Skip to content
GitHub
Start a 14-day trial

Ask in plain English

You don’t need a query language to search your telemetry. Type “slow failed checkout requests” into Traces and press Interpret: I turn the sentence into filters you can see, run the query, and write one line under the bar saying exactly what ran. Every filter I chose is a pill you can edit or remove.

The Traces query bar after interpreting slow failed checkout requests. The bar holds the question as a pill, then three filter pills — service : checkout, duration > p95 and error — each with an edit pencil and a remove ×, followed by Add more… and the Interpret button. Under it the line reads: Traces containing a span from checkout, at or above the window's p95 duration, with an error, in the last hour, newest first. Below are the Rate, Errors, Duration, Services and Endpoints filters and the result count, 43 traces.

The same bar heads all three explorers, each with an example to start from:

Explorer The bar suggests
Traces “slow failed checkout requests”
Logs “errors in checkout in the last hour”
Metric Explorer “cpu utilization over 0.9 in the last hour”

On Logs and Metric Explorer, set the window with the range control: a time in the question, like the “in the last hour” above, doesn’t change it (see Time ranges).

Name things the way your telemetry does — a service, an endpoint, a metric, a level — and describe the rest in your own words.

Press Interpret or Enter. When the question is clear, the query runs straight away; there is no second step.

  • The filters appear as pills in the bar. Click a pill, or its pencil, to change the field, the comparison or the value; × removes it. Either way the query runs again.
  • A line under the bar describes, in plain words, the query that actually ran — the services, conditions, time range and order. If a filter couldn’t be applied, the line says which one.
  • The filter controls below stay in view, so you can refine by hand whenever that’s quicker.

On Traces, each question starts fresh: it replaces the previous question’s filters, and the question itself stays at the front of the bar as a pill. Click that pill to put the text back in the box and rephrase it. A single word, such as an operation name, searches span names directly. On Logs and Metric Explorer, each question adds its filters to the ones you already have; remove a pill with its × to widen the query again.

On Traces, a time in your question — “in the last hour” — sets the range, unless you’ve already picked one. If you have, your range stays and a note says which time from the question was set aside. On Logs and Metric Explorer, the time-range control alone sets the window; pick it there.

I pick a reading and run the query, and a notice titled Interpreted with assumptions shows each reading — the word you typed and what I took it to mean. If part of the question couldn’t become a filter, the notice lists those words under Not applied, with the reason, and the rest of the query still runs.

If nothing in the question maps to your data, nothing runs: your previous results stay on screen, your text stays in the box for a rephrase, and a notice names the words that didn’t fit. Where the rest of the question still narrows the results, the notice offers a best guess without those words — press Apply and run to use it. Otherwise it suggests rephrasing with a name the explorer knows: a service, a metric, a field.

The filters ride in the page address, so a refresh or a link you send a teammate restores the same pills and range. The wording of your question isn’t kept in the link — the filters it produced are.