OpenAI Overhauls Codex With Persistent Cloud Environments, Voice Controls, and Automated Security

OpenAI Expands Codex Beyond the Local Machine

At its DevDay conference on Tuesday, OpenAI introduced an extensive suite of updates to its software engineering agent, Codex, designed to free the tool from the constraints of local machines. The headline feature is the addition of persistent, reusable cloud development environments that developers can access across multiple devices, including desktop computers and mobile phones.

Previously, Codex handled remote workloads through isolated, disposable sandboxes. With the new persistent architecture, environments can be pre-configured with standardized team permissions and project settings, allowing remote tasks to launch faster and providing engineering teams with a shared workspace in the cloud.

Refreshed CLI, Desktop Code Reviews, Security Cloud, and APIs

OpenAI rolled out updates spanning the Codex developer workflow, code review interfaces, security tooling, and developer APIs:

• Codex CLI Redesign: The command-line interface now supports voice commands, allowing engineers to dictate instructions and initiate coding tasks verbally. It also introduces a /agents dashboard for delegating and monitoring multiple concurrent tasks, alongside a cleaner terminal user interface (TUI) optimized for long sessions, improved prompt editing, session resumption, and Git worktree handling.
• ChatGPT Desktop Code Review: Codex has been integrated into a new review interface within the ChatGPT desktop application. Engineers can review automated summaries, inspect diffs, and query Codex about potential bugs before publishing feedback on GitHub pull requests or GitLab merge requests. The tool can also perform automatic reviews asynchronously while users are away from their computers.
• Codex Security Cloud: A dedicated security toolset enables automated vulnerability scans across full GitHub repositories—triggered manually, on a scheduled cadence, or upon incoming commits. Codex evaluates flagged issues, deduplicates alerts, and generates patches in the cloud while the user's laptop is closed. The service includes direct access to models developed under OpenAI's "Daybreak Blue" cybersecurity initiative without requiring a separate application.
• API Enhancements: OpenAI introduced the Decisions API, a real-time decision-making endpoint powered by "Luna" to answer user-defined questions from predefined sets. The company also upgraded its Agents API, adding computer-use capabilities and native integration with Amazon's Bedrock Managed Agents to run OpenAI agents entirely within Amazon Web Services (AWS) environments.

The Push Toward Autonomous Cloud Agents

The announcements reflect an ongoing transition in AI-assisted software development, shifting from autocomplete assistants inside local text editors toward persistent, autonomous cloud agents. By giving Codex the ability to manage long-running tasks, conduct unattended code reviews, and draft security fixes in the cloud, OpenAI is positioning the agent to operate as an active background collaborator across enterprise infrastructure.