Windows Codex Desktop accumulates many residual node.exe processes (159 observed, ~7.26 GB RSS) during desktop/browser workflows
Summary
On Windows, Codex Desktop can accumulate a large number of residual node.exe processes during normal desktop usage. In one live sample, I observed 159 node.exe processes consuming about 7.26 GB of working set memory in total.
The process growth appeared in repeated batches over time rather than as a single one-time spike, and the machine became noticeably heavier as the count increased.
This looks related to the broader Codex Desktop child-process / MCP cleanup issues already reported for other platforms and workflows, but I am filing this separately because the Windows manifestation here is specifically a large buildup of node.exe processes during desktop/browser-oriented activity.
Update: this appears to overlap strongly with #25697 (Windows Desktop: managed node_repl helper fanout after project/chat activation). I added the concrete live-sample evidence there and am closing this as a duplicate so discussion can stay consolidated.
Environment
- Platform: Windows
- App: Codex Desktop
- Observed package path:
C:\Program Files\WindowsApps\OpenAI.Codex_26.602.3474.0_x64__2p2nqsd0c76g0\... - Observed additional runtime path(s):
C:\Users\<user>\AppData\Local\OpenAI\Codex\bin\... - Node executable path for leaked children:
C:\nvm4w\nodejs\node.exe
I did not capture the About dialog directly during sampling, so the version above is inferred from the running process path.
What I observed
At the time of sampling:
159totalnode.exeprocesses- approximately
7262 MBtotal working set across those Node processes - parent/child analysis showed that most
node.exeinstances were direct children of a singlecmd.exeprocess - their creation times clustered into repeated batches instead of random distribution
Observed batch counts by minute:
2026-06-09 12:02->172026-06-09 12:18->172026-06-09 16:57->172026-06-09 17:03->172026-06-09 17:52->172026-06-09 17:54->162026-06-09 18:09->242026-06-09 18:11->172026-06-09 18:12->17
Several of the largest residual node.exe processes were each around 80-114 MB working set.
Process-tree observations
This is the most important part:
- most of the residual
node.exeprocesses had the same direct parentcmd.exe - that
cmd.exeitself appeared to be associated with a Codex / browser-related process chain rather than a normal user-runnpm run devshell - the timestamps of repeated
codexruntime processes and the repeatednode.exebatches were highly correlated
This makes it look more like a Codex Desktop helper / browser / MCP / automation lifecycle problem than a normal project-local Node workload.
Why I think this is app-related
This did not look like a single project intentionally using a normal Node cluster or worker model.
Reasons:
- the processes accumulated in repeated batches over time
- the total count reached
159 - memory reached ~
7.26 GB - the process tree did not resemble a single foreground dev server
- the timing lined up with Codex desktop/runtime activity rather than an explicit user-launched long-running Node workload
Related reports
This may be related to the same general cleanup/lifecycle family as:
#20867Windows desktop becomes unresponsive with MCP auto-start duplicates#20349Codex Desktop accumulates duplicate MCP/helper process trees and multi-GB RSS#18881MCP child processes leak whenMcpConnectionManageris replaced#17832Playwright MCP stdio processes still leak after prior fix#17229Windows app repeatedly spawns Git/conhost processes and leaves residual children
I am filing this separately because this report is specifically about a Windows desktop session ending up with a very large number of node.exe residual processes and multi-GB memory usage.
Expected behavior
When browser helpers, MCP servers, automation sessions, or other Codex-managed Node helpers are no longer needed, Codex Desktop should fully tear them down.
At steady state I would expect:
- no unbounded growth in residual
node.exeprocesses - no repeated batch creation of helper processes that remain alive indefinitely
- helper memory usage to return close to baseline after tasks finish
- closing or idling the desktop app should not leave behind large Node process sets
Actual behavior
Residual node.exe processes accumulate over time and remain alive long enough to push memory usage into multiple gigabytes.
Repro status
I do not yet have a minimal clean repro, but the live process state was easy to observe once the problem appeared.
The issue seems more likely after heavier desktop / browser-oriented usage over time than after a fresh short session.
Useful diagnostic details captured locally
From local inspection of the running system state:
- main desktop process observed:
C:\Program Files\WindowsApps\OpenAI.Codex_26.602.3474.0_x64__2p2nqsd0c76g0\... - additional Codex runtime processes were observed under
C:\Users\<user>\AppData\Local\OpenAI\Codex\bin\... - multiple Codex runtime start times aligned closely with repeated
node.exebatch start times - most leaked Node processes used
C:\nvm4w\nodejs\node.exe
If needed, I can provide a more detailed process table / parent-PID mapping / memory breakdown format based on the live sample.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