Database backup & restore for local development.

Dumps all table data to JSON before an ORM migration reset, then restores it after. Works directly at the database level — ORM-agnostic, framework-agnostic.

ORM migration resets (prisma migrate reset, drizzle-kit push --force-reset, etc.) wipe all data — dropping tables, re-running migrations, and re-seeding. This destroys development data that may be hard to recreate.

db-restore saves your data before the reset and merges it back after, preserving both your development data and any new seed data from migrations.

# Create a profile (one-time)
pnpx github:skydiver/db-restore setup myproject

# Before migration reset: dump your data
pnpx github:skydiver/db-restore myproject dump

# Run your ORM reset
prisma migrate reset   # or drizzle-kit push --force-reset, etc.

# After migration reset: restore your data
pnpx github:skydiver/db-restore myproject restore

pnpx github:skydiver/db-restore setup <name>                Create a new database profile interactively
pnpx github:skydiver/db-restore <name> dump [--out <dir>]   Dump all tables to JSON
pnpx github:skydiver/db-restore <name> restore [--in <dir>] Restore tables from JSON dump
pnpx github:skydiver/db-restore profiles                    List all saved profiles
pnpx github:skydiver/db-restore remove <name>               Delete a profile

Interactive profile creation. Prompts for connection details, tests the connection, and saves the profile. Passwords are never stored — they're prompted before each dump/restore.

$ pnpx github:skydiver/db-restore setup myproject

? Provider: postgres
? Host: localhost
? Port: 5432
? Database: myproject_dev
? User: postgres
? Password: ********

Testing connection... Connected.
Profile "myproject" saved.

Connects to the database, discovers all tables (excluding ORM migration tables), and writes each table's data to a JSON file.

$ pnpx github:skydiver/db-restore myproject dump

┌──────────┬──────┐
│ Table    │ Rows │
├──────────┼──────┤
│ users    │   42 │
│ posts    │  128 │
│ comments │  301 │
├──────────┼──────┤
│ Total    │  471 │
└──────────┴──────┘

Dump saved to ~/.config/db-restore/dumps/myproject (3 files)

If a previous dump exists, you'll be asked to archive it (.tar.gz), discard it, or cancel.

Reads the dump files and writes data back using UPSERT — your development data is merged with any new seed data from migrations.

$ pnpx github:skydiver/db-restore myproject restore

┌──────────┬──────┬──────────┐
│ Table    │ Rows │ Strategy │
├──────────┼──────┼──────────┤
│ users    │   42 │ upsert   │
│ posts    │  128 │ upsert   │
│ comments │  301 │ upsert   │
├──────────┼──────┼──────────┤
│ Total    │  471 │          │
└──────────┴──────┴──────────┘

Restore complete (471 rows across 3 tables)

Lists all saved profiles in a table.

Deletes a saved profile.

1. Connect to the database
2. Discover tables (auto-excludes ORM migration tables)
3. For each table: read columns, primary keys, and all rows
4. Encode special types (bytes, bigint, decimal, datetime, json) as JSON-safe wrappers
5. Write one JSON file per table + a _metadata.json

1. Connect to the database
2. Disable foreign key checks
3. For each table in the dump:
   • Detect schema drift (added/removed columns) and warn
   • Decode type wrappers back to native values
   • Tables with primary keys: UPSERT (insert or update on conflict)
   • Tables without primary keys: TRUNCATE + INSERT
4. Reset auto-increment sequences (PostgreSQL)
5. Re-enable foreign key checks

UPSERT preserves new seed data from migrations while restoring your dump data:

• Rows in dump matching DB primary key — updated with dump values
• Rows in dump with no match — inserted (your dev data)
• Rows in DB with no match in dump — kept (new seed data)

If the database schema changed between dump and restore:

• Column in both dump and DB — included in restore
• Column in dump but not in DB — skipped with warning
• Column in DB but not in dump — uses DB default with warning
• Table in dump but not in DB — skipped with warning

Special database types are encoded as JSON wrappers to prevent data loss:

Primitive types (int, float, string, boolean, null) are stored as-is.

ORM migration tables are automatically excluded from dumps:

~/.config/db-restore/
  profiles/
    myproject.json
  dumps/
    myproject/
      _metadata.json     # provider, timestamp, table list, format version
      users.json         # one file per table
      posts.json
      comments.json

Each table file contains column metadata and all rows:

{
  "table": "users",
  "primaryKeys": ["id"],
  "columns": [
    { "name": "id", "type": "integer" },
    { "name": "email", "type": "character varying" },
    { "name": "created_at", "type": "timestamp without time zone" }
  ],
  "rows": [
    {
      "id": 1,
      "email": "dev@example.com",
      "created_at": {
        "__type": "datetime",
        "value": "2026-01-15T10:30:00.000Z"
      }
    }
  ]
}

Connection profiles are stored in ~/.config/db-restore/profiles/. Passwords are never persisted.

{
  "name": "myproject",
  "provider": "postgres",
  "host": "localhost",
  "port": 5432,
  "database": "myproject_dev",
  "user": "postgres"
}

SQLite profiles store the file path instead of connection details:

{
  "name": "myproject",
  "provider": "sqlite",
  "path": "./data/dev.db"
}

pnpm install
pnpm dev          # Watch mode with tsx
pnpm test         # Run tests with Vitest
pnpm typecheck    # Type check with tsc
pnpm lint         # Lint with Biome
pnpm build        # Bundle with tsup

MIT