> ## Content Index
> Fetch the complete content index at: https://futureinuse.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# OpenAI opens up the Agents API: the same structure behind Codex, now available to agent builders
- URL: https://futureinuse.ai/openai-agents-api/
- Published: 2026-10-14T12:21:54.000Z
- Updated: 2026-10-19T19:50:37.000Z
- Description: In public beta since September 10, the API delivers the harness and infrastructure behind Codex. For managers, it changes the size of the request you can bring to your technical team.
- Author: Rafael Cordeiro
- Tags: AI in News

On September 10, 2026, OpenAI announced the Agents API in public beta for all developers. According to the company, the API gives anyone building agents the same harness and infrastructure that power Codex, OpenAI's coding agent.

## What the harness is

In OpenAI's vocabulary, the harness is the layer that organizes an agent's work, managing context, using tools and coordinating subagents. The company says useful agents need that layer plus infrastructure that keeps them running for days, with environments where they can work with files, run code and save intermediate results.

With the Agents API, the developer describes the task, the model, the tools and the environment, and OpenAI hosts and maintains the harness. According to the announcement, the harness is open source, which means you can inspect the logic that coordinates calls to the model, tools and context.

## Where the agent runs

The execution environment is the builder's call. OpenAI lists three paths, a sandbox managed by OpenAI itself, the builder's own infrastructure, or one of the sandbox partners named in the announcement, such as Cloudflare, Modal, Oracle and Vercel. The options mentioned include fully managed environments or deployments inside the customer's private network, along with different CPU, GPU and memory configurations.

The OpenAI-hosted sandbox, according to the company, runs on the same infrastructure as Codex and ChatGPT and can be configured with files, packages, skills and plugins.

## Three harness features

The announcement highlights recent improvements.

- **Long sessions.** The API automatically compacts earlier context as a session nears its limit, keeping what the agent needs to carry on.
- **Tool use.** A search loads tool definitions as they are needed, and the agent can make parallel calls and filter results in code. The API supports MCP, custom functions and built-in tools such as web search.
- **Subagents.** Complex tasks can be split up and delegated to subagents working in parallel, each with its own context, while the main agent pulls the results together.

OpenAI says it maintains the harness alongside the models and offers versioned access with each release, so agents pick up improvements without developers having to rebuild that layer.

## Cost and status

According to the company, there is no extra fee for using the Agents API, and you pay for the tokens and tools your agents use according to OpenAI's pricing. The product is in beta, and OpenAI says it will iterate on developer feedback until general availability, with no date announced.

## What to bring to the conversation with your technical team

The announcement is aimed at developers, but it affects anyone who commissions internal agents. Four questions help you assess a request.

- **What task the agent will complete end to end**, and who checks the result.
- **Where it will run**, given that company data may pass through the chosen environment.
- **Which tools and data it can access**, and with what permissions.
- **How cost will be tracked**, since billing is by the tokens and tools used, and a long task can consume more than expected.

Since the product is in beta, treat the first agents as pilots with limited scope and human review of the output.

---

## Sources

- OpenAI, "Introducing the Agents API", September 10, 2026.

*Future in Use is an editorial initiative linked to OperOS AI.*