# What OpenAI's new Agents API means for you

> OpenAI released a public beta letting developers build agents using its Codex tooling.

*OpenAI has opened up the same "harness" that powers Codex and ChatGPT for Work, letting any developer build agents that can run for hours or days.*

By Behzad Hosseini · SuggestedTech
Canonical: https://suggestedtech.com/news/what-openai-s-new-agents-api-means-for-you

If you have ever used an AI coding assistant like Codex and wondered how it manages to keep working on a task for a long time without losing track, that management system just became available to any developer who wants to build their own agents.

## What happened

OpenAI released its Agents API in public beta on 10 September 2026, the same morning it launched GPT-Live-1 and ChatGPT for Financial Services. The company describes it as a way to "build and run cloud agents with the Codex harness, fully managed by OpenAI."

A harness is the scaffolding around an AI model that handles the practical side of running a long task: keeping track of what's been done, summarising older parts of the conversation so the agent doesn't run out of working memory, and calling in extra tools or sub-tasks as needed. Until now, that system powered OpenAI's own products, Codex and ChatGPT for Work. Now developers can plug into the same thing with one API call, specifying the task, the model, the tools and the environment it should run in.

The system has four parts: the Agent itself (model, instructions, tools and connections to other services), an optional sandbox environment, a durable session that can persist over long periods, and a stream of events showing what the agent is doing. Agents built this way can keep working for hours or days.

Sandboxes, the isolated spaces where agents actually run their work, can sit on OpenAI's own infrastructure, in a customer's own cloud, or with partner providers including Cloudflare, DigitalOcean, Oracle and Vercel. OpenAI always runs the harness itself, regardless of where the sandbox lives.

Early customers named by OpenAI include Ciridae, Long Lake, WithCoverage, SafetyKit, Dwelly, Hypha, deepsense.ai and Nash.ai.

## What it means for you

If you're a developer, you can now build agents that use OpenAI's own internal tooling rather than assembling that infrastructure yourself. OpenAI says there are no additional fees for using the Agents API beyond what you already pay for model tokens, tools and any container time hosted on OpenAI's servers.

There are limits for now: data is stored only in the US, and OpenAI does not yet support Zero Data Retention, an option some businesses need for compliance reasons.

For everyday users of ChatGPT or Codex, this doesn't change the products you already use. It does mean more outside apps and services may soon be built on the same agent technology, since the tooling is no longer limited to OpenAI's own products.

## What happens next

The release puts OpenAI in more direct competition with Anthropic's own agent tooling for developers, as both companies push to make the infrastructure behind AI agents into a product in its own right. More details are available in [MarkTechPost's report on the launch](https://www.marktechpost.com/2026/09/10/openai-launches-the-agents-api-in-public-beta-putting-the-codex-harness-behind-one-api-call/).

## Key takeaways

- OpenAI's Agents API launched in public beta on 10 September 2026, free of extra fees beyond usage.
- It gives outside developers the same "harness", the software scaffolding that manages an AI agent's work, that runs Codex and ChatGPT for Work.
- Agents built with it can work for hours or days, using sandboxes hosted by OpenAI or partner cloud providers.

## Sources

- [OpenAI Launches the Agents API in Public Beta, Putting the Codex Harness Behind One API Call](https://www.marktechpost.com/2026/09/10/openai-launches-the-agents-api-in-public-beta-putting-the-codex-harness-behind-one-api-call/) — MarkTechPost, 2026-09-10