runid-lys 0.8.2

A modular FastAPI framework for building GraphQL APIs with SQLAlchemy, featuring component-based architecture, JWT authentication, and multi-tenant permissions.

Lys

A modular Python framework for building GraphQL APIs with FastAPI and SQLAlchemy, inspired by Django's app architecture.

Lys provides a component-based system where each application module is self-contained and can be deployed as part of a monolith or extracted into its own microservice.

Features

• Modular app system with automatic component discovery and registration
• GraphQL API via Strawberry with Relay support (Global IDs, cursor pagination)
• Async SQLAlchemy with PostgreSQL, SQLite, and MySQL support
• JWT authentication with refresh tokens, XSRF protection, and rate limiting
• Permission system with role-based, organization-based, and row-level access control
• Multi-tenant architecture with organization-scoped data filtering
• Microservice-ready with service-to-service JWT authentication
• Celery integration for background tasks and periodic jobs
• Fixture system with dependency resolution and environment-based loading

Requirements

• Python >= 3.13
• PostgreSQL (production) or SQLite (development/testing)
• Redis (for Celery workers and rate limiting)

Installation

pip install lys

For optional features:

pip install lys[storage]   # S3 file storage (aioboto3)
pip install lys[mollie]    # Mollie payment integration
pip install lys[dev]       # Development and testing tools

Quick Start

1. Project Structure

A Lys project consists of a server project that configures and runs the framework, and one or more app packages that contain your business logic.

my-project/
├── settings.py              # App registration, database, plugins
├── main.py                  # CLI entry point (Typer)
├── src/
│   └── app.py               # FastAPI application factory
├── .env                     # Environment variables
└── pyproject.toml

2. Create an App

Apps follow a consistent structure. Each module contains up to five component types: entities, services, fixtures, nodes, and webservices.

my_apps/
└── catalog/
    ├── __init__.py
    └── modules/
        ├── __init__.py          # Declares __submodules__
        └── product/
            ├── __init__.py
            ├── entities.py      # SQLAlchemy models
            ├── services.py      # Business logic
            ├── fixtures.py      # Seed data
            ├── nodes.py         # GraphQL types
            └── webservices.py   # Queries and mutations

Declare submodules in modules/__init__.py:

from . import product

__submodules__ = [product]

3. Define an Entity

Lys provides two base entity classes:

• Entity — for business objects with auto-generated UUID primary keys
• ParametricEntity — for reference/configuration data with business-meaningful string IDs

from sqlalchemy import Uuid, String, Numeric, ForeignKey
from sqlalchemy.orm import Mapped, mapped_column, declared_attr, relationship
from lys.core.entities import Entity, ParametricEntity
from lys.core.registries import register_entity


@register_entity()
class ProductCategory(ParametricEntity):
    """Reference table for product categories (ELECTRONICS, CLOTHING, etc.)."""
    __tablename__ = "product_category"


@register_entity()
class Product(Entity):
    """A product in the catalog."""
    __tablename__ = "product"

    name: Mapped[str] = mapped_column(String(255), nullable=False)
    price: Mapped[float] = mapped_column(Numeric(10, 2), nullable=False)

    # Hard FK to parametric entity
    category_id: Mapped[str] = mapped_column(
        ForeignKey("product_category.id"), nullable=False
    )

    # Soft FK to another service's entity (no FK constraint)
    client_id: Mapped[str] = mapped_column(
        Uuid(as_uuid=False), nullable=False,
        comment="Client reference (soft FK)"
    )

    @declared_attr
    def category(self):
        return relationship("product_category", foreign_keys=[self.category_id])

    def accessing_organizations(self) -> dict[str, list[str]]:
        return {"client": [self.client_id]}

    @classmethod
    def organization_accessing_filters(cls, stmt, organization_id_dict):
        filters = [cls.client_id.in_(organization_id_dict.get("client", []))]
        return stmt, filters

Rule: All soft FK fields in Entity subclasses must use Uuid(as_uuid=False) for database-level UUID validation. ParametricEntity subclasses use plain string IDs.

