OpenAI has released the Agents API in public beta, providing developers with a managed service designed for long-running autonomous agents. The system is built on the open-source Codex harness, which OpenAI previously utilized to scale ChatGPT for Work and internal Codex operations.
The API manages agent context, tool utilization, and subagent coordination. It supports infrastructure capable of maintaining agent operations for several days. Developers can configure environments with three distinct sandbox options or choose to run the service without a sandbox.
A typical session involves four stages: creation, task assignment, progress monitoring via webhooks or streaming, and task steering. The service is currently available to all developers, though data processing is restricted to US servers and does not support Zero Data Retention.
The company demonstrated the API's capabilities by creating an incident-investigation agent capable of delegating error and dependency analysis to subagents. The system supports multi-agent configurations with adjustable concurrency limits for subagents.
