Codex Desktop Crashes While Idle as Computer Use Workers Multiply

Open 💬 3 comments Opened Aug 17, 2026 by derickpeters
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of the Codex App are you using (From “About Codex” dialog)?

Version: 26.810.52044, build 6662

What subscription do you have?

ChatGPT Pro 20x

What platform is your computer?

Hardware: Mac mini M4, 16 GB macOS: 15.7.8 (24G824)

What issue are you seeing?

Reproduction: launch application and leave idle; Computer Use instances multiply; crash after approximately five minutes
Activity Monitor: more than 20 Computer Use processes visible
Faulting thread: 315 computer-use
Immediate failure: node::OOMErrorHandler
Dispatch warning: soft limit of 64 blocked synchronous dispatch threads
Incident ID: 2390B917-4DEC-4128-B322-F6540B47B9CA

What steps can reproduce the bug?

Reproduction: launch application and leave idle; Computer Use instances multiply; crash after approximately five minutes

What is the expected behavior?

Codex should launch normally and remain stable while idle. It should create only the Computer Use workers required for active tasks, reuse or terminate them correctly, and prevent their number from continually increasing. Memory and thread usage should remain stable, the application should stay responsive, and it should not crash when no task is running.

Additional information

This is consistently reproducible after restarting the Mac, even when Codex is the only application running and no task is active. Within several minutes, Activity Monitor shows more than 20 Computer Use instances, memory usage increases, and the application crashes.

Environment:

Mac mini M4 with 16 GB RAM
macOS 15.7.8 (24G824)
ChatGPT/Codex version 26.810.52044, build 6662
Bundle identifier: com.openai.codex

The crash report identifies the faulting thread as computer-use, with SIGABRT, node::OOMErrorHandler, and the message: Dispatch Thread Soft Limit Reached: 64 (too many dispatch threads blocked in synchronous operations).

In the attached incident, the application launched at approximately 18:11:43 and crashed at 18:17:16 while idle. This appears to indicate that Computer Use workers are repeatedly being created or restored without being terminated correctly. I have attached the full crash report and Activity Monitor evidence.

Pasted text(2) (1).txt

View original on GitHub ↗

3 Comments

github-actions[bot] contributor · 10 days ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #38939
  • #38841
  • #38813
  • #38771
  • #38760

Powered by Codex Action

kengimai · 10 days ago

Confirmed on a separate Mac mini.

  • ChatGPT/Codex: 26.814.41407 (6720), which is newer than the version in this report
  • macOS: 26.6.1 (25G76)
  • Same failure: EXC_CRASH (SIGABRT) on the computer-use thread, with Dispatch Thread Soft Limit Reached: 64 (too many dispatch threads blocked in synchronous operations)
  • ChatGPT spawns a large number of SkyComputerUseService child processes before aborting
  • Disabling Computer Use in the app UI did not prevent the issue

This also reproduces after reinstalling and restarting the app, so it does not appear limited to the original build or a single machine configuration.

meet-cjli · 9 days ago

Adding another data point — reproduced on BOTH bad builds, fixed by downgrade.

Machine: Mac16,10 (Mac mini M4), 32 GB RAM, macOS 26.x

Timeline:

  • 26.810.52044 → crash loop all day (2026-08-16), ~2 min after every

launch, faulting thread alternates between computer-use and
V8Worker, always node::OOMErrorHandler
reading_mode$...$parse_distilled_html

  • Updated to 26.814.41407 (latest at the time) → still crashed

(2026-08-17 20:42, same signature)

  • Downgraded to 26.803.81509 (framework 151.0.7922.76) → completely

stable for 24h+ now: flat ~0.4 GB memory, exactly 1
SkyComputerUseService process (vs ~190 leaked threads before)

Ruled out on my machine: clearing Cache/Session/Local Storage,
deleting large conversations, disabling automations, quarantining all of
~/.codex (sessions/logs/plugins), full reinstall. Logged-out = stable,
logged-in = crash even with a stripped profile. Disabling Computer Use
in the UI does not help.

The working theory from #38813 (CUA helper version mismatch → blocked
AESendMessage handshakes → worker thread leak → V8 OOM) matches
everything I observed, including why 26.803 + its older helper pairing
is stable.