prisma-sqlmodel-gen

Generate Python SQLModel table models from a Prisma schema.

Who needs this?

Use this package if your TypeScript app uses Prisma, but another Python service needs matching SQLModel models for the same database. It is for teams that want Prisma to stay as the single source of truth instead of hand-maintaining the same schema twice.

What it does

prisma-sqlmodel-gen reads your Prisma schema file and generates Python SQLModel classes from it. It works from the schema file itself, so it does not need to inspect a live database.

Quick start

Install:

npm install -D prisma-sqlmodel-gen

npm package:

• https://www.npmjs.com/package/prisma-sqlmodel-gen

Add a generator block to schema.prisma:

generator sqlmodel {
  provider            = "prisma-sqlmodel-gen"
  output              = "../python_app/generated"
  moduleName          = "models.py"
  emitInit            = "true"
  strict              = "true"
  headerComment       = "true"
  sqlmodelImportStyle = "sqlmodel"
}

Run generation:

npx prisma generate

Or verify generated files in CI:

prisma-sqlmodel-gen --schema ./prisma/schema.prisma --output ./python_app/generated --check

Recommended workflow

{
  "scripts": {
    "db:migrate": "prisma migrate dev && prisma generate",
    "db:deploy": "prisma migrate deploy && prisma generate"
  }
}

--check verifies that committed generated files match the current Prisma schema and exits nonzero on drift.

Minimal example

Prisma schema:

datasource db {
  provider = "postgresql"
}

generator sqlmodel {
  provider = "prisma-sqlmodel-gen"
  output   = "./generated/sqlmodel"
}

model User {
  id    Int    @id @default(autoincrement())
  email String @unique @db.VarChar(255)

  @@map("users")
}

Representative generated output:

class User(SQLModel, table=True):
    __tablename__ = "users"
    id: int | None = Field(
        sa_column=Column(Integer(), primary_key=True, nullable=False, autoincrement=True),
        default=None,
    )
    email: str = Field(sa_column=Column(VARCHAR(255), unique=True, nullable=False))

Support matrix

Runtime targets:

• Prisma 7
• Node 22+
• Python 3.11+
• SQLModel 0.0.38

Datasource targets:

• PostgreSQL
• MySQL

Validation matrix:

• Python 3.11 + PostgreSQL + SQLModel 0.0.38
• Python 3.12 + PostgreSQL + SQLModel 0.0.38
• Python 3.12 + MySQL + SQLModel 0.0.38

Generated artifacts:

• models.py
• __init__.py

Mode:

• strict generation with hard failures for unsupported constructs
• deterministic output intended for commit-and-review workflows

Supported features

Fully supported relational-schema features:

• mapped tables and columns via @@map and @map
• PostgreSQL @@schema
• enums
• PostgreSQL scalar lists and enum lists
• explicit one-to-one and one-to-many relations
• explicit many-to-many via join model
• implicit many-to-many via synthesized PrismaImplicitLink_* models
• relation map, onDelete, and onUpdate
• named/composite foreign keys
• named indexes and unique constraints
• MySQL index length
• index sort where SQLAlchemy can represent it directly
• @updatedAt
• defaults such as autoincrement(), uuid(), now(), and dbgenerated(...) when they map cleanly

Feature matrix:

| Feature | Status | | --- | --- | | @map / @@map | Supported | | @@schema on PostgreSQL | Supported | | @id | Supported | | @@id composite keys | Supported | | @unique / @@unique | Supported | | @@index basic indexes | Supported | | PostgreSQL @@index(..., type: BTree|Hash|SpGist|Brin|Gin) | Supported | | Prisma enums | Supported | | PostgreSQL scalar lists | Supported | | PostgreSQL enum lists | Supported | | Prisma Json on PostgreSQL | Supported as JSONB | | Explicit one-to-one | Supported | | Explicit one-to-many | Supported | | Explicit many-to-many via join model | Supported | | Implicit many-to-many | Supported via synthesized link model | | Multiple named relations between same model pair | Supported | | Self-relations | Supported | | Relation map / onDelete / onUpdate | Supported | | Composite foreign keys | Supported | | @updatedAt | Supported | | PostgreSQL native types used by Prisma SQL schemas | Supported where mapped directly | | MySQL native types used by Prisma SQL schemas | Supported where mapped directly | | MySQL index length | Supported | | Index sort | Supported where mapped directly | | autoincrement() | Supported | | uuid() | Supported | | now() | Supported | | dbgenerated(...) | Supported where SQLAlchemy server defaults can represent it |

