# The dish

## What it makes.

cli.view.fast

# Database migrations that never drift.

Drift diffs your schema against your live database, then writes the migration for you — type-safe, reviewable SQL you can read before it runs. No more hand-edited up/down files, no more “works on my machine” drift between environments.

```bash
$ npm i -D drift-cli
```

- Works with
  - PostgreSQL
  - MySQL
  - SQLite
  - PlanetScale
  - Neon
  - Turso

## Up and running in four commands

Point Drift at your database and let it write the first migration:

```bash
# install
npm i -D drift-cli

# scaffold drift.config.ts + connect your database
npx drift init

# diff your schema against the database and generate SQL
npx drift generate add_orders_table

# preview, then apply — every migration is plain reviewable SQL
npx drift status
npx drift push
```

You write the schema you want…

```typescript
// schema.ts — describe the shape, not the diff
import { table, text, uuid, timestamp, int } from 'drift-cli';

export const orders = table('orders', {
  id:        uuid('id').primaryKey().default('gen_random_uuid()'),
  email:     text('email').notNull(),
  total:     int('total_cents').notNull(),
  status:    text('status').notNull().default('pending'),
  createdAt: timestamp('created_at').notNull().default('now()'),
});
```

…Drift writes the migration.

```sql
-- migrations/0007_add_orders_table.sql  (generated by drift)
-- ↑ 1 table added · 0 columns dropped · safe to run

CREATE TABLE "orders" (
  "id"         uuid PRIMARY KEY DEFAULT gen_random_uuid(),
  "email"      text NOT NULL,
  "total_cents" integer NOT NULL,
  "status"     text NOT NULL DEFAULT 'pending',
  "created_at" timestamptz NOT NULL DEFAULT now()
);

CREATE INDEX "orders_email_idx" ON "orders" ("email");
```

## Why Drift

### A migration tool that gets out of your way

Drift owns one job — keeping your schema and your database in sync — and does it fast, safely, and in plain SQL you can read.

- ### Fast

A Rust diff engine reads your schema and introspects the database in one pass. Generating a migration on a 200-table schema takes under 40ms — fast enough to run on every save.

- ### Type-safe

Your schema is TypeScript, so column names, types and relations are checked at compile time. Rename a column and every query that referenced it lights up red in your editor.

- ### Zero-config

Run drift init and it detects your database from DATABASE_URL, picks the right dialect, and writes a one-file config. No adapters to wire up, no plugins to install.

- ### Framework-agnostic

Drift only owns your migrations — keep using Prisma, Drizzle, Kysely or raw SQL for queries. Postgres, MySQL, SQLite, and serverless drivers all speak the same commands.

- ### Reviewable SQL

Every migration is a plain .sql file you commit to git. Read it, edit it, comment on it in a PR. Nothing is hidden behind a binary format or an opaque migration table.

- ### Safe rollbacks

Each migration ships with a generated down step and a destructive-change guard. drift push --dry-run prints the exact plan; drift rollback reverts the last batch cleanly.

## How it works

### Three commands, every time

The whole workflow is connect, generate, apply. It looks the same on day one and on your thousandth migration — and identical on every machine that runs it.

1.  01
    `$ drift init`

**Connect**

Drift reads DATABASE_URL, detects your dialect, and writes drift.config.ts plus an empty migrations folder. One command, no questions.

2.  02
    `$ drift generate`

**Diff & generate**

Change your schema, then run generate. Drift compares the desired shape to the live database and emits a numbered, reviewable SQL migration with up and down steps.

3.  03
    `$ drift push`

**Apply anywhere**

Run push in CI, on a teammate’s laptop, or against prod. Drift applies only the migrations that haven’t run yet and records them — every environment ends up identical.

## Benchmarks

### Built for the inner loop

Time to generate and apply a migration on a 50-table Postgres schema. Lower is better — measured on an M2 Air, median of 20 runs.

| Tool                          | Time    |
|-------------------------------|---------|
| Drift                         | 38 ms   |
| Hand-written + manual diff    | ~6 min  |
| ORM migrate (Node)           | 210 ms  |
| ORM migrate (Python)         | 640 ms  |

> We deleted 1,400 lines of hand-rolled migration scripts the week we adopted Drift. Generated SQL we can actually read in review, and prod and staging finally stopped drifting apart.

**Priya Raman** Staff Engineer, Northwind

## Your ingredients

### Your agent will ask.

Answer these once. The recipe uses your details instead of placeholder copy.

1. What's the project name and the one-line description of what it does?
2. What's the install command and which ecosystem/language is it for (npm, pip, brew, cargo)?
3. What are the 3-4 key selling points (speed, type-safety, zero-config)?
4. What's the GitHub repo URL and do you have a Discord/community link?

## The method

### The whole prompt.

Paste this into Claude, ChatGPT, Codex, Cursor, or another coding agent. It builds the site and publishes it to Spacefast.