What’s new in OpenAI’s Codex desktop app?

OpenAI updates Codex desktop with computer control and browsing

OpenAI has rolled out a major update to its Codex desktop app, expanding it from an AI coding tool into a more capable “agentic” desktop assistant.

The update includes the ability to control a user’s computer, letting Codex take actions in the local environment rather than only generating code for someone to run. It also adds an in-app browser, so the assistant can navigate and use web content while working.

Key additions

• Computer control for running steps directly on the desktop
• In-app browser to access web resources during tasks
• Image generation capabilities inside the app
• Automation memory to retain context across actions
• Plugin support to extend Codex with additional functionality

Why it matters

This matters because it shifts Codex toward the pattern users now associate with modern software agents: the system can plan and execute multi-step work in a real environment, not just output text. Adding both computer control and browsing increases the range of tasks Codex can attempt—especially workflows that mix coding, research, and iterative refinement.

It also places OpenAI more directly in a competitive landscape of desktop coding and agent tools. Several other tools in the market focus on letting AI do more of the operational work, and Codex’s new features are a bid to become a “workbench” that can drive the browser and the user’s computer from one interface.

The summary available here doesn’t specify limits (such as sandboxing boundaries, what permissions users must grant, or which plugins are currently supported), but the overall direction is clear: Codex is being upgraded to do end-to-end work from the desktop rather than acting only as a code generator.