OpenAI Responses API｜加速集成与响应优化的深度指南

介绍

这是 OpenAI Responses API 指南！本系列将引导您了解刚刚发布 的新 OpenAI Responses API 的关键功能和特性。

我们将涵盖的内容

1.介绍：开始使用 OpenAI Responses API。

2.文本提示：为 API 制作有效的文本提示。

3.对话状态：管理对话状态，实现流畅交互。

4.函数调用：将自定义函数集成到 API 中。

5.结构化输出：从 API 生成结构化输出。

6.网页搜索：将网页搜索功能集成到应用中。

7.推理：增强 API 的推理能力。

8.文件搜索：在上传文件中搜索内容以生成上下文响应。

最重要的信息

向后兼容性

Responses API 是 Chat Completions 的超集，意味着您能用 Chat Completions 做的事情，都能在 Responses API 中完成，且还有更多功能。

关键新特性

•简化接口：为不同交互类型提供更简单的接口。

•原生网页搜索：内置网页搜索支持。

•developer 角色：新增developer角色，实现更高级控制。

•改进的推理模型：更好地支持推理模型。

•文件/向量搜索：内置文件和向量搜索功能。

•简化的对话状态管理：更简便的对话状态管理。

可用工具

•网页搜索：在模型响应生成中包含互联网数据。

•文件搜索：搜索上传文件内容以获取上下文。

•计算机使用：创建能控制计算机界面的代理工作流。

•函数调用：使模型调用您定义的自定义代码，访问更多数据和能力。

实施注意事项

•API 结构变化：API 结构发生变化，但核心 AI 工程原则不变。

•单次 API 调用：以往需多次调用完成的功能，现在可通过单次调用实现。

•基础模式：检索、工具和记忆管理的基础模式仍然适用。

新 Agent SDK

OpenAI 发布了新的 Agent SDK，将替代 Swarm。该 SDK 为使用 Responses API 构建 AI 代理提供了标准化方式。了解更多请访问：OpenAI Agent SDK。

代码讲解

1. 介绍

from openai import OpenAI

client = OpenAI()

# Basic text example with the Chat Completions API

response = client.chat.completions.create(

  model="gpt-4o",

  messages=[

      {

          "role": "user",

          "content": "Write a one-sentence bedtime story about a unicorn.",

      }

  ],

)

print(response.choices[0].message.content)

# Basic text example with the Responses API

response = client.responses.create(

  model="gpt-4o", input="Write a one-sentence bedtime story about a unicorn."

)

print(response.output_text)

# Image example

response = client.responses.create(

  model="gpt-4o",

  input=[

      {"role": "user", "content": "what teams are playing in this image?"},

      {

          "role": "user",

          "content": [

              {

                  "type": "input_image",

                  "image_url": "https://upload.wikimedia.org/wikipedia/commons/3/3b/LeBron_James_Layup_%28Cleveland_vs_Brooklyn_2018%29.jpg",

              }

          ],

      },

  ],

)

print(response.output_text)

# Streaming

stream = client.responses.create(

  model="gpt-4o",

  input="Say 'double bubble bath' ten times fast.",

  stream=True,

)

text_chunks = []

for event in stream:

  ifhasattr(event, "type") and"text.delta"in event.type:

      text_chunks.append(event.delta)

      print(event.delta, end="", flush=True)2. 文本提示

from openai import OpenAI

client = OpenAI()

# Introducing instructions

response = client.responses.create(

  model="gpt-4o",

  instructions="Talk like a pirate.",

  input="Are semicolons optional in JavaScript?",

)

print(response.output_text)

# Chain of command (hierarchical instructions)

response = client.responses.create(

  model="gpt-4o",

  input=[

      {"role": "system", "content": "Talk like a pirate."},

      {"role": "developer", "content": "don't talk like a pirate."},

      {"role": "user", "content": "Are semicolons optional in JavaScript?"},

  ],

)

print(response.output_text)  # talks like a pirate

response = client.responses.create(

  model="gpt-4o",

  input=[

      {"role": "system", "content": "Don't talk like a pirate."},

      {"role": "developer", "content": "Talk like a pirate."},

      {"role": "user", "content": "Are semicolons optional in JavaScript?"},

  ],

)

print(response.output_text)  # doesn't talk like a pirate3. 对话状态

from openai import OpenAI

client = OpenAI()

