OpenAI launches standalone Codex app for agentic coding
The macOS app lets a developer run several Codex coding agents in parallel from one window, each working for up to 30 minutes unsupervised before returning finished code.
- Models & capabilities
- Minor
OpenAI released a standalone Codex app for macOS, moving its coding-agent product out of the browser and code-editor extensions it had previously relied on and into a dedicated desktop application. The app functions as what OpenAI called a command centre for agents: a developer can delegate several coding tasks to separate Codex agents running in parallel, each able to work independently for up to around 30 minutes before returning completed code for review.
At launch the app ran on GPT-5.2-Codex, OpenAI’s coding model at the time, with the company’s next coding model, GPT-5.3-Codex, following three days later. A Windows version was not available at launch; OpenAI said it was in development. TechCrunch reported that OpenAI chief executive Sam Altman described the app internally as among the company’s most-used products.
The release is a further step in OpenAI’s push into agentic software development, competing directly with dedicated coding-agent products such as Anthropic’s Claude Code, and reflects a broader shift among frontier labs from single-turn code completion toward tools that supervise multiple autonomous coding sessions at once.