Databases — let an agent look things up in your database
Connect a read-only database, read its schema, and pre-approve the questions an agent may run. The agent never writes SQL.
Settings → Databases. Needs a plan with Data sources (Pro and up) and
dataSource.manage to configure.
An agent can look things up in your own database… It never writes SQL: it runs questions you have approved.
The principle: the model writes no SQL at run time. It only picks an approved question by name and fills in its parameters. You control exactly what can run.
Connect a database
Connect a database → Name and DSN (a PostgreSQL or MySQL connection string). The DSN is stored encrypted.
Use a read-only account. Every query runs in a read-only transaction, but database-side permissions are the layer you should have anyway.
Read schema pulls the table and column names for the query builder — names only, no data.
Questions the agent may ask
Each data source has a list of questions with three labels: approved, not approved, written for you (generated, not yet approved). Only approved questions can run.
Add a question opens the builder:
Describe it in plain language (monthly revenue for customer X). Ragenta reads the table and column names — never your data — and writes the SQL.
Set Name the agent will use and What it answers (the description the agent picks by).
Edit the SQL if needed. Parameters use $1 (Postgres) or ? (MySQL).
Try it with real values — a dry run with real parameters, shown as a table.
Approve and save.
In an agent
Enable Look something up in your database in Configuration. At run time the agent chooses a question by name and fills its typed parameters; the rows come back as a table it answers from.