ጠይቅ

በራስ ቃላት የሚጠየቅ ጥያቄ: ጥያቄው ወደ መለኪያ ጥያቄ ይተረጎማል፣ ጥያቄው ራሱ በማያ ገጹ ላይ ይቆያል፣ ውጤቱም እንደ ገበታ እና ዓረፍተ ነገር ይመለሳል። የሚታመነው ነገር ተደብቆ አይቀርም።

Asking

Ask opens from the sidebar on any screen. Enter sends, Shift+Enter breaks the line, and a question can be asked in any of the languages the console speaks.

What comes back

The question is translated into the same metrics query language the Analytics workbench speaks, and that query is printed beside the result rather than hidden: you are being asked to trust a machine translation of your own words, so the translation stays inspectable. The query then runs like any workbench question and draws with the same chart, with the model's prose over the numbers limited to describing what the arithmetic under it shows. The elapsed time is measured and shown, and Copy takes the answer text.

When it cannot answer

Every failure keeps its shape: a question the model could not turn into a runnable query says so, a query that failed to run keeps the translation on screen with its own retry, and a plan that does not include asking is a named refusal with a step to the plans page rather than a generic error. Which model answers is decided by the account's plan.

Where this meets the API

The concept is the AI pillar. On the wire: translation is aiTranslateQuery, execution is the ordinary runQuery, and the prose is aiAnswer over the returned points (GraphQL reference). The model can never emit free SQL: it emits the validated query language, and the same validator that guards the workbench guards it.