OpenAI Launches Agents API in Public Beta, Putting the Codex Harness Behind One Call

OpenAI opened a public beta of its Agents API on September 10, making the same managed harness that powers Codex and ChatGPT for Work available to any developer through a single API call.

Rather than sending a request to a raw model, developers send work to a session and OpenAI runs the agent — handling model calls, tool use, conversation history, context compaction, and recovery. Inside a session, the agent can run code, edit files, search the web, apply skills, produce artifacts, and split work across subagents under a set concurrency limit, executing in an OpenAI-managed sandbox, a developer's own infrastructure, or a partner sandbox.

There's no separate fee for the harness itself — usage is billed as model tokens, tool calls, and sandbox minutes — effectively packaging OpenAI's internal agent infrastructure as a product any developer can build on directly.

Photo via Wikimedia Commons, © Markus Spiske, released under CC0.