# openai-python — Official OpenAI Python SDK

> Call the OpenAI REST API from Python 3.9+ with typed request/response models and sync/async clients. Use it as a core SDK for agents and app backends.

## Install

Save as a script file and run:

# openai-python — Official OpenAI Python SDK

> Call the OpenAI REST API from Python 3.9+ with typed request/response models and sync/async clients. Use it as a core SDK for agents and app backends.

## Quick Use

1. Install:
   ```bash
   pip install openai
   ```
2. Run:
   ```bash
   python -c "import openai; print(getattr(openai, '__version__', 'version?'))"
   ```
3. Verify:
   - Make one authenticated request (e.g., list models) and confirm the client handles errors and retries as expected.


---

## Intro

Call the OpenAI REST API from Python 3.9+ with typed request/response models and sync/async clients. Use it as a core SDK for agents and app backends.

- **Best for:** Python developers building LLM apps who want typed clients and a stable upgrade path
- **Works with:** Python 3.9+, OpenAI API, httpx-powered sync/async clients
- **Setup time:** 6 minutes


### Quantitative Notes

- Setup time ~6 minutes (pip install + set API key + one request)
- GitHub stars + forks (verified): see Source & Thanks
- Python requirement: 3.9+ (repo README)


---

## Practical Notes

Keep SDK usage behind a thin adapter in your codebase (one module) so upgrades don’t ripple through your app. Centralize timeouts/retries and log request IDs. For agents, restrict tool schemas and validate all model outputs before execution.

**Safety note:** Never hardcode API keys; use env vars/secret managers and scrub logs.

### FAQ

**Q: Do I need an API key to import it?**
A: No. You can import and build clients without a key; you need a key to call the API.

**Q: Does it support async?**
A: Yes. The library provides synchronous and asynchronous clients.

**Q: How do I manage rate limits?**
A: Use retries/backoff, batch where possible, and log request IDs for debugging.

---

## Source & Thanks

> GitHub: https://github.com/openai/openai-python
> Owner avatar: https://avatars.githubusercontent.com/u/14957082?v=4
> License (SPDX): Apache-2.0
> GitHub stars (verified via `api.github.com/repos/openai/openai-python`): 30,747
> GitHub forks (verified via `api.github.com/repos/openai/openai-python`): 4,777


---

<!-- ZH -->

# openai-python——OpenAI 官方 Python SDK

> 在 Python 3.9+ 中以同步/异步方式调用 OpenAI REST API，并获得类型化的请求/响应模型；适合作为 agent 与后端服务的核心 SDK，并便于集中管理重试/超时与日志。

## 快速使用

1. 安装：
   ```bash
   pip install openai
   ```
2. 运行：
   ```bash
   python -c "import openai; print(getattr(openai, '__version__', 'version?'))"
   ```
3. 验证：
   - Make one authenticated request (e.g., list models) and confirm the client handles errors and retries as expected.


---

## 简介

在 Python 3.9+ 中以同步/异步方式调用 OpenAI REST API，并获得类型化的请求/响应模型；适合作为 agent 与后端服务的核心 SDK，并便于集中管理重试/超时与日志。

- **适合谁（Best for）:** 做 LLM 应用的 Python 开发者，希望使用类型化 client 并保持稳定升级路径
- **兼容工具（Works with）:** Python 3.9+、OpenAI API、基于 httpx 的同步/异步 client
- **安装时间（Setup time）:** 6 分钟


### 量化信息

- 跑通约 6 分钟（pip 安装 + 配 API key + 一次请求）
- GitHub stars + forks（已核验）：见「来源与感谢」
- Python 版本要求：3.9+（仓库说明）


---

## 实战要点

建议把 SDK 使用封装在一层很薄的适配器里（一个模块），避免升级时全局改动。统一配置 timeout/重试并记录 request_id。做 agent 时要收敛工具 schema，并对模型输出做校验后再执行。

**安全提示：** 不要把 API key 写死在代码里；用环境变量/密钥管理，并清理日志中的敏感信息。

### FAQ

**Q: 导入库需要 API key 吗？**
A: 不需要。创建 client/导入不依赖 key；调用 API 才需要。

**Q: 支持异步吗？**
A: 支持，提供同步与异步 client。

**Q: 怎么处理限流？**
A: 用重试/退避策略，能批量就批量，并记录 request_id 便于排障。

---

## 来源与感谢

> GitHub：https://github.com/openai/openai-python
> Owner avatar：https://avatars.githubusercontent.com/u/14957082?v=4
> 许可证（SPDX）：Apache-2.0
> GitHub stars（已通过 `api.github.com/repos/openai/openai-python` 核验）：30,747
> GitHub forks（已通过 `api.github.com/repos/openai/openai-python` 核验）：4,777


---
Source: https://tokrepo.com/en/workflows/openai-python-official-openai-python-sdk
Author: Agent Toolkit