Severe memory growth while resuming a large local Codex thread on macOS
Summary
On macOS, resuming a large local Codex thread after a fresh app restart causes the codex process to grow rapidly from roughly 116 MB to 13.82 GB RSS in about two minutes. This drives memory pressure into red and creates more than 14 GB of swap on a Mac with 8 GB RAM, making the computer nearly unusable.
This has been recurring for several days; it did not occur previously.
Environment
- macOS
- 8 GB physical memory
- ChatGPT desktop app with bundled Codex
- Local Codex thread with a long history
Steps to reproduce
- Quit and relaunch the desktop app.
- Let it restore/resume a large local Codex thread.
- Open Activity Monitor's Memory tab and observe the
codexprocess.
Actual behavior
The process grew in the captured run as follows:
- 13:43: ~116 MB
- 13:44: 2.61 GB → 5.05 GB → 6.28 GB → 7.25 GB → 7.84 GB → 8.65 GB
- 13:45: 13.82 GB
At the peak, memory pressure was red and swap usage reached 14.52 GB. Other Codex renderer/service processes remained in the few-hundred-MB range; the growth was concentrated in the main codex process.
Local app logs correlate the incident with resuming the thread: thread/resume took about 82 seconds.
Expected behavior
Resuming a thread should keep memory bounded and should not consume more RAM and swap than the machine has available.
Attachments
Eight screenshots are attached in chronological order, showing the process and memory-pressure growth.
1 Comment
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action