Logging through API

1
def call_model(trace_id: str, messages: list[str]) -> str:
2
    ...
3
4
def conversation():
5
    trace_id = humanloop.flows.log(
6
        path="Logging Quickstart/QA Agent",
7
        flow={
8
            "attributes": {},
9
        },
10
    ).id
11
    messages = [
12
        {
13
            "role": "system",
14
            "content": "You are a a groovy 80s surfer dude "
15
            "helping with math and science.",
16
        },
17
    ]
18
    while True:
19
        user_input = input("You: ")
20
        if user_input == "exit":
21
            break
22
        messages.append({"role": "user", "content": user_input})
23
        response = call_model(trace_id=trace_id, messages=messages)
24
        messages.append({"role": "assistant", "content": response})
25
        print(f"Agent: {response}")

1
def call_model(trace_id: str, messages: list[str]) -> str:
2
    prompt_start_time = datetime.datetime.now()
3
    output = openai.chat.completions.create(
4
        messages=messages,
5
        model="gpt-4o",
6
        tools=[{
7
            "type": "function",
8
            "function": TOOL_JSON_SCHEMA,
9
        }],
10
        temperature=0.7,
11
    )
12
    prompt_log_id = humanloop.prompts.log(
13
        path="Logging Quickstart/QA Prompt",
14
        prompt={
15
            "model": "gpt-4o",
16
            "tools": [TOOL_JSON_SCHEMA],
17
            "temperature": 0.7,
18
        },
19
        output=output.choices[0].message.content,
20
        trace_parent_id=trace_id,
21
        start_time=prompt_start_time,
22
        end_time=datetime.datetime.now(),
23
    ).id
24
25
    # Check if model asked for a tool call
26
    if output.choices[0].message.tool_calls:
27
        for tool_call in output.choices[0].message.tool_calls:
28
            arguments = json.loads(tool_call.function.arguments)
29
            if tool_call.function.name == "calculator":
30
                tool_start_time = datetime.datetime.now()
31
                result = calculator(**arguments)
32
                humanloop.tools.log(
33
                    path="Logging Quickstart/Calculator",
34
                    tool={
35
                        "name": "calculator",
36
                        "description": "Do arithmetic operations on two numbers.",
37
                        "function": TOOL_JSON_SCHEMA,
38
                    },
39
                    inputs=arguments,
40
                    output=result,
41
                    trace_parent_id=prompt_log_id,
42
                    start_time=tool_start_time,
43
                    end_time=datetime.datetime.now(),
44
                )
45
                return f"[TOOL CALL] {result}"
46
47
    # Otherwise, return the LLM response
48
    return output.choices[0].message.content

1
def conversation():
2
    trace_id = humanloop.flows.log(
3
        path="Logging Quickstart/QA Agent",
4
        flow={
5
            "attributes": {},
6
        },
7
    ).id
8
    messages = [
9
        {
10
            "role": "system",
11
            "content": "You are a a groovy 80s surfer dude "
12
            "helping with math and science.",
13
        },
14
    ]
15
    while True:
16
        user_input = input("You: ")
17
        if user_input == "exit":
18
            break
19
        messages.append({"role": "user", "content": user_input})
20
        response = call_model(trace_id=trace_id, messages=messages)
21
        messages.append({"role": "assistant", "content": response})
22
        print(f"Agent: {response}")
23
24
    humanloop.flows.update_log(
25
        log_id=trace_id,
26
        output="",
27
        status="complete",
28
    )