Repository navigation
@ccusage/codex: It's really slow and uses 30GB of memory #885
Description
I think it should be effectively instant (it takes several minutes on a very fast machine), especially on re-runs, and also not use that much memory. I don't think I've used an obscene amount or anything, maybe 100B tokens.
Thank you
Reacted by denissalanta-codecrafters
Activity
AshyEarl commentedon Mar 17, 2026
May be try my toy project: https://github.com/AshyEarl/modelUsage
doublemover commentedon Mar 17, 2026
Please dont shit up issues with spam. It's also wrong btw
AshyEarl commentedon Mar 22, 2026
Sorry, I shouldn't have posted this here. I just ran into the same issue myself and wrote a small tool hoping to help others with similar problems. My apologies.
ryoppippi commentedon May 17, 2026
The current Codex adapter no longer uses the old @ccusage/codex whole-file loading path. It scans JSONL by markers and uses worker-based parsing so only relevant token_count lines are processed. The streaming work from the #953/#952 area is also present, so the tens-of-GB memory problem reported here is resolved on current main. Further Codex performance improvements should be tracked in #877 and #922.