How it shapes AI answers
When you ask the Assistant a question in chat, Qubit gives it your metric definitions and your saved business objects along with the question. The same information is used when the Assistant generates a dashboard and when Qubit writes Insights.
What the Assistant receives
Metric definitions. For each active metric with a compiled definition: its key, name, description, calculation, filters, measurement type, reporting period, unit, and category. In chat it also gets each metric's latest value. The Assistant is instructed to reuse a matching definition exactly instead of rebuilding the number from raw columns.
Business objects and joins. For each saved business object: its table and database, description, primary key, and dimensions, plus each join with the columns it matches on and its cardinality. Joins that cross two databases are flagged. The Assistant is instructed to join tables on these keys instead of inferring a join from column names, and to prefer certified objects.
What you can control
- Descriptions. The Assistant reads the description of each business object and each metric. Say what one row represents and what a number includes or leaves out.
- Certification. Certified objects are marked as the definition your team vouches for. In a large model, they are listed first.
- Hidden dimensions. A dimension with
is_visible_to_ai: false, set in Code, is left out of what the Assistant receives. - Joins. A declared join gives the Assistant the exact keys to use. Find joins helps you find candidates in your data.
- Data Catalog descriptions. When the Assistant explores your tables, it also reads the table and column descriptions you write in the Data Catalog, and skips columns you exclude from AI analysis there. See Browsing connections.
Other names for the same thing
Qubit has no separate list of synonyms. If your team uses another word for something, such as "clients" for customers, put that word in the description of the business object or metric. When there are many metrics, Qubit lists first the ones whose name, category, or description share words with the question, so the extra word also helps the right metric reach the Assistant.
Without a model
If no business objects are saved, the Assistant still answers from your connected data. It reads the schema, writes its own SQL, and works out joins from column names. Your metric definitions are still used.
Data access
In chat, the Assistant only receives business objects and metrics built on connections and tables you are allowed to read.