Datasources

A datasource is a connection to your database. Everything else — schema search, query generation, dashboards — is built on top of it.

Supported databases

Database	Notes
PostgreSQL	Full introspection, EXPLAIN-based query validation.
MongoDB	Collections sampled to infer field types; queries validated with a zero-row dry run.
MySQL	Introspected via information_schema, validated with EXPLAIN.
Google BigQuery	Validated with a dry-run query job — no data is scanned or billed for validation.
Amazon Redshift	Postgres wire-protocol compatible; introspected via pg_catalog.
Supabase	Treated as Postgres — connect with your Supabase connection string.
Neon	Serverless Postgres — same introspection and validation path as Postgres.
Databricks	Spark SQL identifiers are quoted automatically for you.
Redis	Browsed by key pattern rather than table/row — see the note below.

Note:Redis isn’t a relational store, so it’s browsed differently: instead of tables, you browse by key pattern (e.g. session:*), and pagination beyond the first page of matches isn’t supported the way it is for SQL/Mongo sources.

Connecting

From Datasources → Connect, choose a database type and provide connection details (host, port, database, credentials, or a full connection string depending on the type). Optionally route the connection through an SSH tunnel if your database isn’t publicly reachable.

Credentials are encrypted with AES-256-GCM before they’re written to storage, and only decrypted in memory for the duration of a query. They’re never logged or returned by the API once saved.

Tip:Use Test Connection after saving to confirm InsightBase can actually reach the database before you try to introspect or generate anything against it.

Schema introspection

Introspection reads your tables, columns, types, foreign keys, and row-count estimates, then infers a likely semantic role for each column — a financial amount, a timestamp, a categorical value, an identifier, descriptive text — from naming patterns. Table and column descriptions are embedded (via Voyage AI) so the query-generation pipeline can find relevant schema context by meaning, not just by name.

•Introspection results (the table list) are cached briefly — use Re-introspect after schema changes on your end.
•You can annotate any table or column with a plain-language description to steer future query generation, without touching your database.
•Sample values (a handful of distinct non-null values per column) are captured to help the AI understand categorical fields.

Browsing tables

The table browser lets you page through raw rows without writing SQL — useful for sanity-checking data before you write a manifest against it. Pagination is page/limit based; the total row count is cached briefly per table so repeated page clicks don’t re-run a full count on every click.

Read-only by default

InsightBase never modifies your data. Generated and hand-edited queries alike are restricted to read-only operations, and every query is validated (an EXPLAIN for SQL sources, a zero-row dry run for MongoDB) before it’s saved or executed for real.

Quickstart Dashboards & Manifests