AI Platform · Observability
If it is not traced, it did not happen
Every agent run emits spans for policy load, retrieval, each tool call, the model call with its reasoning budget, the compliance verdict and the trace write. Complaints are answered from the trace, not from memory.
Traces today
612k
Trace completeness
100%
a dropped span fails the run
Tokens today
530M
in + out across the fleet
SLOs breached
1
Sample trace — a Stoxo research answer
Root span 4820ms · span widths are to scale
Latency distribution
Share of runs by wall-clock time
Token burn
Millions of tokens per day, last 8 days
Service levels
Published targets, measured continuously
Stoxo answer p95
target < 10s · actual 9.4s
within budgetSupport first reply p95
target < 30s · actual 18s
within budgetVoice turn latency p95
target < 900ms · actual 870ms
within budgetCompliance verdict p95
target < 5s · actual 6.1s
breachedGateway error budget burn
target < 20%/mo · actual 11%
within budgetTrace completeness
target 100% · actual 100%
within budget