OpenAI is testing a “Persistent mode” for Codex that would let the coding agent continue working proactively until a user puts it to sleep. WIRED found references to the setting in recent changes to the public code base for Codex’s command-line tool.
Features often appear in that repository before reaching the Codex desktop app or ChatGPT Work, but this one has not been broadly released or formally announced. OpenAI confirmed that it is experimenting with the capability while saying there are no immediate plans to launch it. That makes Persistent mode a development signal rather than an available product promise.
A continuously active agent could monitor tasks and handle follow-up work without a new prompt each time. It would also raise practical questions about spending limits, permissions and when the software should stop. Those controls matter more as agents gain access to code, files and external services. For now, users should not assume current Codex installations can run indefinitely or that the experiment will ship in its present form.