Never lose a customer to a missed message
An AI agent trained on your own business, replying in seconds, in any language, on every channel your customers already use.
OpenAI opened public beta access to its new Agents API on September 10, 2026, expanding what outside developers can build with its technology. It puts the infrastructure behind its Codex coding agent directly into developers’ hands through a single API call. The OpenAI Agents API lets outside developers build custom AI agents that can run code, browse the web and coordinate their own subagents. They no longer need to build that scaffolding themselves from the ground up.
What the OpenAI Agents API Actually Does
According to OpenAI’s own announcement, the API brings the same harness and infrastructure that powers Codex to any developer through a flexible interface. Inside a running session, an agent built on the API can execute code, edit files, search the web, apply predefined skills, and produce artifacts. It can also split tasks across subagents up to a set concurrency limit. Those capabilities previously required developers to assemble their own agent infrastructure from scratch.
How the Architecture Is Built
The API is organized around four core objects, as detailed by MarkTechPost’s technical breakdown. An Agent object defines the model, instructions, tools and any connected MCP servers. An optional Environment sandbox handles isolated execution. A durable Session persists state, and a stream of events reports what the session is doing as it works. Developers can run that compute inside an OpenAI-managed sandbox, on their own infrastructure, or through a partner sandbox. Integrations are already live from Cloudflare, DigitalOcean and Oracle.
Why This Matters Beyond Coding Tools
Codex itself remains a specific product aimed at software development. Exposing its underlying harness as a general-purpose API changes that. Developers outside the coding-tools space can now build agents for customer support, research or data analysis. Any workflow that benefits from an AI system capable of taking multi-step action, not just answering questions, is now a candidate. That positions OpenAI more directly against agent-infrastructure offerings from rivals racing to give developers similar building blocks.
What Comes Next in the Beta
The Agents API launched with pay-per-use pricing based on tokens and tool calls. There is no separate subscription layered on top. As the public beta continues, the real test will be adoption. Will third-party developers use the harness at scale for production workloads? And how quickly will OpenAI expand its list of managed-sandbox and infrastructure partners beyond the three announced at launch?
How This Fits the Wider Agent Race
OpenAI is not the only company racing to give developers agent-building infrastructure. Anthropic offers its own agent development kit for building custom Claude-based agents, and Google has been expanding similar tooling around its Gemini models. What sets OpenAI’s move apart is that it exposes the exact harness already proven in a shipping product, Codex. That is different from a separate framework built specifically for external developers. Developers weighing whether to adopt the new API will likely focus on pricing predictability and how easily existing Codex-based workflows can migrate over without a rewrite. OpenAI has not said whether Codex itself will eventually be rebuilt on top of the public Agents API or continue as a separate, parallel product. Infrastructure partners such as Cloudflare and DigitalOcean stand to benefit either way. Developer demand for sandboxed agent compute tends to translate directly into cloud spending, regardless of which underlying model or harness an agent uses.
Frequently Asked Questions
What is the OpenAI Agents API?
It is a public beta, launched September 10, 2026, that gives developers the same harness and infrastructure that powers OpenAI’s Codex coding agent. It is accessible through a single API rather than a standalone product.
What can developers build with it?
Inside a session, the Agents API can run code, edit files, search the web, apply skills and produce artifacts. It can also split work across subagents with a set concurrency limit, according to OpenAI’s own announcement.
How is the API structured?
It is built around four objects. An Agent defines the model, instructions, tools and MCP servers. An optional Environment sandbox handles execution, a durable Session persists state, and a stream of events reports what the session produces as it runs.
Where does the agent’s compute actually run?
Developers can choose to run agent compute in an OpenAI-managed sandbox, their own infrastructure, or a partner sandbox. Integrations are already available from Cloudflare, DigitalOcean and Oracle.
How much does it cost to use?
The Agents API operates on a pay-per-use model based on tokens and tool usage, with no additional flat fee on top of standard API costs.
Is this different from OpenAI’s Codex product?
Codex remains a specific coding-agent product. The Agents API exposes the same underlying harness so developers can build their own custom agents for tasks well beyond coding.
The launch adds to a busy stretch for OpenAI, which has spent September rolling out new capacity, safety governance changes and now a broader developer platform in quick succession. Developers who build on the new API will be watching closely for how OpenAI balances rapid feature releases against reliability. Enterprise customers expect stability from production infrastructure as more companies move agent workloads from experiments into everyday business use. For more on the AI infrastructure race this month, see our coverage of OpenAI’s GPT-6 Astra release and the memory chip shortage squeezing every major AI hardware maker.
Sources
- OpenAI — Introducing the Agents API. https://openai.com/index/introducing-the-agents-api/
- MarkTechPost — 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/
- Blockchain.News — OpenAI Launches Agents API to Streamline AI-Powered Cloud Agents. https://blockchain.news/news/openai-agents-api-launch
Verification your users actually receive.
Send one-time passcodes over WhatsApp with a single API call. Replio can generate, hash and verify the code for you.
