OpenAI Agents API - Cloud agents, run on OpenAI's Codex harness
by•
OpenAI's managed version of the Codex harness to build cloud agents with one API call instead of your own orchestration. Handles long sessions, smart tool use, and subagents working in parallel. Runs on OpenAI or partner sandboxes. No extra fees beyond token and tool usage.
Open source harness, public beta now.
Replies
Agents API is OpenAI's managed version of the Codex harness, the same orchestration and infra behind Codex and ChatGPT for Work, now available as an API. One call creates a production agent, specifying model, tools, and environment, OpenAI handles the orchestration.
What makes it different: It's versioned alongside every model launch, so your harness improves automatically instead of you reworking it each time. The harness itself is open source too, you can inspect exactly how it coordinates model calls, tools, and context.
Key features:
Auto-compacts context so agents can work across long sessions without hitting limits
Tool search loads only relevant tool definitions, cutting token usage
Programmatic tool calling runs calls in parallel and filters results in code
Multi-agent support lets a main agent delegate work to subagents running in parallel
Choice of environment: OpenAI-hosted sandbox, your own infra, or partner sandboxes (Modal, Cloudflare, Daytona, Blaxel, Runloop, Vercel, Oracle, E2B, DigitalOcean)
Who it's for: developers building production agents (research, coding, ops, support) who don't want to build and maintain their own orchestration layer.
Try it at OpenAI Developers · Agents API
P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends
This is cool as heck. Two questions:
Is this billed purely via API billing, or can usage from these agents flow through subscriptions?
Is the eventual goal to offer 24/7 cloud agents inside the ChatGPT app (or your environment of choice) similar to the Grok Bot app + Grok Bots? Or am I misunderstanding?