OpenAI launches Agents API in public beta for developers

OpenAI released its Agents API in public beta on Thursday, providing developers access to the infrastructure that powers its Codex and ChatGPT products through a single API call.

The API enables developers to build cloud-based agents by defining a task, model, tools, and compute environment. OpenAI handles the underlying harness while developers select where the agent operates, including OpenAI-hosted sandboxes, their own infrastructure, or third-party sandbox partners.

The company formed partnerships with Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop, and Vercel to offer compute environment integrations. These options include fully managed environments, virtual private cloud deployments, and various CPU, GPU, and memory configurations.

Technical features include automatic context compaction for extended sessions, tool search to reduce token usage, programmatic parallel tool calling, and multi-agent support that lets a primary agent assign subtasks to subagents running at the same time.

Jack Weissenberger, CTO of Ciridae, said in a statement that using the API raised the company’s evaluation score to 0.85 from 0.71 and produced a 4x latency reduction for subagent workflows.

The Agents API uses an open-source version of the Codex harness, which developers can review through OpenAI’s public codebase. OpenAI states no additional fees apply for using the API itself, with developers paying only for token and tool usage at standard rates.

The company plans to refine the API based on developer feedback before moving toward general availability.