Skip to content
AI.info

The Pulse

Codex Desktop Leaves 323 GB of Crash Dumps on One Mac

A Mac user found about 323 GB of Crashpad files linked to OpenAI’s Codex desktop app, leaving just 1.4 GB free on a 926 GB disk. The app kept working without warning, and the user says the files accumulated on a current build.

Codex Desktop Leaves 323 GB of Crash Dumps on One Mac

AI.info Team · Written with AI ·

323 GB accumulated while the app kept working

OpenAI support said in September that it could not confirm a fix for Codex’s growing Crashpad files. On October 17, a user posted measurements showing the problem still present on a newer desktop build: about 323 GB of crash-report data on a Mac with a 926 GB disk, leaving only 1.4 GB free.

The user, posting as ahmedkhan25 on GitHub, measured the files from October 15 to 17 while running ChatGPT.app 26.1007.71326, build 20398, on macOS 15.7.4 and Apple Silicon. The app remained usable, and the user said it gave no warning before the disk nearly filled.

“Nothing in the app warned me.”

ahmedkhan25, GitHub issue commenter

The report landed in an OpenAI Codex repository issue first filed on June 2, when a user documented 4.9 GB and 54,504 files in a single day. The October measurement shows the two Crashpad folders holding roughly 4.04 million files combined.

Two Crashpad folders held the backlog

The larger folder, under ~/Library/Application Support/com.openai.codex/web/Crashpad, accounted for 180 GB and more than 2.25 million files. Another 143 GB sat in ~/Library/Application Support/Codex/Crashpad, mostly in its pending subfolder.

The user found the buildup by scanning the Mac’s Library directory with du. Renaming both Crashpad folders and deleting them freed more than 50 GB within minutes, up from 1.4 GB of available space. The deletion was still running when the user posted the update, a sign of the burden created by millions of small files.

Crashpad stores diagnostic reports when app processes or helpers fail. In the first report on the GitHub thread, the files appeared in matching pairs: a .dmp minidump and a _sidecar.json file. Users have avoided attaching the dumps publicly because they may contain memory contents, local paths or other private information.

Users report different sizes and triggers

The June report described a background handler problem rather than Codex becoming unusable: the app continued to run while pending reports grew. In a later comment on the same issue, another user reported about 100 GB and roughly one million files in the newer Codex/Crashpad/pending location.

A separate October 8 report documented a 168 GB backlog that began growing again after cleanup. That user later traced the repeated reports to the browser_crashpad_handler crashing and being relaunched; a reboot stopped the growth on that Mac. The account is evidence of one user’s case, not proof that it explains the separate 323 GB measurement.

OpenAI Support had told a Mac user on its developer forum on September 7: “We can’t confirm a Crashpad-specific fix yet.” It asked for app and macOS versions, the folder location and the rate of growth so the company could review diagnostics. The later 323 GB report is from a newer build than the one cited in that support exchange, but the post does not establish what component caused that user’s accumulation.

Cleanup advice has not established a fix

OpenAI Support’s forum response recommended sharing diagnostic details and did not confirm a dedicated fix. In the separate 168 GB case, support advised clearing pending reports with the app closed and keeping the app updated; the reporter said an update was offered but did not confirm reproduction on that version.

The October 17 user asked for an automatic size or file-count limit, a fix for the handler loop and an in-app warning when Codex data grows beyond a few gigabytes. Those are requests in a public bug report, not announced product changes. The clearest result so far is local: deleting the two reported folders restored more than 50 GB, while the app was still running.

Sources

Explore

More articles