OpenAI has launched the Agents API in public beta, giving developers managed access to the agent harness and infrastructure used by Codex. The API is designed for long-running work rather than a single model call: developers specify a task, model, tools and execution environment while OpenAI operates the harness.
The execution layer is deliberately separable from the harness. Agents can run in an OpenAI-hosted sandbox, on a developer's own infrastructure, or through supported sandbox partners. That makes the release more than another tool-calling endpoint: it packages context management, orchestration and recovery while leaving teams some control over where code and files are processed.
For longer jobs, the API automatically compacts earlier context as sessions approach their context limit. It also includes tool search, programmatic tool calling and multi-agent support, allowing a coordinating agent to delegate independent work to subagents in parallel. MCP, custom functions and built-in tools such as web search are supported.
OpenAI says the service is built on the open-source Codex harness. The company is not charging a separate Agents API fee during the beta; developers pay for the models, tools and other metered resources they use. Customer performance figures in OpenAI's launch material are vendor-selected testimonials rather than independent benchmarks, so AiToolMap does not treat them as general performance evidence.
The practical significance is architectural. Teams that previously had to build their own durable agent loop can now buy that orchestration layer as a managed service while retaining choices around execution infrastructure. The public beta is available to developers now, with OpenAI saying it will iterate toward general availability based on feedback.