LangChain Tools and Tool Calling

LangChain tools let a model do more than write text. A tool is a Python function with a name, description and input schema. With tool calling, the model decides which tool to use and with what arguments, and your code runs the function.

Creating a Tool with @tool

from langchain.tools import tool

@tool
def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    fake = {"lahore": "34 C, sunny", "london": "12 C, rain"}
    return fake.get(city.lower(), "unknown")

print(get_weather.name)
print(get_weather.description)
print(get_weather.invoke({"city": "Lahore"}))

get_weather
Get the current weather for a city.
34 C, sunny

The function name becomes the tool name, the docstring becomes the description, and type hints become the argument schema. Good descriptions matter: the model reads them to decide when to call the tool.

Binding Tools to a Model

from langchain_ollama import ChatOllama

llm = ChatOllama(model="llama3.2", temperature=0)
llm_with_tools = llm.bind_tools([get_weather])

msg = llm_with_tools.invoke("Is it raining in London?")
print(msg.tool_calls)

Sample output:
[{'name': 'get_weather', 'args': {'city': 'London'}, 'id': 'call_1', 'type': 'tool_call'}]

The model did not answer directly. It returned a tool call request. The model never runs code itself; your program does.

Completing the Tool Loop

from langchain_core.messages import HumanMessage

messages = [HumanMessage("Is it raining in London?")]
ai = llm_with_tools.invoke(messages)
messages.append(ai)
for call in ai.tool_calls:
    messages.append(get_weather.invoke(call))   # returns a ToolMessage
final = llm_with_tools.invoke(messages)
print(final.content)

Sample output:
Yes, it's raining in London right now, about 12 C.

Invoking a tool with the whole tool call dictionary returns a ToolMessage linked by ID. Sending it back lets the model write the final answer. Agents (next lesson) automate this loop.

Tool Design Tips

• Give each tool one clear job and a specific docstring.
• Use simple argument types; add Field descriptions with an args_schema for complex inputs.
• Return short strings or JSON; long outputs waste context.
• Validate inputs and never let tools run dangerous actions without checks.
• Pick a model that supports tool calling (llama3.2, qwen3, mistral and similar on Ollama).

Frequently asked questions

Does Ollama support tool calling in LangChain?

Yes, ChatOllama supports bind_tools for models that have tool calling capability, such as llama3.1/3.2, qwen3 and mistral.

Where is the @tool decorator in LangChain 1.0?

Import it with from langchain.tools import tool. It is also available from langchain_core.tools.

Can the model call several tools at once?

Yes, many models return multiple tool_calls in one message (parallel tool calling). Run each and send back all ToolMessages.