Agents API (Public Beta)

An OpenAI API for building cloud agents on the same harness that runs Codex, with OpenAI running the agent loop for you. For developers with API access.

Details

OpenAI runs the agent loop on its infrastructure: model calls, tool use, subagent coordination, context compaction, and recovery after failures
You decide what the agent can do and where it runs code and works with files, either in an OpenAI-hosted sandbox or your own environment
Agents can run code, edit files, connect to MCP servers, search the web, hand work to subagents, and resume earlier sessions
No separate fee: usage is billed at standard model, tool, and hosted sandbox rates
Public beta. US data residency only, and not available under zero-data-retention agreements