Agents API (September 2026): Build Cloud Agents with Codex

OpenAI has launched the Agents API in public beta, giving developers an API for running cloud agents with the managed harness that powers Codex. The service is designed to handle session orchestration and context management while an application supplies the task, model, tools, and environment.

A managed harness for long-running agents

An application can create an agent by specifying its task, model, tools, and environment. OpenAI manages the Codex harness, including session orchestration, context compaction, recovery, and the coordination of model calls and tools.

The API retains session state so an agent can continue work across turns without rebuilding its context. OpenAI also describes support for steering an active session, streaming progress, and resuming work after an earlier turn. Context compaction is intended to let a job span multiple context windows without requiring every application to implement its own summarization loop.

Sandboxes and tools

Developers can use an OpenAI-hosted sandbox or connect a sandbox from their own infrastructure or a supported provider. In a hosted sandbox, an agent can run code, edit files, connect to MCP servers, and produce artifacts; the environment can include files, packages, skills, and plugins.

The managed harness supports MCP, custom functions, built-in tools such as web search, and programmatic tool calling. Programmatic calls can run operations in parallel, filter and combine results in code, and keep large intermediate datasets out of the model’s context. Multi-agent workflows can split complex tasks among subagents and bring their results back together.

What the beta means for teams

The Agents API is available in public beta to all developers. OpenAI says there is no additional Agents API fee; users pay for the tokens and tools their agents use. The developer documentation lists model and tool rates plus standard container rates for OpenAI-hosted sandboxes.

The API is not a general-purpose data-residency solution: the current documentation says data residency is supported only in the United States and that Zero Data Retention is not supported. Choosing a self-hosted sandbox does not make the Agents API ZDR-eligible. Teams should review those constraints, sandbox permissions, secrets handling, and deletion behavior before moving sensitive workloads into a long-running agent.

September 29 update: computer use in the Agents API

OpenAI’s API changelog adds computer use to the Agents API. Agents can complete tasks in an OpenAI-hosted browser, while the calling application handles website-access approvals and sign-in. OpenAI’s DevDay recap also says the API brings Codex multi-agent capabilities, tool search, tool calling, and context compaction into applications.

This adds a browser-interaction path to the managed-harness and sandbox options described above. Before enabling it, define which sites and actions the application can approve, and keep authentication under the application’s control as the announcement specifies.

Sources

• OpenAI — “Introducing the Agents API” (September 10, 2026): https://openai.com/index/introducing-the-agents-api/
• OpenAI Developers — “Agents API overview”: https://developers.openai.com/api/docs/guides/agents-api/overview
• OpenAI Developers — “Agents API”: https://developers.openai.com/api/docs/guides/agents
• OpenAI Developers — API changelog: https://developers.openai.com/api/docs/changelog
• OpenAI — “DevDay 2026 Recap” (September 29, 2026): https://openai.com/index/devday-2026-recap
• OpenAI Developers on X — official developer updates: https://x.com/OpenAIDevs