Usage analytics
⚠️
Preview — this feature is being built; the design below is the spec.
The Console’s Analytics page and the analytics endpoints answer the questions finance and platform owners ask most: how much, by whom, on what, and is it working.
| Question | View |
|---|---|
| How are we trending this month? | Summary: requests, tokens, spend, active keys and members, against the previous period. |
| Which models carry our traffic, and how fast are they? | By model: volume, cache hit rate, latency percentiles, error rate. |
| Which systems spend the most? Which keys are idle? | By API key |
| Who uses what, for chargeback? | By member |
| What are web search and code execution costing? | Server tools |
| Are our agents finishing their jobs? | Agents |
How the numbers relate
- Spend in analytics equals the cost report for the same range: credit-billed usage only. Usage inside plan allowances counts tokens but zero dollars.
- Analytics are near real time (a few minutes behind); the cost report is final after the day closes.
- Times are UTC.
Useful habits
- Watch cache hit rate per model: a sudden drop usually means a prompt change broke a shared prefix.
- Watch error rate separately from 429s: errors point at bad requests, 429s at capacity.
- Review idle keys monthly and disable them; an unused key is pure risk.
Was this page helpful?