Know where to look next.
Find the patterns in your AI’s quality, speed, and cost. Move from a change in the numbers to the requests that explain it.
P95 LATENCY
3.8sThe slowdown starts with retrieval.
The requests in this window spend most of their time waiting for the search tool.
THE QUESTION THAT MATTERS
What needs my attention next?
01See the pattern.
02Narrow the question.
03Find the requests behind it.
DASHBOARDS02
Start wide. Then ask a sharper question.
Compare quality, latency, usage, and estimated cost across your recorded activity. Filter by workflow and time range, then inspect the records behind the metric.
Choose what matters
Combine reusable widgets for request volume, latency, model usage, recorded cost, and evaluation results.
Keep the context
Use date ranges and filters to narrow the view and share the same investigation context.
P95 latency for each step of your application.
Daily span volume by operation.
ALERTS03
Set the signal that deserves your attention.
Choose matching events or a count threshold over a time window. Get an in-app or webhook notification when the rule triggers, with a cooldown to limit repeats.
Define a useful trigger
Start with a template or configure event filters and time-window count thresholds for your project.
Control the noise
Use cooldowns to keep repeated matching activity from overwhelming your notifications.
A rule with a clear reason to act
10 failed requests in the last 5 minutes.
View matching traces →THE DISCOVER TOOLKIT
From a signal to a focused investigation.
START WITH ONE REQUEST