OpenAI’s post announces a preview of Codex inside the ChatGPT desktop app for Linux. This is not just a chat client. It is the desktop version of their coding and computer-use workflow, aimed at managing projects, files, browser tasks, terminals, and remote connections from one app. People who already use Codex on Mac or Windows said the GUI’s appeal is straightforward: it is easier to supervise multiple agents, keep separate project contexts, review diffs, use built-in worktrees, and hand work off between desktop and phone than it is in the CLI alone.
The stronger signal was skepticism about the package OpenAI chose to ship. Many comments treated the Linux port as less interesting than the fact that OpenAI folded Codex into a broader ChatGPT app that users describe as slower, more bloated, and less focused than the old standalone Codex or the older "ChatGPT Classic" app. Several reported huge idle memory footprints and lag on Windows and macOS, which made the Linux release feel like more of the same rather than a clean fresh start. That criticism was tied to a broader point: if AI companies claim software development is becoming radically faster, shipping a resource-heavy cross-platform app months later than Mac and Windows undercuts the marketing more than it hurts Linux users.
Security anxiety ran just as hot as performance complaints. A lot of people see a desktop agent with file access and browser control as something closer to a trojan than a normal productivity app. The Linux build appears to use
bubblewrap and
seccomp for a workspace sandbox by default, which some saw as a meaningful baseline. Others argued that app-provided sandboxing is not enough and that if you care about your machine you should still isolate it in a
VM or keep the agent limited to a repo or config directory. That practical framing beat the absolutism. Even commenters who like these tools described using snapshots, separate environments, or approval gates rather than trusting the app outright.
The comments that liked the product were not defending the implementation so much as the workflow. They pointed to concrete wins from desktop control and richer UI, especially for long-running jobs, remote access from a phone, GUI-heavy tasks, and faster human-in-the-loop review. The overall read is clear: Linux support itself is welcome, but the thing being shipped is seen as a powerful and convenient orchestration layer wrapped in a proprietary, heavyweight desktop app that many technical users still do not trust on their main machine.
community.openai.com