Codex limits and OpenAI credits
Mr. Usage uses an unexpired Codex login from its auth.json file, or OpenCode’s ChatGPT OAuth login, to fetch available OpenAI plan limits. The Limits tab displays server-defined usage windows, reset times and additional model or feature allowances without assuming every plan has the same windows.
It can also display the usage-credit balance reported by OpenAI, including granted credits reflected in that balance. Credits are provider billing units for additional usage, not dollars in the app’s API-equivalent cost chart. Mr. Usage does not purchase credits or change your account’s allowance.
This Mac: local Codex and OpenCode history
This Mac counts supported Codex session and archived-session logs, plus OpenAI messages in OpenCode’s database. Per-response records take precedence over running totals. Copied responses and forked OpenCode sessions are deduplicated so repeated records do not inflate usage.
This view is useful for understanding work performed on one computer. It may not include requests from another Mac, an IDE on a different device or cloud tasks. Codex’s experimental compressed .jsonl.zst records are not currently read.
All devices: reconciled account totals
All devices uses ChatGPT’s daily account history across supported CLI, IDE, desktop, cloud and other-machine activity. Complete account totals replace overlapping local Codex and OpenCode activity by UTC day. Local records fill days the account has not yet reported; the app does not add both sources for the same reported day.
The account service supplies daily totals and model shares rather than a complete token-kind breakdown. Input, output, cache proportions and API costs in this view are therefore estimates. Where enough local activity is available, its token mix informs the estimate; otherwise the app uses a typical mix. The app and leaderboard identify account-wide estimates.
Account history can lag live plan limits, and today’s daily total may be absent. This is why All devices and This Mac can temporarily differ, even after a refresh. A corrected account day replaces the earlier value rather than accumulating another copy.
OpenAI API-equivalent cost is not the subscription fee
The app calculates estimated cost using published Standard-tier API rates for supported models and short-context prompts. Cached input is priced separately from uncached input. Reasoning output is already part of output tokens and must not be added again.
A ChatGPT subscription can record zero API cost in a local tool while still representing a nonzero API-equivalent workload. Mr. Usage recomputes that estimate from token counts. Your actual subscription fee, extra credits and any API-key charges remain separate measurements.
Based on the app’s published documentation. Provider behavior and features can vary by installed version. Read as Markdown.