Expanding what developers can do, with the new Codex app for macOS.
Overview
OpenAI introduces the Codex app for macOS, a desktop application designed as a command center for managing multiple AI coding agents simultaneously. The app enables developers to run parallel agent workflows, extend capabilities through skills and automations, and collaborate with agents on long-running software development tasks across the full lifecycle of designing, building, shipping, and maintaining software.
What You'll Learn
1
How to manage multiple AI coding agents in parallel using the Codex desktop app
2
How to extend Codex beyond code generation using skills for tasks like image generation, project management, and cloud deployment
3
How to set up automations that let Codex run background tasks on a schedule
4
Why git worktrees enable multiple agents to work on the same repository without conflicts
5
How Codex's sandboxing and security model works by default and how to configure elevated permissions
Prerequisites & Requirements
- macOS computer (Windows support coming later)
- ChatGPT subscription (Plus, Pro, Business, Enterprise, or Edu) or Free/Go account during limited-time promotion
- Familiarity with git workflows and version control concepts
- Basic experience with AI coding assistants or the Codex CLI(optional)
Key Questions Answered
What is the OpenAI Codex app and how does it differ from the Codex CLI?
The Codex app is a macOS desktop application that serves as a command center for managing multiple AI coding agents simultaneously. Unlike the CLI which handles single-agent interactions, the app provides a visual interface for running agents in parallel across projects, reviewing diffs in-thread, managing skills and automations, and switching between tasks without losing context. It picks up session history and configuration from the CLI and IDE extension.
How does the Codex app handle multiple agents working on the same repository?
The Codex app includes built-in support for git worktrees, which allows multiple agents to work on the same repository without conflicts. Each agent operates on an isolated copy of the code, enabling developers to explore different paths simultaneously. As an agent works, you can check out its changes locally or let it continue making progress without affecting your local git state.
What are Codex skills and how do they extend the agent's capabilities?
Skills are bundles of instructions, resources, and scripts that enable Codex to connect to tools, run workflows, and complete tasks beyond code generation. They allow Codex to handle tasks like implementing Figma designs, managing Linear projects, deploying to cloud platforms (Cloudflare, Netlify, Render, Vercel), generating images with GPT Image, and creating documents. Skills can be shared across CLI, IDE, and app, and checked into repositories for team-wide use.
What are Codex automations and how do they work?
Automations let Codex work in the background on an automatic schedule without manual intervention. They combine instructions with optional skills and run on a user-defined schedule. When an automation finishes, results land in a review queue for human review. OpenAI uses automations internally for daily issue triage, summarizing CI failures, generating release briefs, and checking for bugs.
How much does the Codex app cost and who can access it?
The Codex app is included in ChatGPT Plus, Pro, Business, Enterprise, and Edu subscriptions with usage included and optional additional credits available for purchase. For a limited time, Codex is also available to ChatGPT Free and Go users. During this promotional period, rate limits are doubled for all existing paid plan users across the CLI, web, IDE extension, and app.
What security measures does the Codex app use by default?
The Codex app uses native, open-source, configurable system-level sandboxing identical to the Codex CLI. By default, agents are limited to editing files only in the folder or branch where they're working and using cached web search. Commands requiring elevated permissions like network access require explicit user approval. Teams can configure rules to allow certain commands to run with elevated permissions automatically.
How many developers use Codex and how fast is it growing?
Since the launch of GPT-5.2-Codex in mid-December 2025, overall Codex usage has doubled. In the past month alone, more than one million developers have used Codex. Enterprises and developers are increasingly relying on it for end-to-end development workflows, signaling rapid adoption of AI-powered coding agents.
Can the Codex app customize the agent's communication style?
Yes, Codex offers two selectable personality modes: a terse, pragmatic style focused on execution, and a more conversational, empathetic style for developers who prefer engaging interactions. Both personalities have identical capabilities—only the communication approach differs. The personality can be switched using the /personality command in the app, CLI, and IDE extension.
Key Statistics & Figures
Developer adoption
Over 1 million developers
Used Codex in the past month (as of February 2026
Usage growth
Doubled
Overall Codex usage since the launch of GPT-5.2-Codex in mid-December 2025
Token usage for racing game demo
More than 7 million tokens
Codex consumed to build a complete 3D voxel kart racing game from a single initial prompt
Rate limit increase
2x
doubled
Technologies & Tools
Some links below are affiliate links. We may earn a commission if you make a purchase.
