Skip to main content

Tool calling

ChatModelInput accepts a list of tools. When the model decides to call one, the chatbot returns a tool_response item instead of text, with the name and the arguments of each call. Your code runs the tool and sends the result back as a new message.

OpenAI​

Use the OpenAI tool format:

import json
from intelli.function.chatbot import Chatbot, ChatProvider
from intelli.model.input.chatbot_input import ChatModelInput

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get the current weather in a city",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}]

chat_input = ChatModelInput("You are a weather assistant.", model="gpt-4.1", tools=tools)
chat_input.add_user_message("What is the weather in Paris?")

openai_bot = Chatbot(YOUR_OPENAI_KEY, ChatProvider.OPENAI)
result = openai_bot.chat(chat_input)[0]

if isinstance(result, dict) and result.get("type") == "tool_response":
    for call in result["tool_calls"]:
        name = call["function"]["name"]
        args = json.loads(call["function"]["arguments"])
        print(name, args)  # get_weather {'city': 'Paris'}
else:
    print(result)  # the model answered without a tool

GPT-5 models run on the responses API, which takes the flat tool format: {"type": "function", "name": ..., "description": ..., "parameters": ...}. The reply uses the same tool_response shape.

Anthropic​

Use the Anthropic tool format with input_schema:

tools = [{
    "name": "get_weather",
    "description": "Get the current weather in a city",
    "input_schema": {
        "type": "object",
        "properties": {"city": {"type": "string"}},
        "required": ["city"],
    },
}]

chat_input = ChatModelInput("You are a weather assistant.", model="claude-sonnet-5", tools=tools)
chat_input.add_user_message("What is the weather in Paris?")

claude_bot = Chatbot(YOUR_ANTHROPIC_KEY, ChatProvider.ANTHROPIC)
result = claude_bot.chat(chat_input)[0]

When Claude calls several tools in one turn, all of them are in result["tool_calls"], in the same shape as the OpenAI reply.

Tool Choice​

Set tool_choice to force or disable tool use:

# openai
chat_input = ChatModelInput("You are a weather assistant.", model="gpt-4.1", tools=tools, tool_choice="required")

# anthropic
chat_input = ChatModelInput("You are a weather assistant.", model="claude-sonnet-5", tools=tools, tool_choice={"type": "any"})

Tools in Flows​

To let a model pick between tools and other tasks inside a workflow, see Dynamic tool. To use the tools of an MCP server, see MCP.