OpenAI Ships Codex Desktop App as Agent Command Center

More than one million developers used OpenAI's Codex in the past month, and the company is now betting they need a new kind of software to manage them. On March 4, 2026, OpenAI launched the Codex app for macOS — now also available on Windows — a desktop interface built to run and supervise multiple AI agents at once.

The launch comes with aggressive distribution. For a limited time, OpenAI is including Codex with ChatGPT Free and Go plans, and doubling rate limits on Plus, Pro, Business, Enterprise, and Edu subscriptions. The higher limits apply everywhere Codex runs: the new app, the CLI, the IDE extension, and the cloud.

Why does a desktop app matter?

Since OpenAI launched Codex in April 2025, the product's trajectory has changed what developers expect from agents. Models now handle complex, long-running tasks end to end, and developers orchestrate multiple agents across projects — delegating work, running tasks in parallel, and trusting agents with projects that span hours, days, or weeks.

According to OpenAI, the core challenge has shifted. "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," the company wrote in its announcement.

The Codex app is OpenAI's answer: a "command center for agents." Usage data backs the bet. Since the launch of GPT-5.2-Codex in mid-December, overall Codex usage has doubled.

What does the app actually do?

The app gives developers a focused space for multitasking with agents:

• Agents run in separate threads organized by project, so users can switch between tasks without losing context.
• Developers can review an agent's changes in the thread, comment on the diff, or open it in an editor for manual changes.
• Built-in worktree support lets multiple agents work on the same repository without conflicts, each on an isolated copy of the code.
• The app picks up session history and configuration from the Codex CLI and IDE extension, so existing projects work immediately.

OpenAI frames the app as changing "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."

How far can skills push Codex beyond code?

The bigger strategic signal is skills. Codex is evolving from an agent that writes code into one that uses code to get work done on a computer. Skills — bundles of instructions, resources, and scripts hosted in an open-source repo — extend Codex to tasks like gathering and synthesizing information, problem-solving, and writing. The app includes a dedicated interface to create and manage them, and skills work across the app, CLI, and IDE extension. Teams can check skills into a repository to share them.

OpenAI demonstrated the scope with a stress test. Using an image generation skill powered by GPT Image and a web game development skill, Codex built "Voxel Velocity," a 3D voxel kart racer with eight characters, eight tracks, items, difficulty presets, and drift mechanics — from a single initial user prompt. The agent worked independently, consuming more than 7 million tokens, and took on the roles of designer, game developer, and QA tester, validating its work by actually playing the game. OpenAI then kept prompting it with generic follow-up instructions such as: "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."

Internally, OpenAI says it has built hundreds of skills for work that would otherwise be hard to define consistently — from running evals and babysitting training runs to drafting documentation and reporting on growth experiments. The app ships with a curated library, including:

• Implement designs: pull context and assets from Figma and translate them into production-ready UI code with 1:1 visual parity.
• Manage projects: triage bugs and track releases in Linear.
• Deploy to the cloud: push web apps to Cloudflare, Netlify, Render, or Vercel.
• Generate images: create and edit images via GPT Image for websites, mockups, and game assets.
• Build with OpenAI APIs: reference up-to-date documentation.
• Create documents: read, create, and edit PDF, spreadsheet, and docx files.

What are Automations and personalities?

The app also introduces Automations — scheduled background jobs that combine instructions with optional skills. Finished results land in a review queue. OpenAI uses them internally for daily issue triage, summarizing CI failures, generating daily release briefs, and checking for bugs.

A smaller change addresses developer taste. Codex now offers two personalities — a terse, pragmatic style and a more conversational, empathetic one — with no change in capabilities. Users switch via the /personality command in the app, CLI, and IDE extension.

How does OpenAI handle security?

The app uses native, open-source, configurable system-level sandboxing, the same approach as the Codex CLI. By default, agents can only edit files in the folder or branch where they work and use cached web search; commands requiring elevated permissions, such as network access, prompt the user first. Teams can configure rules to allow specific commands to run automatically.

Who gets access, and at what price?

The macOS app (and now Windows) is available to anyone with a ChatGPT Plus, Pro, Business, Enterprise, or Edu subscription, usable across CLI, web, IDE extension, and app with a ChatGPT login. Usage is included in subscriptions, with additional credits available for purchase. The Free and Go inclusion and the doubled rate limits are temporary.

The pricing move matters beyond OpenAI. Free-tier access to an agentic coding tool puts direct pressure on rivals like GitHub Copilot and Anthropic's Claude Code, and signals that OpenAI views Codex less as a premium developer add-on and more as the front door to its agent platform.

OpenAI states its premise plainly: "The better an agent is at reasoning about and producing code, the more capable it becomes across all forms of technical and knowledge work." The company says making Codex the best coding agent has "laid the foundation for it to become a strong agent for a broad range of knowledge work tasks that extend beyond writing code."

Next on the roadmap: refining multi-agent workflows based on real-world feedback, expanding Automations with cloud-based triggers so Codex runs continuously in the background even when the user's computer is closed, pushing model capabilities, and rolling out faster inference. The gap OpenAI is chasing, as it puts it, is "between what frontier models are capable of and how easily people can use them in practice."