4. Define a Service

Services contain business logic and CRUD operations. They are accessed exclusively through app_manager.

from typing import Optional
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession
from lys.core.services import EntityService
from lys.core.registries import register_service


@register_service()
class ProductCategoryService(EntityService["ProductCategory"]):
    """Service for product category reference data."""
    pass  # Inherits full CRUD from EntityService


@register_service()
class ProductService(EntityService["Product"]):
    """Service for product operations."""

    @classmethod
    async def search(
        cls, session: AsyncSession, query: str, category_id: Optional[str] = None
    ) -> list:
        stmt = select(cls.entity_class).where(
            cls.entity_class.name.ilike(f"%{query}%")
        )
        if category_id:
            stmt = stmt.where(cls.entity_class.category_id == category_id)
        result = await session.execute(stmt)
        return list(result.scalars().all())

    @classmethod
    async def update_price(
        cls, session: AsyncSession, product_id: str, new_price: float
    ):
        product = await cls.get_by_id(product_id, session)
        if product is None:
            raise ValueError("Product not found")
        product.price = new_price
        return product

Rule: Always access entities and services through app_manager, never via direct imports. Use cls.entity_class for the current service's entity and cls.app_manager.get_service("name") / cls.app_manager.get_entity("name") for others.

5. Define GraphQL Nodes

Nodes map entities to GraphQL types using Strawberry.

import strawberry
from strawberry import relay
from lys.core.graphql.nodes import EntityNode, parametric_node
from lys.core.registries import register_node


@register_node()
@parametric_node(ProductCategoryService)
class ProductCategoryNode:
    """Auto-generated GraphQL type for ProductCategory."""
    pass


@register_node()
class ProductNode(EntityNode["ProductService"], relay.Node):
    """GraphQL type for Product."""

    id: relay.NodeID[str]
    name: str
    price: float

    _entity: strawberry.Private["Product"]

    @strawberry.field
    def client_id(self) -> relay.GlobalID:
        return relay.GlobalID("ClientNode", self._entity.client_id)

    @strawberry.field
    def category(self) -> ProductCategoryNode:
        return self._entity.category

6. Define Webservices (Queries and Mutations)

Webservices expose GraphQL operations with built-in permission checking.

import strawberry
from strawberry import relay
from typing import Optional, Annotated
from sqlalchemy import select
from lys.core.graphql.types import Query, Mutation
from lys.core.graphql.registries import register_query, register_mutation
from lys.core.graphql.getter import lys_getter
from lys.core.graphql.connection import lys_connection
from lys.core.graphql.create import lys_creation
from lys.apps.user_role.consts import ROLE_ACCESS_LEVEL
from lys.apps.organization.consts import ORGANIZATION_ROLE_ACCESS_LEVEL


@strawberry.type
@register_query()
class ProductQuery(Query):

    @lys_getter(
        ProductNode,
        access_levels=[ROLE_ACCESS_LEVEL, ORGANIZATION_ROLE_ACCESS_LEVEL],
        description="Get a product by ID."
    )
    async def product(self, obj, info):
        pass  # Handled by lys_getter

    @lys_connection(
        ProductNode,
        access_levels=[ROLE_ACCESS_LEVEL, ORGANIZATION_ROLE_ACCESS_LEVEL],
        description="List products with optional search."
    )
    async def all_products(
        self,
        info,
        search: Annotated[Optional[str], strawberry.argument(description="Search by name")] = None,
    ):
        entity = info.context.app_manager.get_entity("product")
        stmt = select(entity).order_by(entity.created_at.desc())
        if search:
            stmt = stmt.where(entity.name.ilike(f"%{search.strip()}%"))
        return stmt


@strawberry.type
@register_mutation()
class ProductMutation(Mutation):

