OpenAI Agents API: Codex harness for cloud agents in public beta

On September 10, 2026, OpenAI introduced the Agents API in public beta. It exposes the same harness and infrastructure that powers Codex so developers can create cloud agents through an API: specify the task, model, tools, and environment, then let OpenAI host and maintain the harness.

You choose where the agent runs — an OpenAI-hosted sandbox, your own infrastructure, or partner sandboxes (the post names providers including Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop, and Vercel). The harness covers long-session context compaction, tool search and programmatic tool calling (including MCP, custom functions, and web search), and optional multi-agent subagent delegation. The post says the API is powered by the open-source Codex harness, with no extra fee beyond tokens and tools used.

Optional further reading for CS 4550 / CS 5610 — not required for labs or grades. If you experiment with agent APIs alongside Next.js labs, do not treat agent output as a substitute for understanding the React and Node code you submit.

Read original: Introducing the Agents API — OpenAI

Related in the book: Chapter 1