OpenAI Agents API

The Agents API provides a managed Codex harness that lets applications create durable sessions where an agent—defined by model, instructions, tools, and optional MCP servers—runs in an OpenAI-hosted or self-hosted sandbox, executing code, editing files, and generating artifacts. OpenAI handles session orchestration, context compaction, recovery, and billing follows standard model and container rates, while developers configure capabilities, steer execution, and can delegate tasks to up to four concurrent subagents.