OpenAI introduces the Codex app as a desktop command center for multi-agent software work

OpenAI introduced the Codex app on February 2, 2026. The product is a macOS desktop interface designed to manage multiple agents at once, run long-running tasks in parallel, and supervise software work across projects. OpenAI describes it as a command center for agents rather than a single coding assistant. A March 4, 2026 update expanded the app to Windows.

The core workflow is built around project threads and worktrees. Each agent can operate on an isolated copy of the same repository, letting users review diffs, comment on changes, and jump back into an editor without losing track of the bigger job. The app also carries over session history and configuration from the Codex CLI and IDE extension, which ties desktop supervision to the tools developers are already using.

• Parallel management of multiple agents across project-specific threads
• Built-in worktree support to reduce conflicts in the same repository
• Skills management for workflows tied to tools like Figma, Linear, Vercel, PDFs, and spreadsheets
• Automations for scheduled background work such as issue triage, release briefs, and CI failure summaries

OpenAI also used the launch to push a security-by-design message. The Codex app uses the same native, open-source, configurable sandboxing approach as the Codex CLI. By default, agents are limited to editing files in the active folder or branch and using cached web search, while commands that require network access or other elevated permissions must go through an approval flow. That makes the product more suitable for teams that want powerful agents without fully unsupervised execution.

The distribution strategy is broad. OpenAI included Codex across ChatGPT Plus, Pro, Business, Enterprise, and Edu plans, temporarily opened access for ChatGPT Free and Go users, and doubled existing Codex rate limits during the rollout period. OpenAI also said more than one million developers used Codex in the last month, which suggests coding agents are shifting from a specialist tool into a broader developer platform category.