AI Coding Agent
Codex
Core AI coding agent platform available via app, CLI, IDE extension, and cloud
AI Model
Gpt-5.2-codex
Underlying model powering the Codex agent, launched mid-December 2025
AI Image Generation
Gpt Image
Powers the image generation skill for creating game assets, UI mockups, and product visuals
3d Graphics Library
Three.js
Used in the demo racing game built by Codex as a 3D voxel kart racer
Design Tool
Figma
Integrated via skill to fetch design context and translate designs into production-ready UI code
Project Management
Linear
Integrated via skill for bug triage, release tracking, and team workload management
Cloud Hosting
Cloudflare
Supported as a deployment target via Codex skill
Cloud Hosting
Netlify
Supported as a deployment target via Codex skill
Cloud Hosting
Render
Supported as a deployment target via Codex skill
Cloud Hosting
Vercel
Supported as a deployment target via Codex skill
Operating System
Macos
Initial platform for the Codex desktop app (Windows support planned)
Version Control
Git
Worktree support enables multiple agents to work on the same repository without conflicts
Key Actionable Insights
1Use git worktrees through the Codex app to run multiple agents on the same repository simultaneously without merge conflicts. Each agent works on an isolated copy of your codebase, letting you explore different implementation paths in parallel and only merge the results you want.This is particularly valuable for teams evaluating multiple approaches to a problem or working on independent features that would otherwise require branch switching overhead.
2Create and share Codex skills by checking them into your repository to standardize workflows across your entire team. Skills bundle instructions, resources, and scripts so agents can reliably execute tasks according to team preferences, from deployment workflows to documentation standards.OpenAI has built hundreds of internal skills across multiple teams. Team config sharing is documented at developers.openai.com/codex/enterprise/admin-setup#team-config.
3Set up automations for repetitive but important tasks like daily issue triage, CI failure summarization, and release brief generation. Automations run on a defined schedule and deposit results in a review queue, reducing the manual overhead of maintenance-type developer work.OpenAI uses automations internally for these exact use cases. This is especially valuable for teams with recurring operational tasks that consume developer time but don't require real-time human judgment.
4Take advantage of the current promotional period where Codex is available on Free and Go plans and rate limits are doubled on all paid plans. This is a limited-time offer to evaluate whether multi-agent workflows provide value for your development process before committing to ongoing usage.The doubled rate limits apply across all Codex surfaces: the app, CLI, IDE extension, and cloud, making it an ideal time to stress-test the platform at scale.
5Configure project-level or team-level security rules to automatically grant elevated permissions for trusted commands, rather than approving each permission request individually. This balances the default sandbox security model with practical workflow efficiency for established projects.The default sandboxing limits agents to file edits within their working folder and cached web search. Configuring rules via developers.openai.com/codex/rules prevents permission approval fatigue while maintaining security boundaries.
Common Pitfalls
1
Trying to manage multiple parallel agents using traditional IDE or terminal-based tools that were designed for single-threaded human workflows. These tools lack the context-switching and multi-thread management capabilities needed to effectively supervise several agents working simultaneously on different tasks.
The Codex app was specifically built to address this limitation, providing a dedicated interface for multi-agent orchestration with separate threads organized by project.
2
Running multiple agents on the same repository branch without isolation, leading to merge conflicts and inconsistent code states. Without proper isolation, agents can overwrite each other's changes or create conflicting modifications.
Use the app's built-in worktree support to give each agent an isolated copy of the codebase, and review/merge results selectively.
3
Approving elevated permissions individually for every agent action, creating a bottleneck that negates the efficiency gains of parallel agent workflows. This is especially problematic for teams running many agents with common trusted operations.
Configure project-level or team-level security rules to pre-approve trusted commands while maintaining sandbox restrictions for unknown operations.
4
Creating skills only in the app without checking them into the repository, which prevents team members from accessing and benefiting from shared workflow automations. Skills created locally are only available to the individual developer.
Check skills into your repository and use team config to make them available across your entire team, ensuring consistent workflows regardless of which surface (app, CLI, IDE) team members use.
Related Concepts
AI Coding Agents
Multi-agent Orchestration
Git Worktrees
Agentic Workflows
Task Automation
Sandboxed Execution
Developer Experience (dx)
IDE Integrations
CI/CD Automation
Agent Skills And Plugins
Cloud Deployment
Llm-powered Development Tools