OpenAI Launches Standalone Codex Desktop App, Enables Multi-Agent AI Coding on macOS

OpenAI has launched a standalone desktop version of its Codex AI coding tool, initially for macOS with a Windows release forthcoming, marking a strategic shift toward multi-agent software development. The app transforms Codex from a conversational assistant into a command center that orchestrates multiple AI agents across the full software lifecycle—design, coding, testing, and deployment. Codex introduces "worktrees" for isolated, parallel code modifications via Git integration, reusable "Skills" that encapsulate tools and workflows, and scheduled "Automations" for background tasks like issue triage and CI error analysis. In a demo, Codex autonomously built a 3D racing game using over 7 million tokens, acting as designer, developer, and tester. CEO Sam Altman called the output “more impressive than I imagined.” The app features dual interaction modes and a system-level sandbox restricting access by default. Over 1 million developers used Codex in the past month. Free and Go-tier ChatGPT users gain limited access, while paid tiers receive higher rate limits.