OpenAI gives Codex reusable cloud environments that work across devices

OpenAI announced a suite of updates to its Codex software engineering agent at DevDay, emphasizing more persistent cloud-based development workflows. Rather than treating each remote task as an isolated sandbox, Codex can now create reusable cloud development environments intended to start faster and provide shared workspaces with preapproved settings and permissions. OpenAI said these environments can be used from a computer, a phone, or in the cloud.

The Codex command-line interface received a refresh that includes voice controls for initiating and directing tasks. OpenAI also added a new /agents view to help developers delegate and monitor multiple concurrent tasks, and introduced interface improvements for editing prompts, resuming sessions, working with worktrees, and a cleaner terminal UI aimed at improving readability during long sessions.

For code review workflows, Codex is being integrated into the ChatGPT desktop app so users can get summaries, explore diffs, and query Codex about potential issues before submitting feedback to GitHub pull requests or GitLab merge requests. OpenAI also described an automatic code review capability that can run initial reviews while users are away from their machines.

Security tooling is another focus of the release: Codex Security Cloud can scan entire GitHub repositories on demand, on a schedule, or when new commits arrive. The system groups findings, removes duplicates, and can prepare cloud-based fixes even if a developer's laptop is offline. The product includes access to models from OpenAI’s cybersecurity initiative Daybreak Blue without requiring a separate application. OpenAI additionally rolled out API updates, including a real-time Decisions API powered by Luna for answering predefined questions, and an updated Agents API that supports local computer use and Amazon’s Bedrock Managed Agents for running OpenAI agents on AWS.