OpenAI rolls out a macOS Codex desktop app that runs multiple AI coding agents in parallel, anchored by an open source sandbox to make autonomous development secure and enterprise-ready.
OpenAI has launched a Codex desktop app for macOS, positioning it as a command centre for AI coding agents and pairing it with a native, open-source and configurable system-level sandbox to make autonomous development safer and more transparent for enterprises.
The app lets developers run multiple agents in parallel, delegate entire features, automate repetitive engineering work, and supervise systems that operate independently for up to 30 minutes. The shift moves beyond autocomplete tools toward orchestration of long-running, background workflows.
Skills bundle instructions, scripts, and integrations, while Automations schedule recurring jobs. Built-in worktrees isolate each agent’s code to prevent conflicts. Security sits at the core: agents are limited to specific folders and permissions, with explicit approvals required for elevated access.
Alexander Embiricos, Product Lead for Codex, OpenAI, said, “Codex has this sandbox that we’re actually incredibly proud of, and it’s open source, so you can go check it out.”
Internally, OpenAI says Codex has accelerated delivery. Thibault Sottiaux, Codex Team Lead, OpenAI, noted, “A Sora Android app is an example of that where four engineers shipped in only 18 days internally, and then within the month we give access to the world.”
Chief Executive Sam Altman called it transformational: “This is the most loved internal product we’ve ever had. It’s been totally an amazing thing for us to be using recently at OpenAI.”
The app is available now with ChatGPT subscriptions and has already been used by more than a million developers, as OpenAI aims to secure its lead in the fast-growing enterprise AI coding market while blending proprietary models with open, auditable safety infrastructure.
