Skip to content

Ask about your data

A Data Source is a database or data warehouse of your company that the agent can query directly. Someone with the Right to create Data Sources connects it in the Data Studio (Connect a Data Source), and you need access to it to ask about its data.

On Start, where you write the first message of a new Conversation, select Data Sources below the message field. It lists every Data Source you can query, each with a tick: all are selected. Select one to untick it, and the agent won’t use it in this Conversation.

If you can’t query any Data Source yet, the list says so: ask someone who manages Data Sources in the Data Studio to give you access to one.

Your first message fixes the Data Sources for the whole Conversation. Inside a Conversation, the same button lists them, read-only. If you lose access to one later, the agent can no longer query it there and says so if you ask about it.

Ask your question in your own words, such as “What was revenue by region last quarter?”. The agent:

  1. looks at the Data Source’s tables and columns first, so it doesn’t guess names;
  2. writes a query in the database’s own SQL and runs it;
  3. answers from what the query returned. It never makes up or rounds numbers that weren’t in the result.

If the database answers with an error, the agent reads it, corrects its query and tries again.

Each of these shows as a card in the answer. Looked at shows which tables the agent found, or the columns of one table. Queried names the Data Source and the question the query answers; select it to see the SQL, how many rows came back and the first of them, or the database’s error.

A result worth seeing then shows as a Card, with a chart and a table of its rows: see Cards.

A result keeps at most 100,000 rows. The agent sees only the first 50, so it adds up, filters and sorts in the query rather than reading rows one by one.

A query runs at most 60 seconds; then the database cancels it, and the agent tries a cheaper one. Select Stop while a query runs, and the database cancels it right away, not only the answer.

Your Conversation keeps the results of its queries for 7 days. When you delete the Conversation, they go with it for good.

The agent can do whatever the database user behind the Data Source may do, including changing data, and it doesn’t ask first. Whoever connected the Data Source decides that by the database user they chose. The agent changes data only when you ask it to.

Data that is in no Data Source yet, such as a CSV export on your Personal Drive, a file the Sandbox wrote or an Integration’s result, can go into a Managed Database. Then the agent queries it like any other Data Source, and it can go on a Dashboard.

Select the Managed Database for the Conversation and ask, such as “Load my revenue export into a table and add it up by region”. The agent:

  1. saves the data to your Personal Drive as a CSV or Parquet file first, if it came from the Sandbox or an Integration;
  2. loads the file into a table of the Managed Database. It reads the columns and their types from the file, and creates the table, replaces it, or adds the rows to it;
  3. queries the table, as for any Data Source.

It doesn’t ask first: loading writes only into the Managed Database. It loads only CSV and Parquet files of up to 50 MB, only from your own Personal Drive, and only into a Managed Database, never into a database your company connected. A loaded file shows in the Query Log of the Managed Database’s Data Source.

Every query goes into the Data Source’s Query Log: your question, the query, whether it worked and how long it took, never the rows it returned. The people who may accept Context Change Requests on the Data Source can read it.