OpenAI introduced the Agents API, a public-beta managed API for building and running long-running cloud-based AI agents
Feature update ● Confirmed 92% confidence first seen
OpenAI launched the Agents API in public beta, providing a managed interface for orchestrating long-running agent sessions with tools and other capabilities powered by Codex harness. The coverage indicates developers can access the API broadly while OpenAI handles orchestration concerns such as context management and multi-subagent coordination, with notes about data handling constraints (e.g., US-only and no zero data retention support).
Decision brief
- What changed
- OpenAI introduced the Agents API in public beta, a managed API for building and running long-running cloud-based AI agent sessions powered by the Codex harness. The API is broadly available to developers and shifts orchestration tasks such as context management, tool use, recovery, and multi-subagent coordination to OpenAI’s managed service.
- Why it matters
- This gives product and engineering leaders a new managed option for deploying long-running agent workflows without building all orchestration layers themselves, which could reduce internal development effort for agent infrastructure. It also creates a clearer build-vs-buy decision around agent runtime management, but with operational constraints that matter for enterprise use cases, including US-only data handling and lack of Zero Data Retention support in the reported coverage.
- Evidence
- OpenAI’s own blog announced the Agents API as a managed service powered by Codex harness for launching cloud agents. Two secondary outlets, MarkTechPost and TLDR Dev, independently and consistently described it as a public-beta API for long-running agents, with OpenAI handling orchestration functions; MarkTechPost specifically reported the US-only data scope and lack of Zero Data Retention support.
- What remains uncertain
- The coverage does not establish pricing, reliability, performance, security controls beyond the cited data-handling limits, or how production-ready the public beta is for large-scale enterprise workloads. It is also unclear from the provided reporting which deployment modes have feature parity and whether the cited hosting options change compliance, control, or operational responsibilities.
- Monitor next
- Watch for OpenAI’s next product and documentation updates on GA timing, pricing, regional availability, and enterprise data-retention/compliance support for the Agents API.
Analytical support, not advice — assumptions and open questions stated above.