OpenAI upgrades Codex to use desktops, generate images, remember tasks

OpenAI expanded Codex into a more ambitious workstation agent, giving it the ability to operate desktop apps on a user’s computer while adding image generation and memory features that push it further from simple code completion and toward full workflow control.

The company said the updated Codex app functioned as a command center for AI coding and software development, with multiple agents, parallel workflows and long-running tasks. It was available on macOS, and OpenAI also offered a Windows version with native sandbox and PowerShell support, giving the system one interface for working across projects, running parallel agent threads and reviewing results.

Codex first arrived in 2025 as a cloud-based software engineering agent built to handle many tasks in parallel. OpenAI later expanded it into a desktop app, a shift that signaled a broader strategy: not just helping developers write snippets faster, but orchestrating larger pieces of the job inside one environment. OpenAI said ChatGPT Plus, Pro, Business, Edu and Enterprise plans included Codex.

The new features also widened the privacy and governance questions around AI coding tools. Letting an agent use desktop apps means more software, files and credentials may sit within reach of the system’s automation layer. Memory raises the stakes further. OpenAI’s own memory system in ChatGPT can reference saved memories and chat history, which can make the assistant more personalized and persistent, but also more difficult for companies to audit and contain.

Image generation added another layer. OpenAI already supported image generation and image editing through GPT Image or DALL·E models in its API, and bringing that capability into the Codex ecosystem made the product feel less like a coding assistant and more like a general-purpose production workstation. For developers, that could mean fewer handoffs between tools and faster iteration. For CIOs, it could also mean stronger vendor lock-in, since more of the workflow would live inside one company’s stack.

The move put OpenAI directly against Anthropic’s own push into agentic coding. Anthropic has described Claude Code as a system that reads codebases, makes changes across files, runs tests and delivers committed code. Its broader lineup has also grown to include Claude Cowork and desktop and office-app products such as Claude for Word, Excel and PowerPoint, alongside computer-use and autonomy upgrades.

That rivalry now looks bigger than code assistance. It is becoming a contest to control the entire workday, from the desktop to the terminal to the memory of past tasks, with the promise of speed matched by a sharper risk of dependence and overreach.