Supported with provider limitations:

| Feature | Status | | --- | --- | | Scalar lists | PostgreSQL only | | Enum lists | PostgreSQL only | | @@schema | PostgreSQL only | | Dialect native types | Limited to PostgreSQL/MySQL direct SQLAlchemy mappings |

Intentionally unsupported non-isomorphic features:

| Feature | Behavior | | --- | --- | | relationMode = "prisma" | Hard fail | | @ignore / @@ignore | Hard fail | | Unsupported(...) scalar fields | Hard fail | | cuid() / ulid() / nanoid() defaults | Hard fail | | PostgreSQL operator classes / expression-style indexes | Hard fail | | Advanced index algorithms other than supported PostgreSQL BTree|Hash|SpGist|Brin|Gin | Hard fail | | Providers outside PostgreSQL / MySQL | Hard fail | | Features that need Python-only metadata absent from Prisma | Out of scope |

Known limitations

The generator stays in strict mode by default. If a Prisma feature cannot be represented 1:1 in SQLModel/SQLAlchemy metadata or ORM behavior, generation fails instead of degrading silently.

Examples:

• scalar lists outside PostgreSQL
• enum lists outside PostgreSQL
• advanced PostgreSQL index forms such as operator classes or unsupported custom index algorithms
• provider-specific features outside PostgreSQL/MySQL scope
• Prisma-client-only behaviors such as relationMode = "prisma" or @ignore

Diagnostics

Failures include:

• machine-readable diagnostic codes
• model and field context
• source line and column when discovered from the schema text
• a short remediation suggestion when one is available

Example:

ERROR UNSUPPORTED_ADVANCED_INDEX (User.value) [line 7, col 3]:
Prisma index field modifier 'ops' does not map 1:1 to generated SQLModel metadata.
Suggestion: Use only supported sort/length modifiers or manage the advanced index manually in migrations.

Troubleshooting

• UNSUPPORTED_CLIENT_SIDE_DEFAULT Replace client-side defaults such as cuid() with uuid() or a database-generated default.
• UNSUPPORTED_ADVANCED_INDEX Keep unsupported operator classes or expression-style indexes in Prisma migrations instead of generated SQLModel metadata. Supported PostgreSQL type: BTree, Hash, SpGist, Brin, and Gin indexes are generated directly.
• PYTHON_FIELD_NAME_COLLISION Rename Prisma fields that collapse to the same Python identifier after sanitization.
• stale output in --check Run prisma generate again and commit the regenerated Python files.

Python runtime notes

If you use a different SQLModel version, treat it as unvalidated until you run this project's integration suite against that version.

The generated code is validated against SQLModel 0.0.38 on Python 3.11+ with the SQLAlchemy 2.x dependency range that SQLModel 0.0.38 resolves.

Your Python application still needs to install its own runtime dependencies, for example:

pip install "sqlmodel==0.0.38" psycopg[binary]

Choose the database driver appropriate for your environment.

Development

Useful commands:

npm ci
npm run check
npm test
npm run test:coverage
npm run test:integration
npm run test:all

Repository

• Homepage: https://github.com/ganesh-rao/prisma-sqlmodel-gen
• Issues: https://github.com/ganesh-rao/prisma-sqlmodel-gen/issues

Security

Run npm audit before release. At the time of the initial public release, there are unresolved upstream advisories in transitive dependencies used by @mrleebo/prisma-ast, and npm audit currently suggests semver-incompatible Prisma downgrades for some advisories. See SECURITY.md for the current project stance.

Project structure

• Generated Python modules are fully owned by the generator.
• Handwritten Python business logic should live in separate wrapper modules.
• Regenerate after every successful schema migration or schema change.

Internal source layout:

• src/bin.ts package executable that routes Prisma generator invocations to the generator entrypoint and manual commands to the standalone CLI
• src/generator.ts Prisma generator entrypoint
• src/cli.ts standalone CLI
• src/ir/* normalized schema IR exports
• src/compat/* compatibility and diagnostics exports
• src/emit/python/* Python emission exports
• src/map/* Prisma schema parsing and mapping exports

Notes

• This generator currently emits table models only.
• It does not emit Create, Update, or Read DTO classes.