Observe
Read one request end to end as a trace, search the logs and events underneath it, and aggregate many traces into volume, error rate, latency, tokens and cost.
The Observe pillar is where you read what happened: traces, logs and analytics, all views over the one events table.
Traces
A trace is one request seen whole: the agent run, every LLM call, tool call, retrieval and HTTP hop underneath it, joined by a trace ID. The trace view is a waterfall, with an errored span called out so you can see where a request broke rather than where it ended, and an inspector for the selected span’s attributes and payload (subject to your content policy).
Filter the list by service, model, agent, status, duration and more. The
endpoints behind it are GET /traces and GET /traces/{traceId}.
Logs and events
Every app.log, agent.log and custom event lands on the same timeline as the
spans. Free-text search across events spans the whole organisation with
GET /events, so a log line and the trace it belongs to are one click apart.
Analytics
Analytics aggregates the trace stream into the numbers you watch: volume,
success rate, p95 latency, tokens and attributed cost, each with a trend over
the window, plus a status mix and the models in production. The filter bar scopes
every tile at once, and the same filters exist on Traces, so a number and the
requests behind it always agree. The endpoints are GET /analytics/timeseries,
GET /analytics/breakdown and GET /analytics/latency.
To slice by your own business attributes — not just model or status — see Usage & dashboards and custom dimensions.
Related
- Usage & dashboards, slice by custom dimensions and save boards
- Traces (Python SDK), reading traces back from code
- Events model, the store all of this reads from
- Traces API · Analytics API
Gateway
Where traffic gets in. Apps and channels organise your events, keys and tokens authenticate them, and the SDK, OTLP and Sentry paths carry them.
Usage & dashboards
Slice telemetry by any dimension your app emits in the Usage explorer, then save a dashboard tailored to your business — built by you, or authored by your coding agent over MCP.