밑바닥부터 만드는 AI 에이전트 - OpenAI API 만으로 메모리, 도구 호출 구현하기

curl -LsSf https://astral.sh/uv/install.sh | sh
# 또는
brew install uv

from dotenv import load_dotenv
import os

load_dotenv()
print(os.getenv("OPENAI_API_KEY"))

import openai

client = openai.OpenAI()

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {
            "role": "user",
            "content": "Hello, how are you?"
        }
    ]
)

response

ChatCompletion(id='chatcmpl-DoTMqGFDXOErae4jg0rMqbuLlFRbE', choices=[Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(content="Hello! I'm just a program, so I don't have feelings, but I'm here and ready to assist you. How can I help you today?", refusal=None, role='assistant', annotations=[], audio=None, function_call=None, tool_calls=None))], created=1780921660, model='gpt-4o-mini-2024-07-18', object='chat.completion', moderation=None, service_tier='default', system_fingerprint='fp_080e1eab01', usage=CompletionUsage(completion_tokens=29, prompt_tokens=13, total_tokens=42, completion_tokens_details=CompletionTokensDetails(accepted_prediction_tokens=0, audio_tokens=0, reasoning_tokens=0, rejected_prediction_tokens=0), prompt_tokens_details=PromptTokensDetails(audio_tokens=0, cached_tokens=0)))

for choice in response.choices:
    print(choice)

Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(content="Hello! I'm here and ready to help you. How can I assist you today?", refusal=None, role='assistant', annotations=[], audio=None, function_call=None, tool_calls=None))
Choice(finish_reason='stop', index=1, logprobs=None, message=ChatCompletionMessage(content="Hello! I'm just a program, but I'm here and ready to help you. How can I assist you today?", refusal=None, role='assistant', annotations=[], audio=None, function_call=None, tool_calls=None))
Choice(finish_reason='stop', index=2, logprobs=None, message=ChatCompletionMessage(content="Hello! I'm just a computer program, so I don't have feelings, but I'm here and ready to help you. How can I assist you today?", refusal=None, role='assistant', annotations=[], audio=None, function_call=None, tool_calls=None))
Choice(finish_reason='stop', index=3, logprobs=None, message=ChatCompletionMessage(content="Hello! I'm just a computer program, so I don't have feelings, but I'm here and ready to help you. How can I assist you today?", refusal=None, role='assistant', annotations=[], audio=None, function_call=None, tool_calls=None))
Choice(finish_reason='stop', index=4, logprobs=None, message=ChatCompletionMessage(content="Hello! I'm here and ready to help you. How can I assist you today?", refusal=None, role='assistant', annotations=[], audio=None, function_call=None, tool_calls=None))

import openai

client = openai.OpenAI()

PROMPT = """
  너는 아래 3가지 함수를 가지고 있어.
  
  - get_popular() - 인기 항목를 조회.
  - get_details(id) - 상세 정보를 조회.
  - get_credits(id) - 주문 내역을 조회.
  
  이제 다음 3가지 질문에 맞는 함수를 선택해서 함수명과 인자(argument)를 반환하고 이외의 불필요한 대답은 하지마.

  1. 지금 인기있는 항목이 무엇인지 알려줘.
  2. ID 550에 해당하는 항목의 상세정보를 알려줘.
  3. ID 550에 해당하는 주문 내역을 알려줘.
"""

response = client.chat.completions.create(
    model="gpt-4o-mini",
    # n=5,
    messages=[
        {
            "role": "user",
            "content": PROMPT
        }
    ],
)

import openai

client = openai.OpenAI()
messages = []

def call_ai():
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=messages,
    )
    message = response.choices[0].message.content
    messages.append({
        "role": "assistant",
        "content": message
    })
    print(f"AI: {message}")

while True:
    message = input("Send a message to the LLM ...")
    if message == "quit" or message == "q":
        break
    else:
        messages.append({
            "role": "user",
            "content": message
        })
        print(f"User: {message}")
        call_ai()

User: my name is toma
AI: Nice to meet you, Toma! How can I assist you today?
User: what is my name?
AI: Your name is Toma. How can I help you today?

TOOLS = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "A function to get the weather of a city",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {
                        "type": "string",
                        "description": "The name of the city to get the weather of"
                    }
                },
                "required": ["city"]
            }
        }
    }
]

def get_weather(city):
    return f"The weather in {city} is sunny"

FUNCTIONS_MAP = {
    "get_weather": get_weather
}

from openai.types.chat import ChatCompletionMessage

def process_ai_response(message: ChatCompletionMessage):
    if message.tool_calls:
        messages.append({
            "role": "assistant",
            "content": message.content or "",
            "tool_calls": [
                {
                    "id": tool_call.id,
                    "type": "function",
                    "function": {
                        "name": tool_call.function.name,
                        "arguments": tool_call.function.arguments
                    }
                } for tool_call in message.tool_calls
            ]
        })

        for tool_call in message.tool_calls:
            function_name = tool_call.function.name
            function_args = tool_call.function.arguments
            
            print(f"Calling Function: {function_name} with args: {function_args}")
            
            try:
                arguments = json.loads(function_args)
            except json.JSONDecodeError:
                arguments = {}
            
            function_to_run = FUNCTIONS_MAP.get(function_name)

            result = function_to_run(**arguments)
            # function_args는 '{"city": "Seoul"}' 과 같이 생긴 문자열(모델이 주는건 텍스트이기 때문.)
            # 이걸 Python 딕셔너리로 변환하기 위해서 json.loads() 함수를 사용함.
            # get_weather 함수에 전달할 인자는 딕셔너리가 아니라 도시 이름 하나만 받으면 되기 때문에 ** 을 사용.
            # ** 과 같은 표기법은 해당 딕셔너리를 키워드 인자로 바꿔줌 city="Seoul" 와 같은 형태로.

            print(f"Ran function {function_name} with result: {result}")

            messages.append({
                "role": "tool",
                "tool_call_id": tool_call.id,
                "name": function_name,
                "content": result
            })
        
        call_ai()
    else:
        messages.append({
            "role": "assistant",
            "content": message.content
        })
        print(f"AI: {message.content}")

def call_ai():
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=messages,
        tools=TOOLS,
    )
    process_ai_response(response.choices[0].message)

[{'role': 'user', 'content': 'my name is toma'},
 {'role': 'user', 'content': 'my name is toma'},
 {'role': 'assistant',
  'content': 'Nice to meet you, Toma! How can I assist you today?'},
 {'role': 'user', 'content': 'what is my name?'},
 {'role': 'assistant', 'content': 'Your name is Toma.'},
 {'role': 'user', 'content': 'what is the weather in spain'},
 {'role': 'assistant',
  'content': '',
  'tool_calls': [{'id': 'call_D7OM0nlQJPpkNZ5tcP0dZU1c',
    'type': 'function',
    'function': {'name': 'get_weather', 'arguments': '{"city":"Spain"}'}}]},
 {'role': 'tool',
  'tool_call_id': 'call_D7OM0nlQJPpkNZ5tcP0dZU1c',
  'name': 'get_weather',
  'content': 'The weather in Spain is sunny'},
 {'role': 'assistant', 'content': 'The weather in Spain is currently sunny.'}]