At DevDay on September 29, 2026, OpenAI announced reusable cloud environments for Codex: what used to be isolated sandboxes become persistent workspaces, configured once and reusable for every new task. The practical novelty is that the work keeps going even when the laptop is closed.

How they work

A developer prepares an environment from a GitHub project, with its repository, tools and dependencies, and publishes it. Codex analyzes the repository, detects dependencies and drafts the install script; any changes are made in conversation with Codex itself.

From then on each task:

  • runs on a dedicated virtual machine
  • keeps working while the computer sleeps, for example investigating results, removing duplicates and preparing fixes
  • can be resumed from desktop, web or phone, because changes not yet sent to the repository stay with the task

For credentials, the environment supports variables and network secrets: a proxy inserts the real value only when traffic heads to a domain on the allow list.

The other DevDay news for Codex

  • A revamped CLI, with voice commands to start and direct tasks
  • A new code review experience in the ChatGPT desktop app, to read summaries and explore changes on GitHub and GitLab
  • Codex Security Cloud, which scans GitHub repositories on demand or on a schedule and prepares fixes even while the user is offline
  • GPT-6.1 Sol as an available model in Codex, as covered in our article on GPT-6.1 Sol

Who can use it

PlanAccess to cloud environments
ChatGPT Plus and ProYes, with reduced resources (half the processors and memory compared with higher tiers)
Business, Enterprise, Edu, HealthcareYes, with full access

Today's limits

The service launches with some clear boundaries:

  • no GitLab and no self-hosted GitHub Enterprise installations in cloud environments
  • no browser-based coding or computer use inside cloud environments
  • a task recovery window of up to seven days

Why it matters

With dots OpenAI brings always-on agents to a general audience; with Codex Cloud it does the same for people who write software. The thread is the same: AI that does not wait in front of the screen, but works in parallel and returns a result to review. For a small team it means being able to leave fixes, tests and security scans in a queue and find them ready, as long as human review stays on everything that goes into production.

Read also: dots, ChatGPT's always-on agents.

← All articles