Explore: Traces, Logs & Metrics

The Exploration explorers are where you go from a finding to the raw requests, lines and metric series behind it. There are three — Traces, Logs and Metric Explorer — and they share one query shell, so learning one teaches you the others. If you have used Datadog or Honeycomb this will feel familiar; this page covers what is specific to atrim.ai.

They live under Exploration → Traces / Logs / Metric Explorer. You usually arrive here from Critical Paths or Metric Signals — those surfaces hand off a query pre-scoped to what you were looking at — but each opens on its own with recent data loaded.

The shared shell

Every explorer has the same three parts:

• A natural-language-first query bar. Describe what you want in plain English — for traces, “slow failed checkout requests” — and Interpret turns it into a structured query and runs it in one step. The structured filter controls sit right below, always visible, so you can refine the interpretation by hand: the inferred filters appear as editable pills, and a pill that a later edit overrode is shown struck through rather than silently dropped.
• A results area with a Chart / Table / Both toggle (Both by default). The chart is derived from the rows on the current page — its caption says so — not a full-window rollup, so it is a fast read of what you fetched, not a separate aggregate query.
• An Insights rail on the right (open by default) that summarises the window.

Note

Interpreting a query is an LLM call that infers filters — it never executes anything on its own beyond running the query it produced. If you have already pinned a time range, an inferred one is noted and ignored rather than overriding your choice.

Each explorer’s query round-trips through the URL — ?tq= for traces, ?lq= for logs, ?mq= for metrics — so Back, refresh and a shared link all restore the same filters, natural-language text and time range.

Traces

Traces filters on the R.E.D. signals — Rate, Errors and Duration — plus Services and Endpoints (endpoints narrow to the services you pick). Results come as a table with a Traces / Spans / Errors / Avg Duration / Services summary bar and a Group by Trace toggle that flips between individual spans and whole traces.

From any row you can open the trace timeline, view its details, or filter by this row to pivot the query. The timeline itself is a waterfall with the critical path overlaid, a span tree you can search and expand, a minimap, and — below it — the logs that share the trace’s ID. Any one trace can be sent through Deep Trace Analysis, the atrim.ai agent, for a written read of where the time or the error went.

Logs

Logs filters on Levels, Services, Environment, and a body-contains substring. Two projections of the same window sit behind a Rows / Patterns toggle: Rows is the individual lines (Time, Level, Service, Message, Trace, expandable to the full body and attributes); Patterns groups them into templates with a count, severity and example traces, and drilling a pattern filters Rows to it. A Live tail switch polls the Rows view every few seconds for a running feed.

Tip

Logs is the one explorer with an agent behind its Insights rail. Press Analyze and atrim.ai reads the current log window and writes up what it finds. It is on-demand — the rail disarms whenever you change the filters, so editing a query never quietly spends another analysis.

Metric Explorer

Metric Explorer queries the raw OTLP metrics — filter by metric name, type, service, a value band (value > / value <), and environment. The results table lists each series with its shape, sources and min/max; click a row to chart that series, and ⌘/Ctrl-click to overlay a second. A value > or value < filter also draws as a dashed threshold line on the chart. Each point traces back to its exact requests through OTLP exemplars, so you can go from a metric spike to the traces behind it.

If a workspace has not sent OTLP metrics recently, the explorer says so plainly and points you at Traces and Metric Signals instead of showing an empty grid.

What the insights are

Each rail is honest about how it was produced. On Traces and Metric Explorer the Insights rail is a counted read of the rows you fetched — a fast, exact summary and a “what stands out” list computed by arithmetic, labelled as such. On Logs, the rail runs an actual analysis agent over the window. So an insight is never dressed up as more than it is: the explorers show you the raw data faithfully, and where an agent read is offered, it is a real one.

Critical PathsWhere a Traces or Metrics query often starts — request paths ranked, with 'open in Traces / Logs / Metrics' one click away.

Metric SignalsThe detected-metrics surface that hands a scoped query straight into Metric Explorer.