_March 4, 2026 update:_ _The Codex app is now available on Windows._
Today, we’re introducing the Codex app for macOS—a powerful new interface designed to effortlessly manage multiple agents at once, run work in parallel, and collaborate with agents over long-running tasks.
We're also excited to show more people what's now possible with Codex. For a limited time we're including Codex with ChatGPT Free and Go, and we're doubling the rate limits on Plus, Pro, Business, Enterprise, and Edu plans. Those higher limits apply everywhere you use Codex—in the app, from the CLI, in your IDE, and in the cloud.
The Codex app changes how software gets built and who can build it—from pairing with a single coding agent on targeted edits to supervising coordinated teams of agents across the full lifecycle of designing, building, shipping, and maintaining software.
Since we launched Codex in April 2025, the way developers work with agents has fundamentally changed. Models are now capable of handling complex, long-running tasks end to end and developers are now orchestrating multiple agents across projects: delegating work, running tasks in parallel, and trusting agents to take on substantial projects that can span hours, days, or weeks. The core challenge has shifted from what agents can do to how people can direct, supervise, and collaborate with them at scale—existing IDEs and terminal-based tools are not built to support this way of working.
This new way of building coupled with new model capabilities demands a different kind of tool, which is why we are introducing the Codex desktop app, a command center for agents.
The Codex app provides a focused space for multi-tasking with agents. Agents run in separate threads organized by projects, so you can seamlessly switch between tasks without losing context. The app lets you review the agent’s changes in the thread, comment on the diff, and even open it in your editor to make manual changes.
It also includes built-in support for worktrees, so multiple agents can work on the same repo without conflicts. Each agent works on an isolated copy of your code, allowing you to explore different paths without needing to track how they impact your codebase. As an agent works, you can check out changes locally or let it continue making progress without touching your local git state.
The app picks up your session history and configuration from the Codex CLI and IDE extension, so you can immediately start using it with your existing projects.
Codex is evolving from an agent that writes code into one that uses code to get work done on your computer. With skills(opens in a new window), you can easily extend Codex beyond code generation to tasks that require gathering and synthesizing information, problem-solving, writing, and more.
Skills bundle instructions, resources, and scripts so Codex can reliably connect to tools, run workflows, and complete tasks according to your team’s preferences. The Codex app includes a dedicated interface to create and manage skills. You can explicitly ask Codex to use specific skills, or let it automatically use them based on the task at hand.
We asked Codex to make a racing game, complete with different racers, eight maps, and even items players could use with the space bar. Using an image generation skill(opens in a new window) (powered by GPT Image) and a web game development skill(opens in a new window), Codex built the game by working independently using more than 7 million tokens with just one initial user prompt. It took on the roles of designer, game developer, and QA tester to validate its work by actually playing the game.
We’ve included the game, as well as the prompt and skills used to create it, below. You can also try out earlier iterations to see how Codex improved it as it worked for longer.
The game was created by Codex with the develop-web-game skill using this prompt that we've summarized for clarity:
Implement Voxel Velocity as a 3D voxel kart racer using Three.js, with exactly one mode: Single Race (always 3 laps, 1 human vs 7 CPU, and all 8 tracks available immediately with no progression). Build a minimal pre-race flow with only: Track (8), Character (8), Difficulty (Chill/Standard/Mean), optional Mirror Mode, optional Allow Clones, and Start Race, plus an Options menu and an in-race pause menu (Resume / Restart / Quit). Create an arcade driving model with responsive handling, forgiving glancing wall hits, meaningful drifting as the main skill, and a drift-charge system that produces exact boost tiers (Tier 1 0.7s, Tier 2 1.1s, Tier 3 1.5s) while keeping baseline speed “fast-but-readable” and pack passing constant on wide roads. Implement exactly 8 items with one-item capacity, subtle position-weighted distribution, and mild effects (max loss of control ≤1.2s, max steering disabled ≤0.6s) that create goofy chaos without hard stuns, plus off-road slowdowns that are reduced by 50% during boosts. Define the 8 characters with their given stats and AI tendencies, implement CPU difficulty presets and track-authored racing/variation splines, drift zones, and hazard avoidance so AI uses multi-lane width for clean overtakes, and ship HUD/audio essentials (position, lap/final lap banner, minimap, item slot, timer/splits, readable SFX, and one music loop per track).
