ChatGPT Work and Codex: OpenAI tries to explain where conversations are stored

OpenAI published a reference article about how ChatGPT Work works with Codex, in which it tried to explain to users why conversations started in a browser or on a mobile device do not appear in the desktop application — and, in the opinion of independent developer Simon Willison, the explanation turned out to be "unsuccessful."

What OpenAI's reference says

In the cited passage, the company divides ChatGPT Work operations into two modes that differ in where tasks are physically executed and where data is stored.

• Work in browser (web) and in mobile app is performed in the cloud
• Work in desktop app can additionally use local files and other applications on the user's computer — with their permission
• At the time of launch, cloud conversations (cloud Work conversations) do not appear in the desktop version of Work
• Threads started in the desktop app and associated local files remain only on the computer where they were created

What is ChatGPT Work

ChatGPT Work is an OpenAI product aimed at work-related ChatGPT use cases: it combines access through browser, mobile app, and desktop client, as well as integration with Codex — OpenAI's agent tool for automating development tasks. Based on the wording of the reference, the key difference of the desktop version is expanded permissions: it can, with user permission, access files on the computer and other installed applications, while web and mobile versions work exclusively in OpenAI's cloud environment. The very title of the reference article — "ChatGPT Work and Codex" — directly indicates this connection.

Why this caused questions

The formulation describes the architectural separation between cloud and local parts of the product, but does not explain to the user how to track for himself which mode he is currently working in and where to find a conversation he started earlier. Developer Simon Willison, commenting on the quote in his blog, characterized OpenAI's attempt to clarify the mechanics of ChatGPT Work as unsuccessful — that is, in his assessment, the reference does not solve the original problem of misunderstanding, but merely states the fact of data separation between platforms.

The problem is that the reference's wording answers the question "what happens to data," but does not answer the more practical question a user has — "where is my conversation and why don't I see it on another device." For a person who opened ChatGPT Work on their phone in the morning and sat down at a desktop app with Codex in the evening, the lack of a unified conversation history might look like a bug, even though according to OpenAI's description this is the expected behavior of the system at this stage.

What this means

The separation between cloud and desktop conversations in ChatGPT Work is a direct consequence of the fact that the desktop app receives access to local files and programs, while the cloud version does not: from a data security and privacy standpoint this is justified, but from a user experience perspective it means that a unified message history across devices will not exist for now. For teams planning to use ChatGPT Work and Codex simultaneously in both browser and desktop, it means having to manually decide where to start a particular task — in the cloud or locally — and not counting on automatic context synchronization between modes. Until OpenAI changes the architecture or adds explicit synchronization between cloud and local conversations, users should decide in advance which mode — web/mobile or desktop — to conduct a particular workflow in, especially if it is tied to Codex and local project files.