Workspace

Databases

PostgreSQL, SQLite, and MongoDB in the same project. Each connection picks an engine; the workspace changes with it.

A Based project can hold any mix of the three engines. The engine is a field on the connection file (engine = "postgres", "sqlite", or "mongodb"), not a separate app mode. Postgres is the deepest daily driver; SQLite and MongoDB share the same catalog, tabs, and saved-query model.

At a glance

	PostgreSQL	SQLite	MongoDB
Query language	SQL	SQL (multi-statement scripts)	Aggregation pipeline (JSON)
Catalog	Schemas → tables, views, materialized views	Tables, views, triggers	Collections
Data viewer	Paginated `SELECT *`	Paginated `SELECT *`	`find()`, first 200 documents
Structure	Columns, indexes, constraints, stats	Columns, indexes, constraints, DDL	Collection stats / inferred schema
Explain	`EXPLAIN (FORMAT JSON)`	`EXPLAIN QUERY PLAN`	—
Autocomplete	Not yet	Tables, columns, keywords, functions	—
Read-only mode	Not in the connection file	`read_only = true` or a `readonly` tag	Not in the connection file
Inline row editing	Not shipped	Not shipped	Document edit UI not wired

PostgreSQL

Primary SQL engine via sqlx. A project connection needs host, database, and username. port defaults to 5432. Password should be { env = "VAR" }, not a literal in git.

ssl on a connection file is a boolean: true maps to SSL mode Require, false (the default) to Disable. The New Connection wizard exposes the full set: Prefer, Require, Disable, Verify CA, Verify Full.

Ad-hoc SQL editor and saved [sql] queries.
Data viewer with page size from Settings (default 500 rows).
Structure inspector for tables and views.
Explain pane on the query editor.
Test-before-connect in the wizard (latency + server version).

SQLite

File-backed. file is a path relative to the repo root (or absolute). Connection-time PRAGMAs live in an optional [pragma] table and are applied when the pool opens.

Key	Default	Notes
`journal_mode`	`wal`	Skipped when the connection is read-only (would mutate storage).
`synchronous`	`normal`	Also skipped in read-only mode.
`foreign_keys`	`true`	Still applied when read-only.

Set read_only = true (preferred) or a readonly tag to open with mode=ro: no writes, and no empty database created if the file is missing. The connection dashboard also shows a live PRAGMA snapshot.

SQLite is the only engine with SQL autocomplete today. Multi-statement scripts run as a batch; the last result is shown. Query timeout from Settings is enforced on SQLite runs.

MongoDB

Native driver. The connection file needs a url (literal URI or { env = "MONGO_URL" }) and an optional default database.

Catalog lists collections.
Pipeline builder: JSON array of stages, run against a collection.
Saved queries use [aggregate] with pipeline and usually collection.
Document viewer: flattened find(), capped at 200 documents.
Change-stream sampler opens on connect (needs a replica set; samples up to 64 events).

Introspection pipelines whose first stage is database-scoped (for example $listCollections) may omit collection; Based supplies a harmless default handle. Data pipelines that read documents must set collection.

What is not engine-specific

Saved queries, history, the command palette, export to CSV/XLSX, and the table grid chrome are shared. A query’s [target] decides which connection (and therefore which engine) it may run against — see Connections and Editor.