This is a tool that reads the structure of an existing database and generates the appropriate SQLAlchemy model code, using the declarative style if possible.

This tool was written as a replacement for sqlautocode, which was suffering from several issues (including, but not limited to, incompatibility with Python 3 and the latest SQLAlchemy version).

• Supports SQLAlchemy 2.x
• Produces declarative code that almost looks like it was hand written
• Produces PEP 8 compliant code
• Accurately determines relationships, including many-to-many, one-to-one
• Automatically detects joined table inheritance
• Excellent test coverage

To install, do:

pip install sqlacodegen

To include support for the PostgreSQL CITEXT extension type (which should be considered as tested only under a few environments) specify the citext extra:

pip install sqlacodegen[citext]

To include support for the PostgreSQL GEOMETRY, GEOGRAPHY, and RASTER types (which should be considered as tested only under a few environments) specify the geoalchemy2 extra:

To include support for the PostgreSQL PGVECTOR extension type, specify the pgvector extra:

pip install sqlacodegen[pgvector]

pip install sqlacodegen[geoalchemy2]

At the minimum, you have to give sqlacodegen a database URL. The URL is passed directly to SQLAlchemy's create_engine() method so please refer to SQLAlchemy's documentation for instructions on how to construct a proper URL.

Examples:

sqlacodegen postgresql:///some_local_db
sqlacodegen --generator tables mysql+pymysql://user:password@localhost/dbname
sqlacodegen --generator dataclasses sqlite:///database.db

To see the list of generic options:

sqlacodegen --help

The selection of a generator determines the

The following built-in generators are available:

• tables (only generates Table objects, for those who don't want to use the ORM)
• declarative (the default; generates classes inheriting from declarative_base()
• dataclasses (generates dataclass-based models; v1.4+ only)
• sqlmodels (generates model classes for SQLModel)

The following options can be turned on by passing them using --options (multiple values must be delimited by commas, e.g. --options noconstraints,nobidi):

• tables
  • noconstraints: ignore constraints (foreign key, unique etc.)
  • nocomments: ignore table/column comments
  • noindexes: ignore indexes
• declarative
  • all the options from tables
  • use_inflect: use the inflect library when naming classes and relationships (turning plural names into singular; see below for details)
  • nojoined: don't try to detect joined-class inheritance (see below for details)
  • nobidi: generate relationships in a unidirectional fashion, so only the many-to-one or first side of many-to-many relationships gets a relationship attribute, as on v2.X
• dataclasses
  • all the options from declarative
• sqlmodel
  • all the options from declarative

The code generators that generate classes try to generate model classes whenever possible. There are two circumstances in which a Table is generated instead:

• the table has no primary key constraint (which is required by SQLAlchemy for every model class)
• the table is an association table between two other tables (see below for the specifics)

By default, table names are converted to valid PEP 8 compliant class names by replacing all characters unsuitable for Python identifiers with _. Then, each valid parts (separated by underscores) are title cased and then joined together, eliminating the underscores. So, example_name becomes ExampleName.

If the use_inflect option is used, the table name (which is assumed to be in English) is converted to singular form using the "inflect" library. For example, sales_invoices becomes SalesInvoice. Since table names are not always in English, and the inflection process is far from perfect, inflection is disabled by default.

Relationships are detected based on existing foreign key constraints as follows:

• many-to-one: a foreign key constraint exists on the table
• one-to-one: same as many-to-one, but a unique constraint exists on the column(s) involved
• many-to-many: (not implemented on the sqlmodel generator) an association table is found to exist between two tables

A table is considered an association table if it satisfies all of the following conditions:

1. has exactly two foreign key constraints
2. all its columns are involved in said constraints

Relationships are typically named based on the table name of the opposite class. For example, if a class has a relationship to another class with the table named companies, the relationship would be named companies (unless the use_inflect option was enabled, in which case it would be named company in the case of a many-to-one or one-to-one relationship).

A special case for single column many-to-one and one-to-one relationships, however, is if the column is named like employer_id. Then the relationship is named employer due to that _id suffix.

For self referential relationships, the reverse side of the relationship will be named with the _reverse suffix appended to it.

If the built-in generators with all their options don't quite do what you want, you can customize the logic by subclassing one of the existing code generator classes. Override whichever methods you need, and then add an entry point in the sqlacodegen.generators namespace that points to your new class. Once the entry point is in place (you typically have to install the project with pip install), you can use --generator <yourentrypoint> to invoke your custom code generator.

For examples, you can look at sqlacodegen's own entry points in its pyproject.toml.

If you have problems or other questions, you should start a discussion on the sqlacodegen discussion forum. As an alternative, you could also try your luck on the sqlalchemy room on Gitter.