dev.cocore.compute.modelActivity
Aggregate receipt activity per model and per time window. Walks the most-recent indexed receipts (capped at 5000), groups by `body.model`, and tallies requests + tokens across the hour/day/week/month windows, with per-provider breakdowns. The scan cap is reported back as `scanned`/`scanLimit`/`truncated` so consumers can tell an exact total from a lower bound; `oldestScannedAt` is the earliest receipt that contributed to any window.
Details · query
| Kind | query |
|---|---|
| Authority | compute.cocore.dev |
| Indexed |
Query
query · GETOutput
application/jsonoldestScannedAt
string · datetimeEarliest receipt timestamp that contributed to any window, so a consumer can state the coverage of the counts. Absent when nothing counted.
scanLimit
integerCap the scan applied (receipts are walked newest-first).
scanned
integerNumber of indexed receipts this scan walked.
truncated
booleanTrue when the scan filled its cap, so older receipts exist that these totals do not include. The wider windows (week/month) are then lower bounds.
Similar lexicons
semantic matchLexicons whose schemas describe something close to this one. Check them before publishing a new one that overlaps.
-
Network latency rollup derived purely from indexed receipts' signed `startedAt`/`completedAt` pairs — the last ≤100 per group (overall, per-provider, per-model). No side metrics store: the receipts are the source of truth, so the numbers cannot drift from them.