    @lys_creation(
        ensure_type=ProductNode,
        access_levels=[ROLE_ACCESS_LEVEL],
        description="Create a new product."
    )
    async def create_product(self, inputs, info):
        input_data = inputs.to_pydantic()
        service = info.context.app_manager.get_service("product")
        return await service.create(
            info.context.session,
            name=input_data.name,
            price=input_data.price,
            category_id=input_data.category_id,
            client_id=input_data.client_id,
        )

7. Define Fixtures

Fixtures load seed data into the database, with support for environment-based filtering.

from lys.core.fixtures import EntityFixtures
from lys.core.models.fixtures import ParametricEntityFixturesModel
from lys.core.registries import register_fixture


@register_fixture()
class ProductCategoryFixtures(EntityFixtures["ProductCategoryService"]):
    model = ParametricEntityFixturesModel
    delete_previous_data = False

    data_list = [
        {"id": "ELECTRONICS", "attributes": {"enabled": True, "description": "Electronic devices and accessories."}},
        {"id": "CLOTHING", "attributes": {"enabled": True, "description": "Apparel and fashion items."}},
        {"id": "BOOKS", "attributes": {"enabled": True, "description": "Physical and digital books."}},
    ]

8. Configure and Run

settings.py — register apps and configure services:

import os
from lys.core.configs import settings as app_settings
from lys.core.consts.environments import EnvironmentEnum


def configure_app():
    app_settings.configure(
        env=EnvironmentEnum(os.getenv("ENVIRONMENT", "dev")),
        secret_key=os.getenv("SECRET_KEY"),
        front_url=os.getenv("FRONT_URL", "http://localhost:3000"),
        apps=[
            "lys.apps.base",
            "lys.apps.user_auth",
            "lys.apps.user_role",
            "lys.apps.organization",
            "my_apps.catalog",
        ],
        middlewares=[
            "lys.core.middlewares.SecurityHeadersMiddleware",
            "lys.core.middlewares.RateLimitMiddleware",
        ],
        permissions=[
            "lys.apps.base.permissions.InternalServicePermission",
            "lys.apps.user_auth.permissions.AnonymousPermission",
            "lys.apps.user_auth.permissions.JWTPermission",
            "lys.apps.organization.permissions.OrganizationPermission",
        ],
    )

    app_settings.database.configure(
        type="postgresql",
        host=os.getenv("DB_HOST", "localhost"),
        port=int(os.getenv("DB_PORT", "5432")),
        username=os.getenv("DB_USER"),
        password=os.getenv("DB_PASSWORD"),
        database=os.getenv("DB_NAME"),
    )

src/app.py — create the FastAPI application:

from lys.core.managers.app import LysAppManager
from lys.core.consts.component_types import AppComponentTypeEnum
from settings import configure_app

configure_app()

app_manager = LysAppManager()
app_manager.configure_component_types([
    AppComponentTypeEnum.ENTITIES,
    AppComponentTypeEnum.SERVICES,
    AppComponentTypeEnum.FIXTURES,
    AppComponentTypeEnum.NODES,
    AppComponentTypeEnum.WEBSERVICES,
])

app = app_manager.initialize_app(
    title="My API",
    description="Built with Lys",
    version="1.0.0",
)

main.py — CLI entry point:

import typer
from pathlib import Path
from lys.core.clis import run_fast_app

cli = typer.Typer()

@cli.command()
def run(
    host: str = typer.Option("0.0.0.0", "--host"),
    port: int = typer.Option(8000, "--port"),
    reload: bool = typer.Option(False, "--reload"),
    workers: int = typer.Option(1, "--workers"),
):
    run_fast_app(host=host, port=port, reload=reload, workers=workers, app_path=Path("src/app.py"))

if __name__ == "__main__":
    cli()

Run the server:

python main.py run --reload

The GraphQL playground is available at http://localhost:8000/graphql.

