OpenAI opens Agents API public beta, exposing the managed Codex harness

OpenAI ships the Agents API in public beta

On September 10, 2026, OpenAI announced the Agents API in public beta. The company frames it as the same harness and infrastructure that powers Codex, delivered through a flexible API so developers can build and run cloud agents without owning the full orchestration stack.

OpenAI says useful long-running agents need a harness that manages context, uses tools efficiently, and coordinates subagents, plus infrastructure that keeps them running reliably with environments for files, code, and intermediate results. The Agents API is meant to supply that foundation while teams focus on tools, knowledge, and workflows.

Managed Codex harness, your environment

OpenAI hosts and maintains the harness. You choose the agent’s compute environment: an OpenAI-managed sandbox, your own infrastructure, or one of OpenAI’s sandbox partners. The announcement lists first-class integrations with Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop, and Vercel.

For a faster start, OpenAI also introduced an OpenAI hosted sandbox that uses the same sandboxing infrastructure behind Codex and ChatGPT. OpenAI provisions and manages that sandbox so agents can run code, work with files, and produce artifacts.

You can create a session in a single API call by specifying the task, model, tools, and environment. OpenAI’s sample configuration shows multi-agent settings such as multi_agent with max_concurrent_subagents.

Context compaction, tool search, MCP, and subagents

Under “Build with an evolving Codex harness,” OpenAI highlights capabilities shipped with the managed stack:

• Long sessions and compaction. The Agents API automatically compacts earlier context as a session approaches its limit, so workflows can span multiple context windows without custom compaction logic.
• Tool search and efficient tool use. Tool search loads relevant tool definitions as needed to help reduce token usage while preserving the model’s cache. Programmatic tool calling can run calls in parallel, chain operations, and filter results in code.
• MCP and built-in tools. The Agents API supports MCP, custom functions, and built-in tools such as web search.
• Parallel subagents. With multi-agent support, the API can break work into independent pieces and delegate to subagents that run in parallel, each with its own context, while the main agent coordinates results. Configuration examples on the page use max_concurrent_subagents.

OpenAI says the API provides versioned access to harness improvements with each model launch, so teams spend less time reworking orchestration when models change.

Open-source foundation and pricing note

The Agents API is powered by the open-source Codex harness. OpenAI operates and maintains that harness in the API while developers can inspect the public codebase. The company states the API is available in public beta to all developers, with no additional fees for using the Agents APIyou pay for the tokens and tools agents use, as listed on OpenAI’s pricing page.

During the beta, OpenAI plans to iterate toward general availability based on developer feedback. For setup details, see the Agents API overview and quickstart linked from the announcement.

Source: Introducing the Agents API (OpenAI, September 10, 2026)