ChatGPT Desktop App Gains Mac Automation and Task Memory in Codex Update.

OpenAI is expanding the capabilities of its ChatGPT desktop app with a new set of features that blur the line between conversational AI and desktop automation. The company announced on LinkedIn that its Codex interface can now control applications on macOS, connect to third-party tools, generate images, and maintain context across sessions by learning from previous interactions.

The updates position ChatGPT's desktop client as a more persistent and proactive assistant, capable of executing recurring tasks rather than simply responding to one-off prompts. OpenAI states the features are rolling out starting today.

The headline addition is native Mac app control. According to the announcement, Codex can now interact directly with applications installed on a user's computer, though OpenAI has not yet detailed which apps are supported or how permissions are managed. This functionality puts ChatGPT in direct competition with emerging AI automation tools like OpenClaw and Windsurf, which similarly aim to bridge AI interfaces with local system control.

The desktop app is also gaining expanded integration capabilities, allowing connections to external services and workflows. While specifics remain vague, this likely builds on ChatGPT's existing plugin ecosystem and GPT Actions framework, bringing those integrations into the native desktop experience. For users already relying on automation platforms like Zapier or Make, tighter desktop-level integration could reduce context switching.

Perhaps more significant than individual features is the shift toward persistent context. OpenAI says Codex can now "learn from previous actions" and "remember how you like to work." This memory layer suggests the assistant will adapt its behavior based on user patterns, similar to how ChatGPT's existing memory feature works in the web interface but applied to desktop workflows and task execution.

The update also emphasizes handling "ongoing and repeatable tasks," which implies scheduled or triggered automation rather than purely conversational interaction. If executed well, this could transform ChatGPT from a question-and-answer tool into something closer to a programmable personal assistant.

OpenAI's move comes as AI coding assistants expand their scope. Tools like Cursor and Replit have focused on developer workflows, while Claude has emphasized computer use capabilities. ChatGPT's desktop push suggests OpenAI is betting that a general-purpose AI interface with deep system integration will appeal to non-technical users as much as developers.

The update also bundles image generation directly into the Codex desktop app. While ChatGPT has offered DALL-E integration via the web interface for some time, including it natively in the desktop client removes friction for users who switch between text and visual tasks. This matches the strategy of competitors like Gemini, which integrates Google's Imagen model directly into its chat experience.

The LinkedIn announcement offers little technical detail on rollout timelines, supported macOS versions, or whether similar features will come to Windows. OpenAI has not published a corresponding blog post or changelog entry as of this writing, leaving key implementation questions unanswered. Users of the ChatGPT desktop app should see the updates begin appearing in their installations, though exact availability may vary.