OpenAI is making several updates to its Codex AI coding agent. Codex is now able to operate desktop Mac apps with its own cursor, seeing what's on the screen, clicking, and typing to complete tasks.

Codex can run multiple agents on the Mac in parallel, without interfering with the user's own work. OpenAI says developers will find it useful for testing apps, iterating on frontend changes, and more. Codex can now remember preferences, recurring workflows, tech stacks, and other information about each user's personal workflow. With automation improvements, Codex is able to resume work after a pause using existing conversation threads, and it can schedule future work for itself and work on a task across days or weeks. Codex also proposes work using context from projects, memory, and connected plugins.

There is an in-app browser for Codex that allows users to comment directly on pages to provide more precise instructions to the agent. In the future, Codex will get full use of the browser for opening websites, working through user flows, taking screenshots, and inspecting outputs.

Codex has been updated to use gpt-image-1.5 for generating images in the app, which OpenAI says is helpful for creating visuals for product concepts and mockups. Codex now includes support for multiple terminal tabs, addressing GitHub review comments, and opening files directly in the sidebar with rich previews for documents like PDFs and spreadsheets.

Along with these changes, Codex has over 90 new plugins that can combine skills, app integrations, and MCP servers to improve Codex's context gathering and actions.

The updates to Codex are rolling out today to Codex desktop users signed in with ChatGPT. The personalization features are not yet available to Enterprise, Education, EU, and UK users, but will be rolling out soon. Computer use is also not yet available in the EU or the UK.