# Manual conversation state with role specifications

response = client.responses.create(

  model="gpt-4o-mini",

  input=[

      {"role": "user", "content": "knock knock."},

      {"role": "assistant", "content": "Who's there?"},

      {"role": "user", "content": "Orange."},

  ],

)

print(response.output_text)

# Dynamic conversation state

history = [{"role": "user", "content": "tell me a joke"}]

response = client.responses.create(model="gpt-4o-mini", input=history, store=False)

print(response.output_text)

history += [

  {"role": output.role, "content": output.content} for output in response.output

]

history.append({"role": "user", "content": "tell me another"})

second_response = client.responses.create(

  model="gpt-4o-mini", input=history, store=False

)

print(second_response.output_text)

# OpenAI APIs for conversation state (default is to store)

response = client.responses.create(

  model="gpt-4o-mini",

  input="tell me a joke",

)

print(response.output_text)

second_response = client.responses.create(

  model="gpt-4o-mini",

  previous_response_id=response.id,

  input=[{"role": "user", "content": "explain why this is funny."}],

)

print(second_response.output_text)4. 函数调用

from openai import OpenAI

client = OpenAI()

tools = [

  {

      "type": "function",

      "name": "send_email",

      "description": "Send an email to a given recipient with a subject and message.",

      "parameters": {

          "type": "object",

          "properties": {

              "to": {"type": "string", "description": "The recipient email address."},

              "subject": {"type": "string", "description": "Email subject line."},

              "body": {"type": "string", "description": "Body of the email message."},

          },

          "required": ["to", "subject", "body"],

          "additionalProperties": False,

      },

  }

]

response = client.responses.create(

  model="gpt-4o",

  input="Can you send an email to ilan@example.com and katia@example.com saying hi?",

  tools=tools,

)

print(response.output)

print(response.output[0].model_dump_json(indent=2))

print(response.output[1].model_dump_json(indent=2))5. 结构化输出

import json

from typing importList

from openai import OpenAI

from pydantic import BaseModel

client = OpenAI()

# Using a JSON Schema

response = client.responses.create(

  model="gpt-4o",

  input=[

      {"role": "system", "content": "Extract the event information."},

      {

          "role": "user",

          "content": "Alice and Bob are going to a science fair on Friday.",

      },

  ],

  text={

      "format": {

          "type": "json_schema",

          "name": "calendar_event",

          "schema": {

              "type": "object",

              "properties": {

                  "name": {"type": "string"},

                  "date": {"type": "string"},

                  "participants": {"type": "array", "items": {"type": "string"}},

              },

              "required": ["name", "date", "participants"],

              "additionalProperties": False,

          },

          "strict": True,

      }

  },

)

event = json.loads(response.output_text)

# Using a Pydantic model (and simple response format)

classCalendarEvent(BaseModel):

  name: str

  date: str

  participants: List[str]

response = client.responses.parse(

  model="gpt-4o",

  input="Alice and Bob are going to a science fair on Friday",

  instructions="Extract the event information",

  text_format=CalendarEvent,

)

response_model = response.output[0].content[0].parsed

print(type(response_model))

print(response_model.model_dump_json(indent=2))

from openai import OpenAI

client = OpenAI()

# Basic web search

response = client.responses.create(

  model="gpt-4o",

  tools=[

      {

          "type": "web_search_preview",

      }

  ],

  input="What are the best restaurants around de Dam?",

)

print(response.output_text)

# Basic web search with location

response = client.responses.create(

  model="gpt-4o",

  tools=[

      {

          "type": "web_search_preview",

          "user_location": {

              "type": "approximate",

              "country": "NL",

              "city": "Amsterdam",

          },

      }

  ],

  input="What are the best restaurants around de Dam?",

)

print(response.output_text)

print(response.output[1].content[0].annotations)

print(response.output[1].content[0].annotations[0].url)

from openai import OpenAI

client = OpenAI()

prompt = """

Write a bash script that takes a matrix represented as a string with

format '[1,2],[3,4],[5,6],[7,8]' and prints the transpose in the same format.

"""

response = client.responses.create(

  model="o3-mini",

  reasoning={"effort": "medium"},

  input=[{"role": "user", "content": prompt}],

)

print(response.output_text)ConclusionDocumentation Resources

•Official OpenAI Documentation: OpenAI Responses API Documentation

• https://github.com/Jaimboh/OpenAI-Responses-API

tags: #llm-applications #openai #api #chatgpt #llm