Databases

Apps → Databases lists the databases the assistant and your workflows can query. There are two kinds:

• Internal databases — table databases the data extractor built here from your documents. A workflow with a table-building node (data extraction, row analysis) creates one; every run adds or updates rows.
• External databases — databases you run elsewhere, connected read-only: PostgreSQL, MySQL, MariaDB or SQL Server. They are set up in My Data → Databases and appear here with their engine and address.

Internal databases

Open an internal database to see its tables. A table looks like a spreadsheet and is edited like one:

• Double click a cell to edit it (Cmd+Enter to save · Esc to cancel).
• Column settings — click a column header to rename it, change its Output Type, add an Instruction the extractor follows for that column, drag to reorder, duplicate or remove it. Removing a column removes its data from every row.
• Delete selected rows; Delete table removes the whole table.
• Export as .xlsx downloads the table; Save a copy duplicates it (you can include the latest output rows in the copy).
• Table Chat — ask questions about the table in plain language ("total by supplier", "which rows are missing a date?").

Tables built by extraction keep a link to their sources. Select a row (or a cell) and open Extraction sources to see where each value came from: the source PDF opens with the extracted fields highlighted on the page, each marked Located, Approximate or Not located. Sources are recorded at extraction time; re-run the extraction to record them for rows that predate the feature. Highlighting works for PDF files.

Row analysis workflows write to a separate result table rather than changing the source table.

Tables are shared with workflows

A table used as the Table Source of a workflow node is overwritten by that node's runs. Editing a cell by hand is fine, but a later run can replace it — put permanent rules in the column Instruction instead.

External databases

Open a connection to see its Connection details and the Tables the connection is allowed to see; pick a table to preview its first rows. Connections are read-only — only SELECT queries are ever run — and the owner can restrict them to an allowed tables list and a max rows per query. Deleting a connection stops the charts built on it from refreshing.

Using databases elsewhere

• In Chat AI, attach a database or a connection table from the paperclip picker and ask questions — the assistant writes and runs the queries.
• In Workflow AI, the Database nodes read from and write to tables; Data Sources on a table node choose where its rows come from.
• In Studio, pick up to 3 connections as the data behind a live page.

Do it with the API

Internal tables: Tables (GET /microservices/data-extractor/tables/, row updates by source_uid, highlights, table-chat/). External connections: My Data → data sources (/data-sources/, test/, tables/, preview/).