Tuesday, November 10, 2026 Independent reporting on artificial intelligence Newsletter
AI News Beats
AI Companies

OpenAI launches a Codex desktop app for macOS to manage multiple coding agents

OpenAI's new Codex app gives developers a single place to run and supervise several AI coding agents in parallel, with worktrees, scheduled automations and built-in code review.

Slightly opened silver MacBook
Photo: Dmitry Chernyshov / Unsplash

OpenAI has released a Codex desktop application for macOS, designed to give developers one place to run, monitor and direct several AI coding agents working on long-running projects at the same time. The company announced the app in early February 2026.

Codex began in April 2025 as a coding agent that turned natural-language instructions into software tasks. Since then, OpenAI says, it has grown from writing code snippets to handling larger, more complex and longer-running jobs. As developers began using several agents on a single project, the company says conventional IDEs and terminals became awkward places to keep track of everything, and the new app is intended to fill that gap.

Threads, worktrees and review in one window

In the app, each project lives in its own thread, and multiple agents can work within it simultaneously. The developer can see which task has progressed how far, what each agent has changed and what is still running. Code changes can be reviewed inside the app, commented on, or sent back with further instructions.

The app also supports worktrees, which give each agent its own copy of the code so that different experiments can run against the same repository without disturbing the main codebase. According to OpenAI, the app picks up existing settings and history from the Codex command-line tool and IDE extension, so developers do not need to configure it from scratch.

Skills and automations

OpenAI highlights skills as a core part of the app. Skills connect Codex to external tools and workflows, such as generating user interfaces from design files, working with bug trackers or deploying to cloud platforms. The company says it uses Codex internally in similar ways, including preparing documentation, monitoring training runs and handling routine technical work.

Automations let recurring tasks run on a schedule. When one completes, the result is surfaced for review so that a person decides the next step. Developers can also choose how the agent communicates, from short and direct to more explanatory and conversational.

Security and availability

The app runs in a restricted environment with explicit controls over which files and commands the agent can touch. Sensitive operations require permission, and rules can be set at the team or project level. OpenAI says the app is available on macOS and is included with several ChatGPT subscription plans, with free users getting access for a limited period.

The release positions Codex against other agent-oriented coding tools, including Anthropic’s Claude Code and Google’s Antigravity, as AI labs compete to become the layer through which developers supervise automated software work rather than write it line by line. OpenAI’s framing is that the developer decides what to build and the agents carry it out, with the app serving as the control room.

Stay on top of AI

Get the stories that matter in artificial intelligence, straight to your inbox.

Free. One email with the stories that matter. Unsubscribe any time. See our privacy policy.