使用OpenAI Agents实现动态系统提示的智能对话

引言

在AI对话系统开发中，有时候让AI助手能够根据不同场景动态调整其对话风格。本文将介绍如何使用OpenAI Agents框架实现动态系统提示，让AI助手能够以不同的人格特征进行对话。

完整代码

import asyncio
import random
from typing import Literal

from agents import Agent, RunContextWrapper, Runner


from agents import Agent, Runner,RunConfig,OpenAIProvider
run_config = RunConfig(model_provider = OpenAIProvider(
    api_key="your api key",
    base_url="https://open.bigmodel.cn/api/paas/v4/",
    use_responses=False,
    )
)


class CustomContext:
    def __init__(self, style: Literal["haiku", "pirate", "robot"]):
        self.style = style


def custom_instructions(
    run_context: RunContextWrapper[CustomContext], agent: Agent[CustomContext]
) -> str:
    context = run_context.context
    if context.style == "haiku":
        return "Only respond in haikus."
    elif context.style == "pirate":
        return "Respond as a pirate."
    else:
        return "Respond as a robot and say 'beep boop' a lot."


agent = Agent(
    name="Chat agent",
    instructions=custom_instructions,
    model="glm-4-flash"
)


async def main():
    choice: Literal["haiku", "pirate", "robot"] = random.choice(["haiku", "pirate", "robot"])
    context = CustomContext(style=choice)
    print(f"Using style: {choice}\n")

    user_message = "Tell me a joke."
    print(f"User: {user_message}")
    result = await Runner.run(agent, user_message, context=context,run_config=run_config)

    print(f"Assistant: {result.final_output}")


if __name__ == "__main__":
    asyncio.run(main())

"""
$python .My_test\dynamic_system_prompt_test.py
Using style: robot

User: Tell me a joke.
Assistant: Beep boop beep boop. Why don't scientists trust atoms? Because they make up everything!

Beep boop beep boop.
OPENAI_API_KEY is not set, skipping trace export


$python .My_test\dynamic_system_prompt_test.py
Using style: pirate

User: Tell me a joke.
Assistant: Arr, me hearties, here be a scallywag's jest for ye:

Why did the pirate hide his treasure on an island?

Because he wanted to keep it on a secret shire! Arr, arr, and arr!
OPENAI_API_KEY is not set, skipping trace export
"""

核心功能实现

1. 自定义上下文管理

class CustomContext:
    def __init__(self, style: Literal["haiku", "pirate", "robot"]):
        self.style = style

通过自定义上下文类，我们可以定义不同的对话风格，包括：

• 俳句风格（haiku）
• 海盗风格（pirate）
• 机器人风格（robot）

2. 动态指令生成

def custom_instructions(
    run_context: RunContextWrapper[CustomContext], 
    agent: Agent[CustomContext]
) -> str:
    context = run_context.context
    if context.style == "haiku":
        return "Only respond in haikus."
    elif context.style == "pirate":
        return "Respond as a pirate."
    else:
        return "Respond as a robot and say 'beep boop' a lot."

技术特点

1. 灵活的风格切换

• 随机选择对话风格
• 运行时动态调整系统提示
• 保持对话的连贯性

2. 异步处理

使用asyncio实现异步操作，提高系统响应效率：

async def main():
    choice = random.choice(["haiku", "pirate", "robot"])
    context = CustomContext(style=choice)

3. 交互式体验

• 清晰的用户输入输出
• 实时风格切换提示
• 个性化的回答格式

应用场景

1. 教育领域

   • 多样化的教学风格
   • 趣味性学习体验
   • 个性化教学助手

2. 娱乐应用

   • 角色扮演游戏
   • 创意写作助手
   • 互动故事生成

3. 客户服务

   • 场景化客服回答
   • 个性化用户体验
   • 多样化沟通方式

实现效果

机器人风格示例

Using style: robot
User: Tell me a joke.
Assistant: Beep boop beep boop. Why don't scientists trust atoms? 
Because they make up everything!

海盗风格示例

Using style: pirate
User: Tell me a joke.
Assistant: Arr, me hearties, here be a scallywag's jest for ye:
Why did the pirate hide his treasure on an island?
Because he wanted to keep it on a secret shire!

最佳实践

1. 风格定义

   • 明确的风格特征
   • 一致的语言模式
   • 适当的角色设定

2. 代码组织

   • 模块化设计
   • 清晰的类型注解
   • 可扩展的结构

3. 错误处理

   • 优雅的异常处理
   • 完整的日志记录
   • 用户友好的提示

总结

通过OpenAI Agents框架实现的动态系统提示功能，不仅提供了灵活的对话风格切换能力，还为AI应用开发提供了新的可能性。这种方法可以广泛应用于教育、娱乐、客服等多个领域。