Usage & rate limits

See what your keys are doing, understand the hourly limits, and know what to do if you hit one.

Your usage page

Open Usage in the sidebar. It shows:

Range — switch the charts between the last 24 hours, 7 days, and 30 days. Days follow your timezone.
API calls, Skills used, Devices — totals for the range you picked, compared with the period just before it.
Keys — each key's name and prefix. A key that is still active also shows its downloads in the rolling hour against your limit (for example 4 / 10 · 1h).
Top skills — which of your skills have been downloaded most in the selected range.

The limits

There are two limits, both per key and measured over a rolling hour (the last 60 minutes, not the clock hour):

Limit	Applies to	Amount
Downloads	Pack downloads (`read` and `install` of a new version)	Your plan's downloads per hour (see Plans & billing)
All requests	Downloads, manifest fetches (`list`, `update`) and searches together	10 × your downloads-per-hour

For example, on a plan with 10 downloads per hour a key can make up to 10 downloads and up to 100 requests in total in any hour. Administrator accounts have no limits.

What counts

You run	Counts as
`skilvault list`	1 request
`skilvault find …`	1 request
`skilvault read <name>`	1 request, plus 1 download the first time you fetch that version on the machine
Daily local refresh	1 request, plus 1 download per skill that changed

When you hit a limit

The server answers 429 and the client reports the request failed:

HTTP/1.1 429 Too Many Requests
Retry-After: 3600
{"error":"rate limit exceeded — retry within the hour"}

Because the window rolls, older requests age out steadily — you regain capacity gradually, not all at once. If you need more headroom, upgrade your plan, or spread work across keys (each key has its own allowance). See Troubleshooting.

Using fewer downloads

Packs are cached by checksum, so a version is only ever downloaded once per machine. Repeat reads of the same version cost a request but no download.
On CI, cache ~/.cache/skilvault between runs.
Give parallel CI jobs their own key so they do not share one allowance.