Databases, tables and datasets in a chat

Before a Mind answers from data, it looks up what it can reach: the databases and tables of your data layer, and any spreadsheet you uploaded to the chat. This page explains those look-up steps, what they show, and what limits apply.

What a chat can look up

• What the chat reaches follows the Mind and your own access. An Analytical Mind looks up the tables linked to it on its Sources tab; a Warehouse Builder or Lake Builder chat looks up the whole layers. See Link data to a Mind and Warehouse Builder and Lake Builder.
• The first look-up, List Databases, lists each database with its name, its table count and a one-line comment. You normally see one for the Lake layer, with the comment "all mirrored tables from sources", and one for the Warehouse layer, with "all transformed tables".
• If you uploaded spreadsheets to this chat, they join the list as their own temporary database. Its comment reads "Datasets uploaded in this conversation (Excel). Query them exactly as listed."
• The Mind runs these steps on its own while answering, and each appears as a card you can open; the cards are described in Tool cards and Tool History. If the Mind needs a decision from you while it works, it pauses with a question card; see Questions, sleeping and benchmarks in a chat.

Find a table

• List Tables answers for one database: every table the chat can reach there, each with its name and its comment. Names read like "database"."table".
• You do not need a full name. Ask with a part of it: matching ignores case and looks anywhere in the name.
• If a part matches nothing, the step says how many tables it checked and suggests a shorter pattern. Try another word, or ask for the whole database first.
• When a look-up comes back empty, the card reads No databases found or No tables found; open it in full to see the step's own answer.

Read a table's schema

• Describe Table answers for one table: one row per column with the column name, data type, comment, whether it is mandatory, its default value, whether it is a primary key, its order number, whether it is a partition column, and its PII domain.
• A column that holds personal data reports String as its data type, matching the values a query over it returns; its PII Domain entry marks it as personal data.
• In a Builder chat the schema view can add a statistics line for the table: rows, parts, partitions, storage and dependants.
• When the chat cannot find the table, the step answers Table "<name>" either does not exist or is not in Mind's allowed tables list. Check the name and the database first; if the table should be reachable, deploy it and link it to the Mind (Link data to a Mind).

Datasets from your files

• An uploaded spreadsheet becomes tables the chat can query: one per sheet of a workbook, and one per CSV or TSV file. The look-up steps gather them under a temporary database of their own in this chat.
• The schema view of such a table names the file it came from and its row count, and says the data was uploaded in this conversation.
• The comment on the database tells the chat to query the tables exactly as listed, so the names in the list are the ones to use.
• These tables exist only in the chat where you attached the file: other chats cannot look them up or query them. Uploading is described in Attach files and Excel data to a chat.

What the chat does with the rows

• The look-up steps read structure: names, comments, column definitions and table counts. The chat reads row values by running queries; see Query tools and saved queries.
• When the chat writes a report or a slide deck from your queries, it copies figures from the exact results those queries returned in that conversation, not from memory.
• One limit applies when the chat writes: a result with more than 150 rows is not carried into the report or deck. The rows stay in the chat, and the chat reads them from the saved query when it needs them.
• A chart or table built from an incomplete breakdown keeps to what the data holds: it is not filled with invented members, and at most one remainder appears, labelled Other (not broken down in source data).

Questions, answered

What databases can a chat see?

The chat lists a database for the Lake layer, with the tables mirrored from your sources, and one for the Warehouse layer, with the transformed tables; each shows how many tables the chat can reach there. If you uploaded spreadsheets to this chat, a temporary database joins the list. What appears follows the Mind's sources and your own access (Link data to a Mind).

Which tables does a Builder chat see?

Every Lake table in a Lake Builder chat, and every Lake and Warehouse table in a Warehouse Builder chat; no sources need to be linked. See Warehouse Builder and Lake Builder.

Why does the chat not find my table?

The name or the database may be off, or the table is not one the chat can reach. The step answers Table "<name>" either does not exist or is not in Mind's allowed tables list. Check the spelling and the database first. If the table should be reachable, deploy it and link it to the Mind on its Sources tab (Link data to a Mind).

How do I find a table when I only remember part of its name?

Ask with the part you remember: matching ignores case and looks anywhere in the table name. If nothing matches, the step says how many tables it checked and suggests a shorter pattern.

Does a look-up read the rows?

No. The look-up steps read structure: names, comments, column definitions and table counts. Row values reach the chat through queries (Query tools and saved queries).

Why does a personal-data column show "String" as its data type?

A chat reads personal-data columns through a tokenised view, so the schema reports those columns as String; this matches the values a query over the column returns. The PII Domain entry marks the column as personal data.

Why does my spreadsheet appear as a temporary database?

An uploaded spreadsheet becomes tables the chat can query, and the look-up steps group them under a temporary database of their own so you can see what the chat has. The tables stay in that chat only (Attach files and Excel data to a chat).

Why are rows missing from a report or a slide deck?

The result has more than 150 rows, so its rows are not carried into the generated material. The figures still come from the result: the chat reads the rows from the saved query when it needs them.

Why does a generated chart not add up to 100%?

Part of the breakdown is missing from the source data, so the chart shows the members it has and is not filled with invented numbers. At most one remainder is added, labelled Other (not broken down in source data).