OpenAI推出了Agents API，为开发者提供了一套完整的托管解决方案来构建和部署持久的云端AI代理1。该API承担了会话管理、编排、上下文压缩和任务恢复等核心功能，允许应用程序定义工具集并选择执行环境1。代理可在OpenAI托管或自托管的沙箱环境中运行，支持代码执行、文件编辑以及MCP服务器连接等能力1。
该API支持多代理并发运行，最多允许4个并发子代理，实现复杂的任务委派和协作1。开发者可通过配置模型、指令、工具和MCP服务器来定制代理行为1。定价结构基于底层模型的API费率、OpenAI工具的标准费率以及容器运行费率1。目前该服务仅支持美国数据驻留，不支持零数据保留（ZDR）选项1。OpenAI为开发者提供了Python、JavaScript、Go、Java、Ruby和cURL等多种编程语言的SDK支持1。
OpenAI has introduced the Agents API, a new service enabling developers to build persistent cloud-based AI agents using a managed Codex framework 1. The API handles critical infrastructure tasks including session management, orchestration, context compression, and task recovery, while applications provide the tools and select their execution environment 1.
The Agents API supports configuration of models, instructions, tools, and MCP servers 1. Agents can execute code, edit files, and connect to MCP servers while running in either OpenAI-hosted or self-hosted sandbox environments 1. The platform enables multi-agent functionality, supporting up to four concurrent sub-agents for task delegation and parallel execution 1. Pricing is based on model API rates, standard rates for OpenAI tools, and container fees 1.
The service currently operates with data residency limited to the United States and does not support zero data retention options 1. OpenAI provides multi-language SDK support including Python, JavaScript, Go, Java, Ruby, and cURL 1.
评论
还没有评论，欢迎留下第一条。