Architecture

Component System

Lys organizes code into five component types, loaded in this order:

Order	Component	Role	Base Class
1	Entities	SQLAlchemy models	Entity, ParametricEntity
2	Services	Business logic, CRUD	EntityService[T]
3	Fixtures	Seed data	EntityFixtures[T]
4	Nodes	GraphQL type definitions	EntityNode[T]
5	Webservices	Queries, mutations, endpoints	Query, Mutation

Components are discovered automatically via the @register_entity(), @register_service(), @register_node(), and @register_fixture() decorators.

App Manager

LysAppManager is the central orchestrator. It loads apps, registers components, and creates the FastAPI application. Database schema migrations are handled by Alembic.

All entities and services are accessed through the app manager:

# Get a registered entity class
user_entity = app_manager.get_entity("user")

# Get a registered service class
user_service = app_manager.get_service("user")

# Inside a service, use cls.app_manager
class OrderService(EntityService["Order"]):
    @classmethod
    async def create_order(cls, session, user_id, items):
        product_service = cls.app_manager.get_service("product")
        # ...

Monolith to Microservices

Lys apps are designed to be independently deployable. A typical evolution:

Phase 1 — Monolith: All apps in a single process.

apps=[
    "lys.apps.base",
    "lys.apps.user_auth",
    "lys.apps.organization",
    "my_apps.catalog",
    "my_apps.analytics",
    "my_apps.billing",
]

Phase 2 — Microservices: Each app (or group of apps) in its own service. Apps communicate via GraphQL with service-to-service JWT authentication.

┌──────────────────┐    ┌───────────────────┐    ┌──────────────────┐
│   Auth Service   │    │ Catalog Service   │    │ Billing Service  │
│                  │    │                   │    │                  │
│ lys.apps.base    │    │ lys.apps.base     │    │ lys.apps.base    │
│ lys.apps.user_*  │    │ my_apps.catalog   │    │ my_apps.billing  │
│ lys.apps.org     │    │                   │    │                  │
└────────┬─────────┘    └────────┬──────────┘    └────────┬─────────┘
         │                       │                        │
         └───────────── GraphQL + JWT ────────────────────┘

Each service registers only the apps it needs. Lys handles service-to-service authentication via InternalServicePermission and ServiceAuthUtils.

Permission System

Lys implements a layered permission system:

Level	Description	Class
Anonymous	Public endpoints, no auth required	AnonymousPermission
JWT	Authenticated user with valid token	JWTPermission
Role-based	User has a specific role	ROLE_ACCESS_LEVEL
Organization	User belongs to the right organization	ORGANIZATION_ROLE_ACCESS_LEVEL
Owner	User owns the resource	OWNER_ACCESS_LEVEL
Internal	Service-to-service calls	InternalServicePermission

Permissions are evaluated in chain. Each webservice declares its required access levels:

@lys_getter(
    ProductNode,
    access_levels=[ROLE_ACCESS_LEVEL, ORGANIZATION_ROLE_ACCESS_LEVEL],
    is_licenced=True,
    description="Get a product."
)

Row-level filtering is handled by implementing accessing_users(), accessing_organizations(), and organization_accessing_filters() on entities.

Celery Workers

Lys integrates with Celery for background tasks. Workers load only entities and services (no nodes or webservices).

worker.py:

from settings import configure_app
from lys.core.configs import settings
from lys.core.celery_app import create_celery_app

configure_app()
celery_app = create_celery_app(settings)
app = celery_app

Task definition:

from celery import shared_task, current_app

@shared_task
def process_catalog_import(file_id: str):
    app_manager = current_app.app_manager
    product_service = app_manager.get_service("product")

    with app_manager.database.get_sync_session() as session:
        # Process import...
        session.commit()

    return {"status": "completed"}

Run the worker:

celery -A worker worker --loglevel=info

Configuration

Environment Modes

Mode	debug	testing	GraphQL Introspection	Log Level
DEV	True	True	Enabled	DEBUG
DEMO	False	False	Enabled	INFO
PREPROD	False	False	Disabled	WARNING
PROD	False	False	Disabled	ERROR

