OpenAI Codex Adds Reusable Cloud Environments Across Devices
OpenAI upgrades Codex with reusable cloud workspaces, voice controls, automated code reviews, security scans, and new agent APIs across devices for teams.
Summary
At Dev Day on Tuesday, September 29, 2026, OpenAI made Codex more portable through reusable cloud development environments accessible from computers, phones and the cloud. Unlike isolated remote task sandboxes, the environments are persistent and configurable, promising faster starts and shared team workspaces with approved settings and permissions. A refreshed Codex CLI adds voice-directed tasks, an /agents view for delegating and tracking concurrent work, improved prompt editing, session resumption and worktrees, plus a cleaner terminal interface for long sessions.
Codex now supports code review inside the ChatGPT desktop app, where users can read summaries, inspect changes and ask about potential issues before commenting on GitHub pull requests or GitLab merge requests. It can also perform initial reviews automatically. Codex Security Cloud scans entire GitHub repositories on demand, on schedules or after commits, then investigates findings, removes duplicates and prepares cloud fixes while users are offline. It includes Daybreak Blue cybersecurity models without a separate application.
OpenAI also introduced the Decisions API, which uses Luna to answer user-defined questions from predefined choices. The updated Agents API adds computer use and supports Amazon Bedrock Managed Agents, allowing OpenAI agents to run entirely on AWS.
Positives
- Reusable Codex environments persist across tasks and work from computers, phones or the cloud.
- Codex CLI now accepts voice direction and provides an /agents view for managing multiple delegated tasks.
- ChatGPT desktop integrates Codex summaries, change exploration and automatic initial reviews for GitHub and GitLab workflows.
- Codex Security Cloud scans whole GitHub repositories and prepares fixes even while users are offline.
- Amazon Bedrock Managed Agents can now build OpenAI agents that run entirely on AWS.
Risks & concerns
- Luna returns only predefined answers to user-defined questions through the Decisions API.
- Codex Security Cloud repository scanning is specified only for GitHub, with GitLab scanning support unmentioned.
- Automatic Codex reviews provide an initial pass, leaving users to assess issues before submitting feedback.
- Pricing, general availability and detailed rollout timing were not disclosed.