Linux Codex Desktop IAB Chromium grows to 21.9 GiB, is killed by systemd-oomd, and exits the app

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

26.818.61809

Bundled browser and Chrome plugins are also 26.818.61809.

What subscription do you have?

Personal ChatGPT account (exact tier omitted; this appears to be a local process-memory issue).

What platform is your computer?

Linux 6.8.0-51-generic x86_64 — Ubuntu 24.04.4 LTS, GNOME on X11, 32 GiB RAM and 8 GiB swap.

What issue are you seeing?

Codex Desktop's Browser Use / in-app-browser Chromium scope grew to a 21.9 GiB memory peak and 3.0 GiB swap peak. Linux systemd-oomd killed that Chromium scope. The main ChatGPT/Codex Desktop app exited less than one second later.

This was initially mistaken for ordinary Google Chrome consuming the memory, but the systemd scopes distinguish them:

• Suspected Codex IAB scope: app-org.chromium.Chromium-2462411.scope — 21.9 GiB memory peak, 3.0 GiB swap peak, OOM-killed.
• Ordinary Google Chrome scope: app-gnome-google-chrome-2327726.scope — only about 190 MiB swap at the decision point.
• Codex Desktop scope: app-gnome-chatgpt-2462411.scope — ended about 0.7 seconds after the Chromium OOM kill.

The shared 2462411 suffix and the preceding Codex log entry showing IAB startup strongly suggest the large Chromium scope belonged to the Codex in-app browser rather than foreground Google Chrome.

Relevant redacted journal timeline:

09:07:44 browser_use_iab_backend_startup_started backend=iab platform=linux release=26.818.61809
09:07:44 browser-use native pipe listening pipePath=<redacted>
09:07:44 browser_use_iab_backend_startup_ready backend=iab platform=linux release=26.818.61809

09:34:28 systemd-oomd: Considered 157 cgroups for killing, top candidates were:
09:34:28 systemd-oomd:   app-org.chromium.Chromium-2462411.scope (Swap Usage: 665.5M)
09:34:28 systemd-oomd:   app-gnome-chatgpt-2462411.scope (Swap Usage: 370.0M)
09:34:28 systemd: app-org.chromium.Chromium-2462411.scope: systemd-oomd killed some process(es) in this unit.
09:34:28 systemd: app-org.chromium.Chromium-2462411.scope: Failed with result 'oom-kill'.
09:34:28 systemd: app-org.chromium.Chromium-2462411.scope: Consumed 5h 13min CPU time, 21.9G memory peak, 3.0G memory swap peak.
09:34:29 systemd: app-gnome-chatgpt-2462411.scope: Consumed 2h 42min CPU time.

Immediately after the kill, the system recovered to about 28 GiB available memory, supporting attribution to the terminated scope rather than a persistent machine-wide leak.

What steps can reproduce the bug?

I do not yet have a deterministic short reproduction.

Observed sequence:

1. Start Codex Desktop on Ubuntu/GNOME X11.
2. Use Codex normally with Browser Use enabled and start an in-app-browser (iab) session.
3. Leave the app active across longer-running work and multiple tasks.
4. The IAB-associated Chromium scope grows until systemd-oomd kills it; Codex Desktop then exits.

Local session-store context, in case it is relevant:

• ~/.codex/sessions: 3.1 GiB total
• 325 files
• Largest individual rollout: about 149 MiB
• Several rollouts are over 100 MiB

No session contents are attached.

What is the expected behavior?

Browser Use/IAB memory should remain bounded. If the browser backend exceeds a safe threshold, it should be restarted or fail gracefully without exhausting system memory or terminating the full Codex Desktop app.

Additional information

Possibly related reports:

• #22591 — Codex Chrome native host unbounded memory growth on Windows
• #38566 — repeated Codex Desktop Chromium/V8 OOM crashes
• #29510 — Codex app-server grows to 30–40 GiB with large rollout history
• #21233 — Linux Codex runtime killed by cgroup OOM

This report differs by being a Linux Codex Desktop iab/Chromium reproduction on build 26.818.61809, with systemd scope attribution and a 21.9 GiB measured peak.