OpenAI has updated its codex tool with new functions that allow you to act more integrated with the computer, operating desktop applications, as well as generating images and remembering past experiencesamong other news.
As announced last March, the company led by Sam Altman plans to unify its ChatGPT assistant, with the Codex application and its Atlas web browser, in a computer “super app” that simplifies the user experience optimizing its resources and allowing it to compete even more with companies like Anthropic.
Within this framework, OpenAI has moved towards this super application with an update to its Codex service that, beyond its abilities to generate code, can now work more integrated with the computer, work with more tools and applications for everyday use, generate images, remember user preferences, learn from previous actions and perform repetitive and continuous tasks.
This was shared by the technology company in a statement, where it specified that it has also added a more comprehensive support for developer workflowssuch as reviewing pull requests, viewing multiple files and terminals, or connecting to remote development environments.
Among the new features of this update, OpenAI has highlighted Codex’s ability to use the computer in the background to interact with all installed applications. Thus, the tool can click, write with its own cursor and detect each action.
Furthermore, this ability also allows several agents to work in parallel on the computer without interrupting the work the user is carrying out with other applications.
On the other hand, OpenAI has also highlighted how Codex can work natively with the web. This is because it has a built-in browser from which users can give precise instructions to the agent directly on the pages. That is, for example, in ‘frontend’ development, it allows you to change components of a website simply by selecting the part you want to vary and describing the change.
Another of the added features is that Codex can use the gpt-image-1.5 model to generate and modify images. This also allows you to combine screenshots and code to generate content, as well as being able to create visuals for product concepts, interface designs, mockups, or even games, in the same workflow.
Codex will now also be able to reuse existing conversations, preserving previously established context. That is, it will be able to take into account what has been done previously to schedule future tasks and automatically activate to continue with long-term tasks autonomously.
Following this line, a preview version of the memory function, which allows Codex to remember useful information from previous experiencesincluding personal preferences, corrections, or information that took time to collect.
In addition to all this, OpenAI has released 90 new plugins for Codex that allow combining functionalitiesas well as the integration of applications and MCP servers, so that the tool can have more options to collect context and, thus, perform actions in all its tools accurately.
These new plugins include Microsoft Suite, GitLab Issues, Neon by Databricks, Render and Superpowers, among others, as specified by the company.
As a result, Codex will be able to proactively propose tasks thanks to project context, connected plugins, and memory. For example, you may suggest revisiting a previous project.
As the director of Codex, Thibault Sottiaux, explained in a press conference reported by Engadget, This release is aimed at developers and will be expanded for more widespread use soon. Thus, the new capabilities are designed to offer developers ease in managing AI agents that can operate on broader tasks more proactively.
The technology company has clarified that the application now also includes support for responding to GitHub review commentsas well as running multiple terminal tabs and connecting to remote development environments using SSH.
Likewise, users will also be able to open files with rich previews for PDFs, spreadsheets, slides or other documents directly in the sidebar.
OpenAI has highlighted that all of this allows “moving more quickly through all stages of the software development life cycle”, from writing code to checking results or reviewing changes, including collaborating with agents, all in the same workspace.
The update is now available in the Codex desktop app for users with a ChatGPT account. However, some features such as personalization, contextual suggestions, memory or computer control will be available in Europe and the United Kingdom later to ensure they comply with regulations regarding data use. In addition, computer usage is initially available on macOS.
