OpenAI Codex Adds Cloud Environments, Voice CLI and Code Review

OpenAI shipped the broadest Codex update yet at DevDay 2026: the coding agent now runs on your computer, remotely from your phone, or entirely in the cloud from any device, backed by reusable development environments that give whole teams a shared setup with approved settings and permissions. Alongside the runtime expansion, the Codex CLI was rebuilt with voice-controlled task steering and multi-task delegation, a dedicated code-review view landed in the ChatGPT desktop app, and OpenAI released the Codex harness as open source on GitHub.

This piece covers what changed in each surface — cloud, CLI, and review — plus availability by plan and what to verify before moving your team's workflow onto the new pieces. The separate Codex Security Cloud announcement gets its own write-up; this one focuses on the daily development loop.

Codex everywhere: what actually changed

The core shift is that Codex is no longer tied to the machine it was installed on. A task can start on a laptop, be checked from a phone, and finish in the cloud; the environment it runs in is now a first-class, reusable object rather than whatever happened to be on the developer's box. OpenAI says reusable environments help tasks start quickly and give teams a shared setup with settings and permissions that have been approved once, centrally.

That team angle is the operational story: instead of every engineer hand-rolling their own agent sandbox — with whatever access that sandbox happens to have — a lead can define the environment once and everyone inherits it. Cloud Codex is available across Plus, Pro, Business, Healthcare, Education, and Enterprise plans, so the capability is not gated to top tiers.

The revamped CLI: voice, /agents and worktrees

The Codex CLI picked up voice-controlled task initiation and steering — you can talk a task into existence and redirect it as it runs — plus an /agents view for delegating and tracking several tasks at once. Session resuming means an interrupted run can pick up where it left off, and git worktree integration lets parallel tasks work on isolated checkouts of the same repository instead of colliding in one working tree. The CLI updates are available on all plans.

OpenAI also open-sourced the Codex harness on GitHub — the scaffolding that drives the agent loop. For teams building internal tooling around coding agents, that matters more than any single feature: the piece you previously had to reverse-engineer or reimplement is now inspectable and forkable.

Code review inside the ChatGPT desktop app

The new code-review view turns the desktop app into a review surface: it shows changes across multiple projects, summarizes them, lets you walk the diffs, and lets you interrogate Codex about potential issues before feedback goes to GitHub or GitLab. The interesting half is automatic reviews — Codex can scan changes in the background while you are offline, so the review conversation starts with findings already on the table.

Like the CLI, code review is available on all plans. It slots in front of your existing review process rather than replacing it: the output is still feedback a human sends to GitHub or GitLab, which keeps the accountability line where it was while cutting the time from diff to first substantive comment.

Availability, and how this fits the agent push

Every piece announced here is broadly available rather than enterprise-gated: cloud Codex on Plus through Enterprise (including Healthcare and Education), the CLI and code review on all plans. That breadth is deliberate — Codex is also the execution layer for dots, the always-on agents OpenAI launched the same day, and tasks that dots start in Codex count against the same plan usage. The more surfaces Codex covers, the more places an always-on agent can actually do work.

The same keynote also introduced GPT-6.1 Sol — available in Codex today at near-Astra coding quality for a fifth of the price — and an Ultrafast tier that pushes Astra to roughly 300 tokens per second in Codex for Enterprise and Pro 500 subscribers. Faster models plus portable environments plus background review is a coherent bet: shorten every loop in the development cycle at once.

What to verify before rolling it out

Treat shared environments as the piece worth piloting first: they encode security decisions — what the agent can reach, with which permissions — so the approval step deserves the same scrutiny as CI credentials. Check how reusable environments interact with your secrets management before pointing them at production repositories, and confirm how session resume and worktrees behave in your monorepo layout before making the CLI the default entry point for the team.

Watch the open-sourced harness repository for how quickly external contributions land, whether automatic background reviews stay reliable at scale, and how the review view handles very large diffs. One caveat from launch day itself: at least one prominent developer reported a rough Codex Cloud experience hours before the keynote, a reminder that the cloud path is new — keep local Codex as a fallback while the reliability record accumulates.