OpenAI Opens the Codex Harness to All Developers With Agents API

OpenAI has launched the Agents API in public beta, giving every developer access to the same harness and infrastructure that powers Codex and ChatGPT — at no additional cost beyond the tokens and tools agents consume. The release, available today to all developers, marks the first time OpenAI has productized the internal machinery it built to run agents for millions of Codex and ChatGPT for Work users as a simple, flexible API.

The move matters because agent infrastructure, not raw model capability, has become the bottleneck for production deployments. Long-running agents need context management, tool orchestration and reliable compute environments that survive for days. OpenAI is now selling that entire stack as a service, positioning itself against cloud providers and agent-framework startups that have been building equivalent plumbing.

What can developers build with a single API call?

According to OpenAI, developers can create a production-ready agent in a single API call by specifying four things: the task, the model, the tools and the environment. OpenAI hosts and maintains the harness, while developers choose where the agent's compute runs.

That choice comes in three flavors:

• An OpenAI-managed sandbox, leveraging the same sandboxing infrastructure that powers Codex and ChatGPT
• The developer's own infrastructure
• A sandbox from one of OpenAI's ecosystem partners

"The Agents API gives you a strong foundation for building agents on top of our optimized agent harness and infrastructure, so you can focus on the tools, knowledge, and workflows that make your agent unique," OpenAI said in its announcement.

The company frames this as a division of labor: OpenAI handles the plumbing, developers handle the differentiation.

Who are the sandbox partners?

OpenAI has signed first-class integrations with nine ecosystem providers: Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop and Vercel. The partnerships exist because different workloads demand different compute, storage and deployment options.

The partner integrations cover three dimensions:

• Fully managed environments or deployments within a customer's own VPC
• Specific file and secret storage mechanisms
• Different CPU, GPU and memory configurations, with performance, cold-start and cost profiles matched to a company's workflow

For developers who want speed over configuration, the new OpenAI hosted sandbox provisions and manages a secure environment where agents can run code, work with files and produce artifacts. Developers can configure these sandboxes with their own files, packages, skills and plugins.

What does the harness actually solve?

OpenAI's core argument is that harness engineering wastes developer time. "Taking advantage of new model capabilities often means reworking your harness, taking valuable time away from improving your application," the company said. The Agents API provides versioned access to harness capabilities with each model launch, and OpenAI maintains and continuously improves the harness alongside its models.

Three recent improvements illustrate what the harness handles:

Automatic context compaction. The Agents API automatically compacts earlier context as a session approaches its context limit, preserving information the agent needs to continue. This lets developers build workflows that span multiple context windows without writing their own compaction logic — a persistent pain point for anyone running agents on long tasks.

Tool search and programmatic tool calling. Tool search loads relevant tool definitions as needed, which OpenAI says reduces token usage and cost while preserving the model's cache. Once tools are available, programmatic tool calling lets agents run calls in parallel, chain related operations, and filter or combine results in code — so agents process large volumes of data and bring only relevant results back into context. The API supports MCP, custom functions and built-in tools like web search.

Multi-agent parallelization. With multi-agent support, the API can break complex tasks into independent pieces and delegate them to subagents that work in parallel. Each subagent maintains its own context and stays focused on its assignment, while the main agent coordinates the work and merges results. OpenAI says this speeds up research, analysis and coding tasks without requiring developers to build their own orchestration.

Why does the open-source component matter?

The Agents API runs on the open-source Codex harness, which OpenAI describes as giving developers "visibility into the core logic that coordinates model calls, tools, and context." OpenAI operates and maintains the harness in the API, but developers can inspect and learn from the public codebase on GitHub.

That transparency is a deliberate pitch to enterprise buyers and skeptical developers. Agent behavior is hard to debug when the orchestration layer is a black box, and an inspectable harness lowers the barrier for teams that need to understand exactly how model calls, tools and context interact before deploying agents in production.

How much does it cost?

There are no additional fees for using the Agents API during the public beta. Developers pay only for the tokens and tools their agents use, as outlined on OpenAI's pricing page. That pricing structure signals OpenAI's intent: the company wants the harness to become the default substrate for agent workloads, monetized through model and tool consumption rather than a platform premium.

The launch includes customer testimonials on OpenAI's announcement page, and the company promises rapid iteration: "During the public beta, we'll iterate quickly based on your feedback as we work toward general availability. Let us know what's working, where you're running into friction, and what you need to build and run your agents in production."

What happens next?

The stakes extend beyond OpenAI. The company is betting that the harness layer — compaction, tool orchestration, subagent coordination, sandbox management — becomes the strategic choke point of the agent economy, and it is giving it away at zero marginal cost to capture that position. Rivals building agent platforms and independent harness providers now face a competitor whose infrastructure is battle-tested at the scale of millions of Codex and ChatGPT users and whose core logic is open source.

If the public beta converts developers at the pace OpenAI expects, the general availability release will determine whether the Agents API becomes the standard runtime for production agents — or whether the market fragments across the nine sandbox partners and competing frameworks already in the field.