# How do you use the Codex app on macOS? Date: 2026-02-11 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/use-codex-app-macos/ **Answer:** Sign in with your ChatGPT account, add a project folder, and start one thread per task; each agent can work in its own copy of the code, and you review its changes as a diff before keeping them. Start with small, separate tasks, read every diff, and add skills or scheduled Automations once the basics feel routine. The Codex app, released by OpenAI for macOS on February 2, 2026, is a desktop window for directing **coding agents**: AI models that read your code, edit files and run commands to complete a programming task. Where a chat assistant suggests code for you to paste, Codex makes the changes itself and then shows you what it did. The app's particular strength is running several of these agents at once, each on its own task, while you supervise. Codex is included in ChatGPT Plus, Pro, Business, Enterprise and Edu plans, and for a limited time OpenAI is also offering it to Free and Go users, with doubled rate limits on paid plans during that period. ## Three ideas to understand first **Threads.** Each task runs in its own thread, grouped under the project it belongs to. You can switch between threads without losing the context of any of them. **Worktrees.** A Git **worktree** is a second working copy of the same repository, on its own branch. The app has built-in support for them, so two agents can change the same project without tripping over each other, and your own checkout stays untouched until you decide to bring changes in. **The sandbox.** By default, agents may edit files only inside the folder or branch they are working in, use cached web search, and must ask your permission before running commands that need more access, such as reaching the network. You can write rules that let specific commands run automatically for a project or team. ## A worked first session Imagine a small personal website kept in a Git repository: a few HTML pages, a stylesheet and a script that builds a list of recent posts. You want three things done. **1. Sign in and add the project.** Open the app, sign in with your ChatGPT account and add the website's folder as a project. Make sure the repository has no uncommitted changes, so you can always return to a clean state. **2. Start three threads, one per task.** Keep each request narrow and say how you will judge success: ```prompt Thread 1: The "Recent posts" list on index.html shows posts in the wrong order. Find the cause in build-posts.js, fix it so the newest post comes first, and explain the fix in two sentences. ``` ```prompt Thread 2: Add simple automated tests for build-posts.js covering ordering and empty input. Use the test tool already in package.json if there is one; ask before installing anything new. ``` ```prompt Thread 3: Read the site and update README.md with accurate setup and build instructions. Do not change any other file. ``` Run threads 2 and 3 in their own worktrees, so they do not touch the files thread 1 is editing. **3. Review each diff.** When a thread finishes, open its changes. A **diff** shows exactly which lines were added or removed. You can comment on specific lines to ask for changes, or open the files in your own editor and adjust them by hand. Keep only what you understand. **4. Bring the changes together.** Check out each set of changes locally, run the site and the new tests, and commit them the way you normally would. An illustrative outcome: thread 1 finds that posts were sorted as text rather than as dates, thread 2 adds four tests and asks once for permission to run them, and thread 3 rewrites the README. Your session will differ, but the pattern of narrow tasks, separate copies and careful review carries over. > **Try it:** On a project you know well, start two threads with tasks you could do yourself in ten minutes each. Compare the diffs with what you would have written, and note one thing you would ask the agent to do differently next time. ## Skills and Automations Once single tasks feel comfortable, two features extend the app. **Skills** are folders of instructions, resources and scripts that teach Codex a repeatable workflow, such as your team's documentation style or deploying to a host like Netlify or Vercel. The app has a screen for creating and managing them, and a skill you create works in the app, the CLI and the IDE extension alike. **Automations** run a set of instructions, optionally with skills, on a schedule you choose; results arrive in a review queue. OpenAI says it uses them internally for chores such as daily issue triage and summarizing failed builds. You can also pick how Codex talks to you: the `/personality` command switches between a brief, to-the-point manner and a chattier, warmer one, without changing what it can do. ## How does it compare with Codex in the terminal? The **Codex CLI** is the same agent in a command-line form. It runs locally, is open source, and installs with `npm install -g @openai/codex` or `brew install --cask codex`; you start it by typing `codex` in a project folder and choosing **Sign in with ChatGPT**, or by setting up an API key. It suits people who already live in the terminal, and it fits naturally into scripts and remote servers. The app and the CLI share more than a name. The app reads your session history and settings from the CLI and the IDE extension, so existing projects appear without setup. Skills and the `/personality` command work in both, the sandboxing approach is the same, and usage counts against the same ChatGPT plan. The difference is in supervision. In the CLI you typically work with one session per terminal window. The app is built for several agents in parallel, with worktrees, a visual diff for each thread and a queue for scheduled work. A practical split for many developers is to use the terminal for quick, focused changes and the app when several tasks are running at once and each needs a careful review. ## Sources - Introducing the Codex app: https://openai.com/index/introducing-the-codex-app/ - openai/codex README: https://github.com/openai/codex/blob/49342b156d55f0b7d7b7e63eae6702456966d998/README.md - OpenAI launches Codex app for macOS to speed up software projects: https://siliconangle.com/2026/02/02/openai-launches-codex-app-macos-speed-software-projects/