OpenAI has released the Agents API in public beta, exposing the managed Codex harness that powers long-running agents as a single API call. The service allows developers to create durable agent sessions with built-in context compaction, tool search, and multi-agent coordination.
- The API manages agent state, including sessions, turns, and items, reducing integration effort compared to the Agents SDK or Responses API.
- Agents can run in an OpenAI-hosted sandbox, a self-hosted environment via `codex exec-server`, or one of nine partner sandboxes like Cloudflare and Vercel.
- Built-in features include automatic context compaction, programmatic tool calling for parallel execution, and subagent delegation with independent contexts.
- Early customer results report metrics such as a 4x latency reduction for Ciridae and 60% lower cost per case for SafetyKit.
The API is available to all developers in public beta with US-only data residency, though Zero Data Retention is not supported.