OpenAI Debuts Agents API to Power Autonomous Workflows

OpenAI has unveiled its new Agents API, a managed service designed to give developers direct access to the Codex harness. The API simplifies the creation of autonomous AI agents by handling session persistence, orchestration, context compaction, and recovery. Developers only need to provide the tools and define the execution environment, allowing the AI to perform complex tasks like editing files, running code, and connecting to Model Context Protocol (MCP) servers.

The system is built around four core concepts: agents, environments, sessions, and events. Developers can configure agents to use advanced models like gpt-6-astra and enable multi-agent orchestration. For instance, the API supports delegating independent research tasks to up to 4 concurrent subagents. Agents can operate within OpenAI-hosted sandboxes or self-hosted environments, utilizing capability directories such as workspace skills to execute commands.

In terms of pricing, OpenAI bills model usage at the selected model's standard API rates. Standard rates also apply to OpenAI tools, while OpenAI-hosted sandboxes are billed at standard container rates. Currently, the Agents API only supports data residency within the United States and does not support Zero Data Retention (ZDR), even if developers choose to use a self-hosted sandbox.

This new API enables several practical use cases for developers, such as building automated Slack bots, data analysts that query warehouses with read-only SQL, and GitHub issue investigators that reproduce bugs. By retaining session state across turns, the API allows developers to steer agents continuously without needing to rebuild the entire conversation context from scratch.