dsh-commandcode-quota

LIMITATIONS

Known limitations

- **Polling, not push.** The panel refreshes every 60 seconds (15 seconds while a window is approaching its cap, and never once a window is spent), with a 15-second host cache on top. Credit changes can lag by up to a minute. - **No per-model allowance breakdown.** Command Code allocates a per-model share of the monthly budget, but the `/alpha` endpoints do not expose that table, so the card reports the total only. - **No history.** Every read is a live snapshot; nothing is stored except the one cached report. - **Command Code only.** This does not replace DSH's own local token statistics (`$DSH_HOME/dsh-usage/`). - **The catalog follows one plan at a time.** A check fetches the page for the plan you are on, plus the shared pricing page that maps plans to models and carries the plan-level estimate — not every tier's page. Switch plans and the next check picks the new one up. - **Provider and Teams publish no per-model table.** Both show the plan-level figure only (Provider is metered, Teams publishes one number). Ultra has no page of its own; it is shown from Max's table and labelled as an inference. - **The vendor's own latency is the read's deadline.** Healthy, all four endpoints answer in about a second; each gets 30 s before its reading is dropped, and `COMMANDCODE_QUOTA_TIMEOUT_MS` overrides that. A dropped reading is never filled in with an older number: the card says how many are missing and names them on hover. - **Framework seams.** See [Compatibility](#compatibility) — a future dsh release will need a look.