To inject AI into more programming processes, OpenAI is launching the Codex CLI, a coding “agent” designed to run locally from terminal software.
Announced Wednesday alongside Openai’s latest AI models, the Codex CLI, links Openai’s models to local code and computing tasks, Openai says. Through the Codex CLI, Openai models can write, edit code on the desktop, and perform certain actions such as moving files.
The Codex CLI appears to be a small step towards Openai’s wider agent coding vision. Recently, the company’s CFO, Sarah Frier, described what she called an “agent software engineer.” It aims to build Openai, a tool that can take a description of the app’s project, effectively create it, and even create it.
Codex Cli doesn’t go that far. However, it is called a client (command line interface) that integrates OpenAI models and ultimately includes O3 and O4-Mini and handles code and computer commands.
Openai also says it is open source.
“(Codex CLI) is a lightweight, open source coding agent that runs locally in terminals,” an Openai spokesperson told TechCrunch via email. “The goal provides users with a minimal, transparent interface (code and tasks) for directly linking models.”
In a blog post provided to TechCrunch, Openai added: “By passing screenshots or low-fidelity sketches to the model, you can get the benefits of multimodal inference from the command line, accessing code and locally (via the Codex CLI).
To promote the use of Codex CLI, Openai plans to invest $1 million in API grants into qualified software development projects. The company says it will award $25,000 in API credits to selected projects.
Of course, AI coding tools carry risks. Many studies have shown that code generation models cannot fix security vulnerabilities and bugs, and even introduce them. It’s best to keep that in mind before providing AI access to sensitive files and projects. The entire system needs to be kept small.