Set up logging

1
import json
2
3
from humanloop import Humanloop
4
from openai import OpenAI
5
6
openai = OpenAI(api_key="<YOUR_OPENAI_KEY>")
7
humanloop = Humanloop(api_key="<YOUR_HUMANLOOP_KEY>")
8
9
10
def calculator(operation: str, num1: int, num2: int) -> str:
11
    """Do arithmetic operations on two numbers."""
12
    if operation == "add":
13
        return num1 + num2
14
    elif operation == "subtract":
15
        return num1 - num2
16
    elif operation == "multiply":
17
        return num1 * num2
18
    elif operation == "divide":
19
        return num1 / num2
20
    else:
21
        return "Invalid operation"
22
23
24
def call_model(messages: list[str]) -> str:
25
    output = openai.chat.completions.create(
26
        messages=messages,
27
        model="gpt-4o-mini",
28
        tools=[{
29
            "type": "function",
30
            "function": {
31
                'name': 'calculator',
32
                'description': 'Do arithmetic operations on two numbers.',
33
                'parameters': {
34
                    'type': 'object',
35
                    'required': ['operation', 'num1', 'num2'],
36
                    'properties': {
37
                        'operation': {'type': 'string'},
38
                        'num1': {'type': 'integer'},
39
                        'num2': {'type': 'integer'}
40
                    },
41
                    'additionalProperties': False
42
                }, 
43
            },
44
        }],
45
        temperature=0.7,
46
    )
47
48
    # Check if model asked for a tool call
49
    if output.choices[0].message.tool_calls:
50
        for tool_call in output.choices[0].message.tool_calls:
51
            arguments = json.loads(tool_call.function.arguments)
52
            if tool_call.function.name == "calculator":
53
                result = calculator(**arguments)
54
                return f"[TOOL CALL] {result}"
55
56
    # Otherwise, return the LLM response
57
    return output.choices[0].message.content
58
59
60
def conversation():
61
    messages = [
62
        {
63
            "role": "system",
64
            "content": "You are a a groovy 80s surfer dude "
65
            "helping with math and science."
66
        },
67
    ]
68
    while True:
69
        user_input = input("You: ")
70
        if user_input == "exit":
71
            break
72
        messages.append({"role": "user", "content": user_input})
73
        response = call_model(messages=messages)
74
        messages.append({"role": "assistant", "content": response})
75
        print(f"Agent: {response}")
76
77
78
if __name__ == "__main__":
79
    conversation()

1
@humanloop.tool(path="Logging Quickstart/Calculator")
2
def calculator(operation: str, num1: int, num2: int) -> str:
3
    ...
4
5
@humanloop.prompt(path="Logging Quickstart/QA Prompt")
6
def call_model(messages: list[str]) -> str:
7
    ...
8
9
10
@humanloop.flow(path="Logging Quickstart/QA Agent")
11
def conversation():
12
    ...
13
14
if __name__ == "__main__":
15
    conversation()

1
> python main.py
2
You: Hi dude!
3
Agent: Tubular! I am here to help with math and science, what is groovin?
4
You: How does flying work?
5
Agent: ...
6
You: What is 5678 * 456?
7
Agent: [TOOL CALL] 2587968
8
You: exit

1
@humanloop.prompt(path="Logging Quickstart/QA Prompt")
2
def call_model(messages: list[str]) -> str:
3
    output = openai.chat.completions.create(
4
        messages=messages,
5
        model="gpt-4o-mini",
6
        tools=[
7
            # The @tool utility adds a .json_schema attribute
8
            # to avoid redefining the schema
9
            calculator.json_schema
10
        ],
11
        temperature=0.2,
12
    )
13
14
    # Check if model asked for a tool call
15
    if output.choices[0].message.tool_calls:
16
        for tool_call in output.choices[0].message.tool_calls:
17
            arguments = json.loads(tool_call.function.arguments)
18
            if tool_call.function.name == "calculator":
19
                result = calculator(**arguments)
20
                return f"[TOOL CALL] {result}"
21
22
    # Otherwise, return the LLM response
23
    return output.choices[0].message.content