OpenAI’s Agents API gives developers the infrastructure behind Codex and ChatGPT

By NewSunSEO | September 16, 2026

Developers can now build cloud-based agents on the same infrastructure that runs Codex and ChatGPT, with OpenAI’s new Agents API entering public beta. The API lets agents run for hours at a time, execute code, and process files, opening a path to production-ready autonomous workflows for any team willing to call the endpoints.

What the Agents API does

The Agents API is a developer surface for building agents that operate in persistent cloud sessions rather than short single-turn calls. Three capabilities stand out:

• Long-running execution. Agents can run for hours, which makes tasks like multi-step research, code refactors, and large file processing feasible without custom orchestration.
• Automatic context management. The system handles context windows on the developer’s behalf, so an agent that reads thousands of pages or processes a long repository does not require hand-tuned truncation.
• Parallel tool use and sub-agent delegation. Agents can call several tools at once and delegate parts of a job to other agents, which lets a single program behave more like a coordinated team than a single thread.

Where the agents actually run

Developers pick the execution environment. OpenAI offers hosted sandboxes, and partners including Cloudflare, Vercel, and Oracle are also supported. Running code closer to users or inside an existing enterprise stack is now a configuration choice rather than a build-it-yourself project.

There are no extra fees for the sandbox or the partner environments. Billing is based solely on token usage, which keeps cost forecasting tied to model consumption instead of seat counts or compute hours.

What is already wired in

The API runs on the open-source Codex tool and includes several integrations by default.

• Model Context Protocol (MCP) support for connecting external tools and data sources.
• Custom functions so a developer can expose any internal API as a tool the agent can call.
• Built-in tools including web search, so an agent can fetch fresh information without a separate retrieval pipeline.
• File processing for workflows that need to read, transform, or write documents.

Why the same infrastructure matters

Until now, the agent patterns shipping inside Codex and ChatGPT were a closed product surface. Putting them behind a public API means a startup, an enterprise team, or an independent developer can build the same class of autonomous workflows that OpenAI itself runs internally. The hours-long runtime and the sub-agent delegation model are the practical differences: most agent demos today run for minutes and operate as a single loop, and that ceiling is the main thing the new API raises.

How to get started

The Agents API is documented at developers.openai.com, where the overview and the guide to OpenAI-hosted sandbox environments walk through authentication, environment selection, and tool registration. Because Codex uses the same underlying setup, existing Codex configurations can carry over with light changes, lowering migration costs for teams already running agents on the Codex side.

FAQ

What is OpenAI’s Agents API?

The Agents API is a public-beta developer interface from OpenAI for building cloud-based agents that run for hours, execute code, and process files on the same infrastructure that powers Codex and ChatGPT.

How much does the Agents API cost?

There are no extra fees for using OpenAI-hosted sandboxes or partner environments such as Cloudflare, Vercel, and Oracle. Billing is based solely on token usage.

What can agents built with the API do?

Agents can run for hours, manage context automatically, use tools in parallel, delegate work to sub-agents, call MCP-connected tools and custom functions, perform web search, and process files.