Introducing the Agents API
OpenAI launches the Agents API in public beta, providing a managed harness and infrastructure for building production-ready cloud agents with tools, environments, and multi-agent support.
OpenAI has introduced the Agents API in public beta, offering developers a managed harness and infrastructure to build and run cloud agents. The API simplifies agent creation by allowing a single API call to specify tasks, models, tools, and environments, with OpenAI hosting the harness. Developers can choose compute environments, including OpenAI-managed sandboxes or their own infrastructure, to suit different workloads and deployment needs.
The Agents API integrates with ecosystem providers such as Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop, and Vercel, offering first-class support for varied requirements. OpenAI also provides a hosted sandbox leveraging the same infrastructure behind Codex and ChatGPT, enabling secure and performant environments for agents to run code, work with files, and produce artifacts without additional setup.
Versioned access to model capabilities is included, ensuring agents benefit from continuous improvements to the harness without requiring developers to rebuild their systems. Recent enhancements include context management for longer sessions, automatic context compaction to preserve critical information, and tool search to reduce token usage and costs while maintaining performance.
The API supports MCP, custom functions, and built-in tools like web search, along with multi-agent support for parallel task delegation. Developers can inspect the open-source Codex harness codebase, and the Agents API is available today in public beta with no additional fees beyond token and tool usage costs as outlined on the pricing page.