Visual PostgreSQL studio

Draw the schema.Ship the migration.

DataForge turns tables and relationships drawn on a canvas into reviewed, transactional SQL, then lets you query, explain and edit the result against a real PostgreSQL database.

Create a free account

The demo gives you a private copy of an e-commerce database. Break anything; it is yours alone.

Database: PostgreSQL 17, real and private
Demo data: 6 tables · 17,000+ rows
Migrations: One transaction per apply
Queries: Read-only unless you say so

Workflow

From canvas to commit, every step shows its SQL.

The draft is diffed against the live catalog by object identity, so a renamed column is a rename, not a drop and an add.

01 Design
Add tables, set types, keys, defaults and identity columns. Drag a column onto another table's key to create a foreign key. Invalid references are flagged before any SQL exists.
- reviews.product_id → products.id
- ON DELETE CASCADE
- rating smallint NOT NULL DEFAULT 5

02 Review the generated SQL

-- Create table public.reviews with 4 columns
CREATE TABLE public.reviews (
    id bigint GENERATED ALWAYS AS IDENTITY NOT NULL,
    product_id bigint NOT NULL,
    rating smallint NOT NULL DEFAULT 5,
    body text,
    CONSTRAINT reviews_pkey PRIMARY KEY (id)
);

-- Link public.reviews to public.products
ALTER TABLE public.reviews ADD CONSTRAINT reviews_product_id_fkey
    FOREIGN KEY (product_id) REFERENCES public.products (id) ON DELETE CASCADE;

03 Apply, then query
1. BEGIN
2. pg_current_xact_id() → 1042
3. step 1 of 2 · CREATE TABLE
4. step 2 of 2 · ADD CONSTRAINT
5. COMMIT
6. status → applied
A migration is marked applied only after COMMIT returns. If the app stops mid-apply, the recorded transaction id settles the outcome with pg_xact_status().

Under the hood

The database enforces the rules. The interface only explains them.

Read-only by role: Reads run as a role that only has SELECT, inside READ ONLY transactions. Nobody guesses from the SQL text.
One statement per message: User SQL goes through the extended query protocol, which refuses multi-statement strings.
A watchdog, not a hope: statement_timeout first; then the server cancels and terminates the backend itself.
Clean connections: DISCARD ALL before a connection returns to the pool. No SET or temp table leaks into the next run.
Private sandboxes: Each project gets its own database and roles, with CONNECT revoked from everyone else.
Sealed credentials: AES-256-GCM with the connection id as associated data. Passwords never reach the browser.
Exact diffs: Tables and columns are identified by OID and attnum, so renames are detected, not guessed.
Streaming I/O: CSV imports and exports stream through cursors and batched unnest() inserts with savepoints.

App: Next.js 16 · React 19 · TypeScript
Database: PostgreSQL 17 · node-postgres · pg-cursor
Interface: React Flow · Monaco · TanStack Table
Runtime: Docker Compose · isolated networks

Try it

A real database is one click away.

A private copy of an e-commerce database, created in under a second
Six saved example queries, from window functions to JSONB
Write mode, migrations and imports all work against your copy
Removed automatically after a day