Ferro: The Pydantic x Rust ORM

Ferro is an async ORM for Python: you declare Pydantic models, and a Rust engine builds the SQL, runs it and turns rows into model instances. Its runtime dependencies are Pydantic and pydantic-settings.

Here is a model, a query and the result:

import asyncio

from ferro import Field, Model, connect, engines


class User(Model):
    id: int | None = Field(default=None, primary_key=True)
    username: str
    is_active: bool = True


async def main():
    await connect("sqlite::memory:", auto_migrate=True)

    async with engines.session():
        await User.create(username="alice")
        await User.create(username="bob", is_active=False)

        active_users = await User.where(lambda user: user.is_active == True).all()
        print(f"Found {len(active_users)} active users.")


asyncio.run(main())

Running it prints:

Found 1 active users.

Every ORM call runs inside async with engines.session():, which routes it to the connection you opened. auto_migrate=True creates the table on connect; it suits development and tests.

What Ferro does

• The Rust engine generates the SQL and parses rows outside the GIL, then fills each model's fields directly, without calling __init__.
• Everything is async, on pyo3-async-runtimes and SQLx.
• Models are Pydantic v2 models, so validation and IDE support work as they do for any Pydantic class.
• Within a session, one database row is one Python object (the identity map).
• It supports SQLite and PostgreSQL. Both are compiled into the engine, so no driver packages are needed.

Installation

Ferro ships pre-compiled wheels for macOS, Linux and Windows:

pip install ferro-orm
# With the migrations CLI (ferro migrate)
pip install "ferro-orm[cli]"
# With the Alembic bridge
pip install "ferro-orm[alembic]"

Schema changes

Auto-migrate keeps a development database current. For production, ferro migrate new writes reviewed migration files and ferro migrate up applies them; see Schema management. To use more than one database, register named connections with connect(..., name="..."); see Connections.