Docs › Reference

What Ask understands

Every kind of question the box can read, with examples.

The Ask box reads a question by matching its words against your repository's own vocabulary: the names of its types, attributes, roles, domains and metrics. It does not guess. Whatever it understood is shown in the sentence above the results, and anything it could not read is searched for as plain words.

Because the vocabulary comes from your metamodel, your repository may use different words from the examples here. If your administrator renames Tables to Datasets, which datasets hold customer email works with no other change.

How a question opens

The first words decide what kind of answer you get.

A question that opens any other way is read as a list, and so is a search of plain words.

Wildcards

Two characters mean something of their own, wherever they appear in the box.

* on its own is the whole repository, with the rail on the left counting it by kind — the quickest way to see what is in there. It is meant to be narrowed, and everything else in this page still applies on top of it: * in Banking, * from Snowflake, * tables, or * with conditions added underneath.

A pattern is matched against two things: an asset's name and its external key. Those are the two things anybody pattern-matches on. It is not matched against definitions, comments or the words around an asset — that is what words are for, and %cust% and customer are different questions.

The underscore is an ordinary character here, not a single-character wildcard as it is in SQL, because half the names in a harvested repository are CUSTOMER_ID and ORDER_DATE. So %_id finds names and keys that end in _id, and nothing else.

A pattern needs at least three characters beside the %. ab% is refused, with a sentence saying why, rather than served slowly: two characters means reading every asset in the repository, and an answer that takes a minute is not an answer.

Wildcards narrow, like every other part of a question. orders cust% is the things matching orders among those whose name or key starts with cust; two patterns narrow each other the same way. Each one gets its own ✕ in the sentence above the results.

The same two wildcards work on a list with conditions and in a saved search, and mean the same thing there.

Kinds of thing

Name a type, in the singular or the plural, and the answer is narrowed to that type: tables, column, business terms. The sentence shows it as Showing Tables. If you name two types, the first one is the kind of answer you want.

Words such as assets, things, everything and anything mean you do not want the answer narrowed to one type.

Marks and values

Name a yes-or-no attribute and Ask keeps only the assets where it is yes. Columns with PII gives columns marked Contains PII. The last word of an attribute's name is enough.

For an attribute with a fixed list of values, name the value, with or without the attribute: business terms that are approved and status approved both give whose Status is Approved. Text attributes are not read this way; their words are searched for.

In a domain

Put a domain's name after in, within, inside, under or across, and the answer is narrowed to everything that domain answers for, including what sits inside the things it owns. A domain answers to:

• its full name, such as Banking & Financial Services;
• its short code, such as BFSI;
• the first word of a longer name, such as Banking;
• any synonyms recorded for it.

A word that two domains would both answer to belongs to neither. The domain's name on its own is still an ordinary word: typing customer looks for customers. When a word you typed is exactly a domain's name, Ask offers Everything in that domain on the right.

From a source

Put a source after from, in or of, and the answer is narrowed to what that source catalogued. Words such as catalogued, harvested or brought in before it are read with it. A source answers to:

• the product's name, such as Power BI or Snowflake, for everything any connection to it catalogued;
• a connection's name or code, such as Power BI MDA or POWERBI_MDA;
• the name of a system a harvest found, such as Snowflake — ACME account.

So how many assets catalogued from Power BI counts every asset a Power BI source brought in, and tables from Snowflake lists only Snowflake's tables. Only harvested assets count: something written in KRISID and filed under a Snowflake schema did not come from Snowflake. A source's name on its own is still an ordinary word.

You can also put the source's name straight in front of what you are asking for — Snowflake assets, Snowflake tables, Power BI reports — which means the same as tables from Snowflake. Both parts are read: the source narrows it, and the kind narrows it again. The name has to stand immediately before the kind for this; a source's name on its own is still an ordinary word, so you can still search for something that happens to be called Snowflake.

Catalogued is a thing that happens once, so each asset belongs to exactly one source in these answers: the one whose harvest first brought it in. A table your BI tool merely reports on is still the warehouse's, and is not counted twice. An asset harvested so long ago that no source was recorded against it is counted from the system it sits inside, which is the only case where where-it-sits decides the answer.

What is connected to something

Phrases such as mapped to, linked to, related to, connected to, associated with, belongs to, part of, attached to and owned by ask what is linked to something. The answer lists everything one link away, in either direction, and says how many: 7 things linked to Customer. Put a type before the phrase to narrow the answer: which tables are mapped to the Customer domain.

A question such as what are the assets in the Customer domain, which names no kind of thing, is read the same way.

Where is it

A where question is about real, physical things.

• If the thing you ask about is harvested, such as a column, Ask names what it sits in, all the way up.
• If it is a business thing, such as a term, Ask lists the physical assets linked to it.

Either way the answer gives the name to query, a query to start from with a Copy SQL button, and under to run it, which roles could read it. If the table holds marked columns, a careful line says so before the query. Ask never runs the query.

Asking for a number

Name a metric from your semantic model and ask for a number, and Ask composes a query instead of listing assets. It does that when you:

• slice it with by, per or split by and a dimension: net revenue by region;
• give a period: last quarter, this month, last year, the last 30 days, 2025;
• ask for a grain: monthly, by week;
• open with how much or how many.

The answer shows the query, the grain it used and which roles could run it. It is never run. If you slice by something the model does not have, Ask refuses and lists what it can be sliced by. A metric named on its own, with no slice or period, is found like any other asset. See Metrics and asking for a number.

Examples

Taking a reading off

Every part of the sentence has a ✕. Click it to take that part off and keep the rest. Click just the words instead to ignore the whole reading and search for your words alone, and read it as a question to go back.

Good to know

• Small words such as the, of, with and please are ignored.
• When the glossary says one of your words means something else, Ask searches for both and says so under also searched for.
• A question with conditions that finds nothing is never a dead end. Ask shows the nearest answers, each with its count.

Related

• Ask: search and questions
• Metrics and asking for a number
• Troubleshooting and FAQ
• Glossary