Database

app_settings.database.configure(
    type="postgresql",        # postgresql, sqlite, mysql
    host="localhost",
    port=5432,
    username="user",
    password="password",
    database="mydb",
    pool_size=10,             # Connection pool size
    max_overflow=20,          # Extra connections allowed
    pool_pre_ping=True,       # Verify connections before use
    pool_recycle=3600,        # Recycle connections after 1 hour
    ssl_mode="require",       # PostgreSQL SSL mode
)

Plugins

Lys supports plugin-based configuration for optional features:

app_settings.configure(
    plugins={
        "cors": {
            "allow_origins": ["http://localhost:3000"],
            "allow_methods": ["*"],
        },
        "auth": {
            "cookie_secure": True,
            "access_token_expire_minutes": 5,
            "refresh_token_expire_hours": 24,
        },
        "file_storage": {
            "provider": "s3",
            "bucket": "my-bucket",
        },
        "rate_limit": {
            "requests_per_minute": 60,
        },
    }
)

Testing

Running Tests

# Unit tests (mocked dependencies, fast)
pytest tests/unit/ -v

# Integration tests (real database, forked for isolation)
pytest tests/integration/ --forked -v

# E2E tests (full FastAPI app with httpx client, forked)
pytest tests/e2e/ --forked -v

# Combined coverage (ALWAYS use separate processes — required for singleton isolation)
pytest tests/unit/ --cov=src/lys --cov-report=
pytest tests/integration/ --forked --cov=src/lys --cov-append --cov-report=
pytest tests/e2e/ --forked --cov=src/lys --cov-append --cov-report=term-missing

Test Structure

tests/
├── unit/                    # Mocked dependencies
│   └── apps/
│       └── catalog/
│           ├── test_product_entities.py
│           ├── test_product_services.py
│           └── test_product_services_logic.py
├── integration/             # Real DB (SQLite in-memory), forked
│   └── apps/
│       └── catalog/
│           └── test_product_service.py
├── e2e/                     # Full app with httpx AsyncClient, forked
│   └── test_product_api.py
├── fixtures/                # Shared test helpers (create_all_tables, etc.)
│   └── database.py
└── conftest.py              # Shared fixtures

Integration and E2E tests run in forked subprocesses (pytest-forked) to isolate the SQLAlchemy registry singleton between test runs.

Built-in Apps

Lys ships with several pre-built apps:

App	Description
lys.apps.base	Languages, emailing, jobs, logging, access levels, webservice management
lys.apps.user_auth	User entity, login/logout, JWT tokens, password reset, email verification
lys.apps.user_role	Role management, role-based permissions
lys.apps.organization	Multi-tenant client/organization support, organization roles
lys.apps.licensing	Subscription plans, license rules, payment integration
lys.apps.file_management	File upload/download with S3 storage, file import processing
lys.apps.ai	AI conversation management, text improvement, tool generation from GraphQL

Documentation

Detailed documentation is available in the docs/ directory:

Developer Guides:

• docs/guides/creating-an-app.md — App structure, modules, registration, and component loading
• docs/guides/entities-and-services.md — Database models, business logic, CRUD, and fixtures
• docs/guides/graphql-api.md — Nodes, webservices, decorators, inputs, and pagination
• docs/guides/permissions.md — Authentication, authorization, and row-level access control

Functional Specifications:

• docs/FRS/auth.md — Authentication system (login, tokens, refresh, logout)
• docs/FRS/jwt_permissions.md — JWT permission system and access control
• docs/FRS/webservice_management.md — Webservice configuration and access levels
• docs/FRS/internal_service_communication.md — Service-to-service authentication

License

Apache License 2.0 — see LICENSE for details.