OpenAI Codex for Mac adds screen control, browser, image tools

OpenAI announced an update to its Codex desktop app for macOS that adds background computer control, an in-app browser, integrated image generation and more than 90 plugins for developer workflows. The rollout begins today for users who sign in with a ChatGPT account; some personalization and computer-control features are not yet available in the European Union and the United Kingdom.

The background computer-control feature allows Codex to view the screen, move the cursor, click and type inside macOS applications. OpenAI described the capability as useful for frontend iteration, app testing and workflows that do not expose APIs. Multiple agents can run concurrently without interrupting a user’s active work.

The in-app browser lets users annotate pages to give agents precise instructions. OpenAI wrote that the browser controls are focused on frontend and game development initially, with plans to broaden support over time.

Image generation is built into the Codex workflow via the gpt-image-1.5 model. Users do not need a separate API key to generate images; usage is covered by the linked ChatGPT account.

The update includes more than 90 new plugins that connect Codex to tools such as Atlassian Rovo, CircleCI, CodeRabbit, GitLab Issues, the Microsoft Suite and Neon by Databricks. These integrations let agents combine skills, app access and server connections across a developer’s existing toolset.

New productivity and workflow features include support for multiple terminal tabs, handling GitHub pull-request review comments, and SSH connections to remote development machines (SSH is in alpha). A summary pane tracks agent plans, sources and artifacts, and files open in a sidebar with previews for PDFs, spreadsheets, slides and documents. A proactive mode uses context from connected plugins, stored memory and active projects to surface a prioritized action list that draws from Google Docs comments, Slack threads, Notion pages and codebase context.

OpenAI reported that more than 3 million developers use Codex each week. The company positioned the desktop app as a single environment that combines computer control, browsing, image generation and coding while requiring a ChatGPT account for access.

Competing projects have introduced similar agent features. OpenClaw, created by developer Peter Steinberger, ran persistent agents locally and linked messaging apps, files, browsers and shell commands; the project received about 60,000 GitHub stars in 72 hours and later moved to an open-source foundation after Steinberger joined OpenAI in February. Anthropic’s Claude Code offers a terminal-based coding assistant that can read full codebases, edit files, run tests and commit to GitHub; Anthropic added a macOS computer-use research preview for Pro and Max subscribers in March.

The update is available to Codex desktop users signed in with a ChatGPT account, with some features restricted in the EU and UK during the initial rollout.