You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
My 5 hour window has been dropping despite no messages being sent. Simply logging in and opening the Codex app caused a 40% usage drop.
The Codex analytics dashboard shows the same 40% drop, but also shows no usage for June 20th.
Yet:
What steps can reproduce the bug?
Log out of Codex app.
Log into Codex app.
See your 5 hour window magically disappear.
What is the expected behavior?
No usage should drop until a message is sent, and limits shouldnt drop 40% just by opening the app.
Additional information
Note that the 5 hour window began on June 20th, 1:07 pm (as proven by resetting at 6:07 pm), which is when the app was opened. despite no message being sent also being reflected in the Codex analytics dashboard.
Quota drops without user messages need charge receipts by origin. Split foreground prompts from background startup, plugin sync, MCP init, thread resume, index rebuild, retry, and server-side adjustment.
For each quota-window delta, record timestamp, request id if any, client surface, model, quota units charged, token buckets when available, and origin. Then the UI can say "40% used by background plugin startup" or "server-side adjustment" instead of making users infer from an empty message history.
Activity
github-actions commentedon Jun 20, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
mohammad-malik commentedon Jun 20, 2026
Note that the CLI does not seem to be affected.
Necmttn commentedon Jun 21, 2026
Quota drops without user messages need charge receipts by origin. Split foreground prompts from background startup, plugin sync, MCP init, thread resume, index rebuild, retry, and server-side adjustment.
For each quota-window delta, record timestamp, request id if any, client surface, model, quota units charged, token buckets when available, and origin. Then the UI can say "40% used by background plugin startup" or "server-side adjustment" instead of making users infer from an empty message history.
Generated with ax.