OpenAI Launches Public Beta for Its Managed Cloud-Based Agents API
OpenAI has launched the public beta of its Agents API, enabling developers to build production-grade AI agents that execute code, handle files, and run multi-step tasks in cloud environments. Built on the open-source Codex execution framework, the service offers managed sandboxes with no extra API infrastructure fees beyond standard token and tool usage. This release significantly lowers the complexity of deploying autonomous AI workflows by offloading sandbox infrastructure, code execution safety, and long-running context handling to OpenAI. Furthermore, integrations with partners like Cloudflare, DigitalOcean, and Vercel allow developers to seamlessly scale agents across diverse computing resources. The API features automated context compression to maintain state across hours or days of execution and supports multi-agent architectures with up to three concurrent sub-agents. Developers can use OpenAI Hosted Sandbox or host custom sandboxes with partners like E2B, Modal, and Oracle to adjust compute configurations.
## BACKGROUND
AI agents are system architectures powered by large language models that autonomously interact with tools, execute code, and process multi-step goals over extended sessions. Building agent applications historically required developers to manage secure isolated environments, such as MicroVMs or containers, to safely run untrusted LLM-generated code and manage token limits.