=tools,
    verbose=True,
    handle_parsing_errors=True,
    callbacks=[FileCallbackHandler("logs/genai.log"), ConsoleCallbackHandler()]
).withConfig(config={"user_id": 123, "conversation_id": 1234})

This setup will help you trace the configuration passing process and identify any issues. If the problem persists, consider reviewing the tool's implementation and the configuration structure for any discrepancies ^[1][2][3].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}