Codex was then continuously reprompted from a random list of ten generic prompts to keep working on the problem. An example of one of the prompts is:
Your job is to add new features so the game matches the original more closely. First, play the game and identify what’s missing vs. the original. Then pick a few missing features and implement them. After each feature, thoroughly test it by playing the game and confirm it works. If you notice any bugs while playing, prioritize fixing them too.
At OpenAI, we’ve built hundreds of skills internally to help multiple teams confidently delegate work to Codex that would otherwise be hard to define consistently—from running evals and babysitting training runs to drafting documentation and reporting on growth experiments.
The Codex app includes a library of skills for tools and workflows that have become popular at OpenAI, with a few highlighted below. You can find the full list in the open source repo(opens in a new window).
* Implement designs:Fetch design context, assets, and screenshots from Figma(opens in a new window) and translate them into production-ready UI code with 1:1 visual parity.
* Manage projects:Triage bugs, track releases, manage team workload, and more in Linear(opens in a new window) to keep projects moving.
* Deploy to the cloud: Have Codex deploy your web app creations to popular cloud hosts like Cloudflare(opens in a new window), Netlify(opens in a new window), Render(opens in a new window), and Vercel(opens in a new window).
* Generate images:Use the image generation skill(opens in a new window) powered by GPT Image to create and edit images to use in websites, UI mockups, product visuals, and game assets.
* Build with OpenAI APIs:Reference up-to-date documentation(opens in a new window) when building with OpenAI APIs.
* Create documents:A set of skills for reading, creating, and editing PDF(opens in a new window), spreadsheet(opens in a new window), and docx(opens in a new window) files with professional formatting and layouts.
When you create a new skill in the app, Codex can use it wherever you work: in the app, CLI or in your IDE extension. You can also check skills into your repository to make them available to your entire team. Read more about sharing skills using team config here(opens in a new window).
With the Codex app, you can also set up Automations that let Codex work in the background on an automatic schedule. Automations combine instructions with optional skills, running on a schedule you define. When an Automation finishes, the results land in a review queue so you can jump back in and continue working if needed.
At OpenAI, we’ve been using Automations to handle the repetitive but important tasks, like daily issue triage, finding and summarizing CI failures, generating daily release briefs, checking for bugs, and more.
Setting up an automation to periodically create new skills
Developers have different preferences in how they work with an agent. Some want a blunt, execution-focused partner; others prefer more communicative, engaging interactions. Codex now lets developers choose between two personalities—a terse, pragmatic style and a more conversational, empathetic one, without any change in capabilities, to fit the approach you like the most. Just use the /personality command in the app, CLI, and IDE extension.
We’re integrating security by design across the entire Codex agent stack. The Codex app uses native,open-source(opens in a new window) and configurable system-level sandboxing just like in the Codex CLI. By default, Codex agents are limited to editing files in the folder or branch where they’re working and using cached web search, then asking for permission to run commands that require elevated permissions like network access. You can configure rules(opens in a new window) for your project or team that allows certain commands to automatically run with elevated permissions.
The Codex app is available starting today on macOS. Anyone with a ChatGPT Plus, Pro, Business, Enterprise or Edu subscription can use Codex across the CLI, web, IDE-extension and app with their ChatGPT login. Usage is included in ChatGPT subscriptions, with the option to purchase additional credits if needed.
For a limited time, Codex will also be available to ChatGPT Free and Go users to help build more with agents. We’re also doubling rate limits for existing Codex users across all paid plans during this period.
Enterprises and developers increasingly rely on Codex for end-to-end development. Since the launch of GPT‑5.2‑Codex in mid-December, overall Codex usage has doubled, and in the past month, more than a million developers have used Codex. We’ll continue to expand where and how developers can use Codex, including making the app available on Windows, pushing the frontier of model capabilities, and rolling out faster inference.
Within the app, we’ll keep refining multi-agent workflows based on real-world feedback, making it easier to manage parallel work and move between agents without losing context. We’re also building out Automations with support for cloud-based triggers, so Codex can run continuously in the background—not just when your computer is open.
Codex is built on a simple premise: everything is controlled by code. The better an agent is at reasoning about and producing code, the more capable it becomes across all forms of technical and knowledge work. Yet a key challenge today is the gap between what frontier models are capable of and how easily people can use them in practice. Codex is designed to close that gap by making it easier to direct, supervise, and apply the full intelligence of our models to real work. We’ve focused on making Codex the best coding agent, which has also laid the foundation for it to become a strong agent for a broad range of knowledge work tasks that extend beyond writing code. We’re excited to see what you build with Codex!
