Observability overview
Operate › Overview sums up a tenant’s traces over a time range: how many calls the tenant’s models took, how slow the slowest 5% of calls were, what the tokens went to, and how many calls ended in an error.
Before you start
Section titled “Before you start”The overview is built from traces, so it needs tracing on for the platform. When it is off, the page says Observability is not configured. See Traces.
Who can open it
Section titled “Who can open it”Admin, Builder and Auditor, and platform admins.
What you see
Section titled “What you see”- The page header, with Time range (Last 24 hours, Last 7 days or Last 30 days; Last 7 days by default) and the tenant picker.
- Five figures for the range:
| Figure | What it is |
|---|---|
| Model calls | Traced calls to the tenant’s models. |
| Latency, slowest 5% (p95) | The 95th-percentile latency at the worst point in the range. |
| Tokens | Tokens used, all models together. |
| Cost trend | An estimate from the traces, to follow a trend. |
| Errors | Calls that ended in an error. |
- Three charts:
- Calls over time: model calls per hour (last 24 hours) or per day (7 or 30 days).
- Tokens by model: tokens per model across the range.
- Latency p95 over time: the 95th-percentile latency, in milliseconds, per hour or per day.
When tracing is on and nothing was recorded in the range, the page says No traces in this range: run a query in the Playground, an assistant or an agent, or widen the range.
If one chart cannot be loaded, it says so in place and the rest of the page still shows. The page lists what failed under Parts of this overview could not be loaded, with Try again.
What you can do
Section titled “What you can do”Change the time range
Section titled “Change the time range”- Open Time range.
- Choose Last 24 hours, Last 7 days or Last 30 days.
The range is kept in the page address, so you can share the view as a link.
Look at another tenant
Section titled “Look at another tenant”Choose it in the tenant picker. The picker appears when you can see more than one tenant.
Investigate a spike
Section titled “Investigate a spike”- Note when it happened on Calls over time or Latency p95 over time.
- Open Traces and find the requests from that time.
Limits
Section titled “Limits”| Limit | Value |
|---|---|
| Time ranges | 24 hours, 7 days, 30 days |
| Chart granularity | Hourly for 24 hours, daily otherwise |
| Data | What the traces still hold within their retention window |