Docs / Build an application

studioSupported

Studio is not an admin panel. It is a deliberately small, read-only browser for the tables declared by the active zmdb.config.ts, served only on IPv4 loopback.

Limits first#

It binds to 127.0.0.1 and has no authentication or --host flag.
It accepts GET only. There is no write mode behind another flag.
It lists only the config's schema set; it does not introspect the database.
It never accepts SQL, free-text filters, undeclared tables, or undeclared sort columns from the browser.
Columns tagged Sensitive are omitted. Every other value is raw and unmasked.
Pages use LIMIT/OFFSET, default to 25 rows, and are capped at 50.
The UI is server-rendered HTML with no browser framework, script bundle, or asset build.

Do not point Studio at a production database expecting an authorization layer, general-purpose redaction, audit workflow, or editing controls. The loopback socket is the access boundary.

Start the installed command#

The selected config must declare the same driver thunk used by migration and push commands. This transcript was captured from the built executable against a generated SQLite project:

$ yarn zmdb studio --port 4545
http://127.0.0.1:4545

The emitted application served its declared-table index and rejected a write verb before database access:

$ curl -sS -o /dev/null -w '%{http_code}\n' http://127.0.0.1:4545/
200

$ curl -sS -o /dev/null -w '%{http_code}\n' \
    -X POST http://127.0.0.1:4545/
405

The measured index contained:

Local raw-data viewer
Declared tables
posts
2 visible columns

Omit --port to choose an ephemeral port. A fixed port does not widen the bind. If the requested loopback socket cannot be opened, the command fails; it never retries on another interface.

What the browser shows#

The index comes exactly from the config's schema files. From there you can:

page a table and sort by one visible declared column;
open a row through its declared primary key;
follow a declared relation to another configured table.

Logical property and table names stay in the browser. Physical names remain inside compiler-generated SQL. Unknown query parameters and malformed row keys are refused before a query runs.

Every page includes the raw-data warning. Sensitive columns are structurally excluded through the emitted JSON Schema property list, including from row keys and pagination links; Studio does not apply masking rules to anything else.

What it is not#

Studio does not discover tables that exist only in the database, edit data, accept arbitrary SQL, or replace psql, DataGrip, TablePlus, Beekeeper Studio, sqlite3, or another wire-protocol client. Those remain the right tools for database-wide introspection and administration.

For remote access, put the loopback viewer behind an authenticated tunnel such as ssh -L. Studio itself will not expose an unauthenticated database browser to the network.

---

See also: Config File · pull · CLI Overview

← Previous

export

CLI & Scaffolding