OpenAI Launches Agents API in Public Beta for Long-Running Cloud Agent Workflows

IT Home reported on September 11 that OpenAI announced the launch of the public beta of its Agents API on September 10 local time, allowing developers to access a cloud-based AI agent execution environment managed by OpenAI via an API.

The service leverages the agent execution framework and infrastructure behind Codex, enabling developers to specify tasks, models, tools, and execution environments to build AI applications capable of handling multi-step tasks.

OpenAI stated that developers can create production-grade agents with just a single API call. Agents can run code in the cloud, process files, store intermediate results, and continuously execute tasks spanning hours or even days. OpenAI maintains the agent execution framework, allowing developers to focus on the tools, knowledge, and workflows needed for their specific applications.

The Agents API supports OpenAI-hosted sandbox environments, as well as deployment on developers' own infrastructure or partner-provided sandboxes. OpenAI has partnered with providers including Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop, and Vercel to offer varied CPU, GPU, memory, and storage configurations.

IT Home noted that OpenAI also introduced the OpenAI Hosted Sandbox service, allowing developers to configure files, software packages, skills, and plugins for agents without building their own sandbox infrastructure. Used for running code, processing files, and generating artifacts, this environment uses the same sandbox infrastructure as Codex and ChatGPT.

For long-running tasks, the Agents API provides context management capabilities that automatically compress earlier information when conversations approach context limits, helping agents continue executing tasks across multiple context windows. The API also supports tool search and programmatic tool calling, allowing agents to load tool definitions on demand, execute calls in parallel, chain operations, or filter results to minimize unnecessary context consumption.

Multi-agent support is another feature of the Agents API. Developers can decompose complex tasks into multiple independent subtasks to be processed in parallel by sub-agents, with a primary agent coordinating and synthesizing the results. In examples provided by OpenAI, developers can configure up to three concurrent sub-agents.

The Agents API is built on the open-source Codex execution framework. OpenAI stated that developers can review the public repository to understand the core logic that orchestrates model calls, tools, and context, while OpenAI will maintain the framework and continually improve agent capabilities alongside model updates.

The Agents API is currently open in public beta to all developers at no additional API fee—users only pay for the model tokens and tools they use. OpenAI noted that it will continue to iterate based on developer feedback throughout the public beta and gradually work toward a general release.