OpenAI has released a separate Codex app for macOS that lets you work with multiple AI agents in parallel: distribute tasks between them, track progress, and return to long projects without losing context. At the launch, the company also opened Codex to users of ChatGPT’s free plans and increased limits for paid plans.
OpenAI has released a separate Codex app for macOS that lets you work with multiple AI agents in parallel: distribute tasks between them, track progress, and return to long projects without losing context. At the launch, the company also opened Codex to users of ChatGPT’s free plans and increased limits for paid plans.
According to OpenAI, the new application is designed as a separate workspace for managing AI agents. While Codex previously worked mainly as a code assistant or editor extension, the company now offers an interface in which you can simultaneously manage multiple tasks and control the work of each agent individually.
In Codex for macOS, each task opens in a separate workflow. The user sees exactly what the agent did, can review the changes, leave a comment, or manually edit the result in their editor. This is designed for long processes that last not minutes, but hours or days — for example, feature development, release preparation, or support for large projects.
A separate part of the application is dedicated to the so-called skills — pre-configured action scenarios. In essence, these are instructions and tools that allow Codex to not just write code, but perform more complex tasks: work with designs, prepare documents, analyze data, or deploy projects in the cloud. The user can choose a specific scenario or let the agent decide what to apply.
Codex also introduces Automations, which are automatic tasks that run on a schedule and run in the background. For example, an agent can regularly check for errors, prepare reports, or update documentation, and the result appears in the review queue.
OpenAI says security is built into the architecture. By default, agents only have access to working files and cannot perform dangerous actions without user permission. These rules can be flexibly configured for individual projects or teams.
This is OpenAI’s tool for automating code and technical tasks, which the company has been developing since 2025. After the release of the GPT-5.2-Codex model in late 2025, OpenAI recorded a sharp increase in the use of the service. According to its estimates, over 1 million developers used Codex in the last month alone. The macOS version is available now, and the company promises to launch a client for Windows later.
Previously, dev.ua wrote about how over a month ago OpenAI introduced a series of GPT-5.2 models, among them GPT-5.2-Codex — an agent programming model for complex software engineering in real-world scenarios. A few weeks ago, the model became available via the OpenAI API, and now it has also appeared on GitHub Copilot.
