OpenAI Expands Codex With Desktop Control and Memory Features

OpenAI announced a major update to Codex, adding new capabilities that allow the AI assistant to interact directly with desktop environments and remember user preferences over time.

The updated Codex can now operate on a user’s computer by seeing, clicking, and typing with its own cursor. Multiple agents can work in parallel on macOS without disrupting user activity, enabling tasks such as testing applications or adjusting frontend designs. A new in-app browser lets users issue commands directly on web pages, with plans to extend this feature for broader browser control.

The release also introduces an image generation feature using OpenAI’s gpt-image-1.5 model, allowing developers to create mockups and product visuals within their workflows. More than 90 new plugins have been added, integrating tools like GitLab Issues, Microsoft Suite, and Atlassian Rovo for task management and automation.

A preview of Codex’s memory function lets the assistant recall context from previous sessions, including user preferences and corrections, to improve task continuity. The update also supports scheduling future work and resuming long-term projects automatically. These features are currently rolling out to Codex desktop users signed in with ChatGPT, with enterprise and education access planned soon.

The move positions OpenAI to compete more directly with Anthropic, whose Claude Code product has recently gained traction among enterprise users.