OpenAI Agents SDK
last releaseApril 15, 2026
goblin vibe check:
good if you want codex-style agent behavior in your own stack without soldering the runtime together yourself
model-native agent harness for building long-running agents with sandbox execution, filesystem tools, shell access, memory, and portable manifests. useful when you want openai-flavored agent infrastructure without hand-rolling the execution layer yourself.
native sandbox execution for files, commands, and long-horizon workcodex-like filesystem tools plus shell, apply-patch, skills, and mcp supportmanifest abstraction for portable workspaces across sandbox providersdurable execution with snapshotting and rehydration support
key features
native sandbox execution for files, commands, and long-horizon workcodex-like filesystem tools plus shell, apply-patch, skills, and mcp supportmanifest abstraction for portable workspaces across sandbox providersdurable execution with snapshotting and rehydration support
spec & usage
official update published april 15, 2026
launches first in python with typescript support planned later
works with built-in support for providers including blaxel, cloudflare, daytona, e2b, modal, runloop, and vercel
limitations
strongest fit if you already want to build around openai models and primitives
scope:
codeworkflowwrapperautomationagentapicloudbusiness
launchApril 15, 2026
last releaseApril 15, 2026