Force to call a tool

[TaQuangTu]

Hi Lang-graph developers, you are doing a great project!

I want to discuss about how to force a graph to always call a tool before invoking any other tool based on the context.
In our current implementation, when creating an agent with create_react_agent function, I'm noticed that there is no option for us to specify tool_choice parameter, hence cannot achieve that expected behavior. Is there any reason preventing us from implementing it or is there another way to achieve it?

Thanks!

[eyurtsev]

See this section in the docs: https://langchain-ai.github.io/langgraph/how-tos/tool-calling/#force-tool-use

from langchain_core.tools import tool

# highlight-next-line
@tool(return_direct=True)
def greet(user_name: str) -> int:
    """Greet user."""
    return f"Hello {user_name}!"

tools = [greet]

agent = create_react_agent(
    # highlight-next-line
    model=model.bind_tools(tools, tool_choice={"type": "tool", "name": "greet"}),
    tools=tools
)

agent.invoke(
    {"messages": [{"role": "user", "content": "Hi, I am Bob"}]}
)

[TaQuangTu]

Hi @eyurtsev E
Thank you for your prompt response!
The solution you sent would required me to provide args, which i want to fill dynamically at runtime by LLM, also it does not help me to achieve the behavior of "force to call a tool then invoke one or more tools intelligently".

[eyurtsev]

Ah you'll need to build an agent from scratch for now and put this logic in the call model node. It'll give you the flexibility to force a tool call and then change the bound tools.

Cc @sydney-runkle we can add this to the feature list