Ollama LLM not actually calling the tool, but correctly identifying the need for tool

[RETR0-OS]

I am trying to test the tool-calling functionality of langchain with the Ollama Chat model. I am using llama3-groq-tool-use:latest as the LLM.

Here is my code:

from langchain_ollama import ChatOllama
from langchain_core.tools import tool

@tool
def secret_code(username: str) -> str:
    """
    A function to generate a secret code based on the username of the user.
    :param username:
    :return: the secret code
    """
    return f"Your secret code is: {username[::-1]}"

llm = ChatOllama(
    model="llama3-groq-tool-use:latest",
    temperature=0.3
)

llm_with_custom_tool = llm.bind_tools([secret_code])

result = llm_with_custom_tool.invoke("Hello, my username is John. Can you get me my secret code?")
print(result)
print(result.tool_calls)

I am getting the following output:

content='<tool_call>\n{"id": 0, "name": "secret_code", "arguments": {"username": "John"}}\n</tool_call>' additional_kwargs={} response_metadata={'model': 'llama3-groq-tool-use:latest', 'created_at': '2025-06-08T13:23:14.6043758Z', 'done': True, 'done_reason': 'stop', 'total_duration': 7987012700, 'load_duration': 5166449600, 'prompt_eval_count': 175, 'prompt_eval_duration': 1047000000, 'eval_count': 26, 'eval_duration': 1767000000, 'model_name': 'llama3-groq-tool-use:latest'} id='run--79c5f8d2-c0ba-4a29-9ee4-e06f393b88f0-0' usage_metadata={'input_tokens': 175, 'output_tokens': 26, 'total_tokens': 201}
[]

The LLM correctly identifies that it needs to call a tool but does not actually call it. How do I fix this issue?

[Kana2581]

This result is correct. You need to use a toolnode to execute this result. you could see https://github.com/langchain-ai/langchain-academy/blob/main/module-1/agent.ipynb

[RETR0-OS]

Hello, @Kana2581
Thank you for your response. I was following the tutorial posted with the Langgraph docs, and they state that:

Now, create a function to run the tools if they are called. Do this by adding the tools to a new node called BasicToolNode that checks the most recent message in the state and calls tools if the message contains tool_calls

However, my response does not have a tool_calls attribute associated with it. That is where the issue arises. Do you have any idea as to why I am getting the tool object in the content instead of the tool_calls? Shouldn't the content contain the processed result of the tool call?

Thank you

[Kana2581]

If you're following the example from the LangGraph tutorial, a successful tool call typically results in at least three messages:

1. The first AIMessage will include the tool_calls attribute, indicating which tools need to be invoked.
2. The second message is a ToolMessage, which contains the results returned by the tool.
3. The third AIMessage contains only the content, with no tool_calls, as it represents the final response after processing the tool result.