Add tools to text chat

[KyrianC]

flow

1. Define tools according to your own functions

Tools must be defined accoding to json-schema definition

function get_weather(location: str, units: Literal["Celsius", "Fahrenheit"] = "Celsius"):
  ...

tools = [
    {
        "name": "get_weather",
        "description": "Get The weather data from a location",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {
                    "description": "The geographical location for which weather information is requested.",
                    "type": "string",
                },
                "units": {
                    "description": "The units of measurement for the temperature.",
                    "type": "string",
                    "enum": ["Celsius", "Fahrenheit"],
                    "default": "Celsius",
                },
            },
            "required": ["location"],
        },
    }
]

map_tool_to_func = {
  "get_weather": get_weather
}

2. Ask chat for question related with tool, and pass tool_definitions

tool_call_result = chat(
    text="What is the weather in Brest?",
    chatbot_global_action=None,
    previous_history=history,
    temperature=1,
    max_tokens=1000,
    model=model,
    available_tools=tools,
)

messages = tool_call_result.standardized_response.message
tool_calls = messages[1].tool_calls

# IMPORTANT: providers needs you to add to history otherwise it won't work
history = [msg.model_dump() for msg in messages]

print(tool_calls)

output:
[{"id": "8KS3Mz2wmYGqUcfncF0ymjAUxozUH", "name": "get_weather", "arguments": '{"location": "Brest"}'}]

3. Call your own tool with passed arguments

tool_results = []
for tool_call in tool_calls 
    arguments = json.loads(tool_call['arguments'])
    tool_call_result = map_tool_to_func[tool["name"]](**arguments)
    # {"temperature": 10, 'units': "Celsius", "weather": "rainy"}
    tool_results.append({"id": tool_call["id"], "result": tool_call_result})

4. Get text response back from Chat

chat_result = chat(
    text="",
    chatbot_global_action=None,
    previous_history=history,
    temperature=1,
    max_tokens=1000,
    model=model,
    tool_results=tool_results,
)

print(chat_result.standardized_response.generated_text)

output:
The Weather in Brest is currently rainy with a temperature of 10 degrees Celsius.

[KyrianC]

Don't hesitate to tell me if you think the flow is not simple enough.
like we must add previous messages to history for it to work.

Another way would be to make the user repeat the first request, and just add tool_resultsto it. so I could reconstruct a history from that. That's what cohere is doing.

[Daggx]

Don't hesitate to tell me if you think the flow is not simple enough. like we must add previous messages to history for it to work.

Another way would be to make the user repeat the first request, and just add tool_resultsto it. so I could reconstruct a history from that. That's what cohere is doing.

The flow may not be simple but if the users are familiar with how function tooling works with other providers than its seems clear