Databases
Every table in your memory in one place. Data files you import stay tables, row for row, with search, filters, sorting, charts and CSV export. Memory also builds its own tables of the people, organizations and products it knows about, with every value traced to its sources.
Databases holds two kinds of table, in one list and one grid:
- Imported tables: a data file you import stays a table, with every row and column as it was. Ask Nova about it in a chat and it answers from the table itself, with exact numbers.
- Databases built from memory: if a dozen conversations mention the same company, you get one row for it, with its size, its people and its last known status, and a link from each value to the memories that support it.
Open Memory, stay on the RAG tab, and switch the explorer to Databases.
The list
- The Folders / Databases switch, New database, Import table, and a Find a database… box.
- Imported tables come first, each with its row count, then the databases built from memory, each with its entity count. A New tag marks a database that was found recently.
- With nothing yet it reads No databases yet. A search that finds nothing reads No database matches.
- The first database opens by itself.
Import a table
Press Import table and pick a file: .csv, .tsv, .xlsx, .xls, .ods, .parquet, .jsonl or .ndjson. When it is in, the list shows the file's name and how many rows it brought. A file that is not a table is refused with Not a table and its name. You do not need a chat open.
Filters
Filters shows how many filters are on, with Clear to drop them all. Each column is a section you open with its chevron, marked with an icon for its type. A section with a filter on opens by itself and shows a badge with the count, and its × clears that filter.
- A column of values lists them with a tick, a count and a bar showing each one's share. The values you picked sit on top as chips. With more than eight values, Find a value searches them.
- A number or date column has From and To. For numbers, the boxes show the column's lowest and highest value.
- Other text has Contains….
- A database built from memory with several workspaces also has Namespace.
Above the rows, each filter shows as a pill you can remove, such as Region is West, Sales 100 to 500 or Order Date 2017-01-01 to 2017-01-31, with Clear all.
The table
- The header says Imported table or Database, then the name and description. An imported table also says what One row is, and counts the rows that match out of all of them (for example 3,203 of 9,994 rows), its columns, and the name Nova uses for it in a chat (In chat: …). A database built from memory counts its entities, populated and available fields, and how much of it is filled in.
- Sort (imported tables) opens a menu: Original order, then each column with Ascending and Descending. The button then names the sort, such as Profit ↓. You can also click a column header: once for ascending, again for descending, a third time for the original order.
- Columns picks the fields to show. Delete removes an imported table and its data, after you confirm. Refresh rebuilds the databases.
- The search box looks through every column of the table.
- The footer shows 25 rows a page, for example Showing 25 of 9994 and Page 1 of 400, with arrows to move between pages. With nothing matching it reads No rows match the current search and filters.
Select and export rows
The checklist button (Select rows) adds a column of ticks; it turns into × while it is on. The tick in the header selects the whole page. A bar shows how many rows are selected, with Export CSV to download them and Clear. Your selection is kept as you move between pages, and clears when you change the search, a filter or the sort.
Analytics
An imported table has two views, Table and Analytics. Analytics charts the rows that match your search and filters, so filter first and the charts follow.
- Rows matching: how many rows are charted, out of all of them, and the share.
- Columns charted: how many columns have a chart. Columns with a single value and ID-like columns are skipped.
- One card per column, with its name, how many different values it has and how many rows:
- a category or yes/no column: a donut with the total in the middle and a legend of each value's count and share;
- a number: a bar per value when it has 24 values or fewer, otherwise a histogram of 12 ranges, each with a table of the counts;
- a date: a timeline by month, or by year when it spans more than 60 months;
- free text: the 12 most common values, then Other, with bars showing their share.
While it works it reads Summarizing the rows….
Databases built from memory
A database built from memory has three tabs, Table, Sources and Relations. Its table starts with Entity and ends with Evidence, the number of sources behind each row.
- Pick it in the list.
- Narrow it with Filters or the search box.
- Click an entity to open its pane: Known properties, Statements and Evidence.
- Click a value to see its Current value, its status, a Conflict flag when one is raised, and its History.
- Open the Sources tab for the source excerpts behind the selected item, or Relations for its links to other things (subject, relation, object).
How they fill up
- They fill automatically as memories are processed. There is no button to start or force it.
- The background processing that builds them runs on the Default workspace. A new or separate workspace may show "No databases yet".
- Fields are typed: text, number, date, choice, link or yes/no.
Correct what memory got wrong
- Correct value saves a new current value. The old one stays in History, so nothing is lost.
- Rename an entity when memory picked the wrong name.
- Retire an entity to hide it from database views. Its sources stay in memory.
Create one
- Press New database in the explorer.
- Give it a Database name, and answer What kind of things belong here? Both are required.
- Press Create.
The definition you write tells the system which things belong in the table as memories are processed.