Introducing the Agents API

OpenAI has introduced the Agents API, designed to facilitate the creation of long-running agents. This API provides a robust harness for managing context, efficiently utilizing tools, and coordinating subagents. It also offers infrastructure for reliable operation over extended periods, enabling agents to work with files, run code, and save intermediate results. Key features include Tool search for reduced token usage and programmatic tool calling for parallel operations and data processing. The Agents API supports MCP, custom functions, and built-in tools like web search, with a public beta currently underway.

Unlike previous approaches, OpenAI's new Agents API specifically addresses the challenges of building long-running agents by providing dedicated infrastructure for context management, tool use, and subagent coordination.