advanced-alchemy

Advanced Alchemy

Code Style Rules

• Use Mapped[...] for columns and T | None for optional fields.
• Keep business transformations in service lifecycle hooks.
• Prefer the inner Repo service pattern and advanced_alchemy.* imports.
• Use from __future__ import annotations when it matches the project; 1.11 supports it in model modules.

Match-Your-Framework — read first

advanced-alchemy ships first-party extensions for five web frameworks. If your project uses one of these, jump directly to the matching integration guide and skip the others:

• Litestar — SQLAlchemyPlugin with full DI, session store, CLI. The rest of this SKILL.md covers Litestar by default; also see references/litestar_plugin.md.
• FastAPI → references/fastapi-integration.md — AdvancedAlchemy(config=..., app=app), Depends(alchemy.provide_session()) DI, provide_service()/provide_filters(), Alembic CLI via assign_cli_group.
• Flask → references/flask-integration.md — AdvancedAlchemy(config=..., app=app) or init_app() factory, pull-based alchemy.get_sync_session(), async-via-portal.
• Sanic → references/sanic-integration.md — AdvancedAlchemy(sqlalchemy_config=..., sanic_app=app) (note: sqlalchemy_config= kwarg, not config=), sanic-ext DI, request.ctx sessions.
• Starlette → references/starlette-integration.md — AdvancedAlchemy(config=..., app=app), request.state session access, lifespan wrapping.

Transaction configuration is framework-specific. Litestar uses before_send_handler; FastAPI, Flask, Starlette, and Sanic use commit_mode="manual", "autocommit", or "autocommit_include_redirect". Read the matching framework guide, then references/commit-modes.md and references/multi-database.md.

The rest of this SKILL.md covers framework-agnostic topics: base classes, repositories, services, filters, custom types, caching, replicas, operations, and Alembic migrations.

Overview

Advanced Alchemy is NOT a raw ORM — it is a service/repository layer built on top of SQLAlchemy 2.0+ with opinionated base classes, audit mixins, and deep framework integrations (Litestar, FastAPI, Flask, Starlette, Sanic). It provides:

• Base models with automatic id, created_at, updated_at fields
• Repository pattern for type-safe async CRUD
• Service layer with lifecycle hooks (to_model_on_create, to_model_on_update)
• Framework plugins for automatic session/transaction management
• Custom types: EncryptedString, FileObject, DateTimeUTC, GUID, Bool, Vector, TOTPSecret, OneTimeCode
• Alembic integration for migrations via CLI

Quick Reference

Base Classes

Base Class	PK Type	Audit Columns	When to Use
UUIDAuditBase	UUID v4	created_at, updated_at	Default choice for most models
UUIDBase	UUID v4	None	Lookup tables, tags, no audit needed
UUIDv7AuditBase	UUID v7	created_at, updated_at	Time-sortable IDs (preferred over v6)
BigIntAuditBase	BigInt auto-increment	created_at, updated_at	Legacy systems, integer PKs
NanoIDAuditBase	NanoID string	created_at, updated_at	URL-friendly short IDs
IdentityAuditBase	database identity	created_at, updated_at	Native IDENTITY columns
DefaultBase	None (define yourself)	None	Custom primary keys with AA table naming

Repository Pattern

Repository	Purpose
SQLAlchemyAsyncRepository[Model]	Standard async CRUD
SQLAlchemyAsyncSlugRepository[Model]	CRUD + automatic slug generation
SQLAlchemyAsyncQueryRepository	Complex read-only queries (no model_type)

Service Layer

Service	Purpose
SQLAlchemyAsyncRepositoryService[Model]	Full CRUD with lifecycle hooks
SQLAlchemyAsyncRepositoryReadService[Model]	Read-only (get_many, get, count, exists)

Key lifecycle hooks: to_model_on_create, to_model_on_update, to_model_on_upsert.

Custom Types

Type	Purpose	Notes
FileObject	Object storage with lifecycle hooks	Tracks file state across session; auto-deletes on row delete via StoredObject tracker
PasswordHash	Hashed password storage	Supports Argon2, Passlib, and Pwdlib backends; hashes on assignment
EncryptedString	Transparent encryption at rest	Pass a stable key explicitly; the random default is deprecated
UUID6 / UUID7	Time-sortable UUID variants	UUID7 preferred for standardized timestamp-ordered identifiers
DateTimeUTC	Timezone-aware UTC datetime	Stores as UTC; raises on naive datetimes
Bool	Dialect-aware boolean	Uses Oracle 23c native BOOLEAN when SQLAlchemy exposes it; falls back to stock SQLAlchemy Boolean
Vector	Dialect-aware vector storage and distance operators	Oracle 23ai VECTOR, PostgreSQL/CockroachDB pgvector, JSON fallback without distance operators
TOTPSecret / OneTimeCode	MFA and single-use code storage	TOTPSecret encrypts shared secrets; OneTimeCode hashes codes and requires an explicit hashing backend

Repository Service Layer

SQLAlchemyAsyncRepositoryService is the primary service base class. Key behaviors:

• Dict-to-model conversion: pass raw dict to create(), update(), upsert() — the service converts via to_model_on_create / to_model_on_update lifecycle hooks before persistence
• Bulk operations: create_many(data), update_many(data), upsert_many(data), delete_many(item_ids) — batched in a single transaction; delete_many() accepts raw primary keys, composite-key tuples/dicts, model instances, or mixed lists
• Lifecycle hooks: to_model_on_create, to_model_on_update, to_model_on_upsert — override to transform input data, hash passwords, normalize strings, etc.

Mixins

Mixin	Fields Added	When to Use
AuditColumns	created_at, updated_at	Add timestamps to a model with a custom primary key
SlugKey	unique slug column	Pair with a slug repository; the mixin does not generate values
UniqueMixin	as_unique_async() / as_unique_sync()	Session-cached select-or-create after defining unique_hash() and unique_filter()
SentinelMixin	hidden sa_orm_sentinel column	Deterministic ordering for SQLAlchemy bulk inserts; not optimistic locking

Litestar Integration

Use SQLAlchemyPlugin (composite of SQLAlchemyInitPlugin + SQLAlchemySerializationPlugin) for full integration:

• SQLAlchemyPlugin: registers engine/session providers, a Litestar before_send hook, and ORM type encoders in one call
• SQLAlchemyDTO: generates Litestar DTOs directly from ORM models with include/exclude field control
• Type encoders: automatic serialization of datetime, UUID, Decimal, Enum, and custom column types
• Exception handling: set_default_exception_handler=True (the default) registers RepositoryError handling through the plugin

Workflow

Step 1: Define the Model

Choose the appropriate base class from the quick reference table. Use UUIDAuditBase unless you have a specific reason not to. Define columns with Mapped[] typing.

Step 2: Create the Repository

Create a repository class with model_type set to your model. Use SQLAlchemyAsyncRepository for standard CRUD, SQLAlchemyAsyncSlugRepository if the model uses SlugKey.

Step 3: Build the Service

Create a service class with an inner Repo class. Set match_fields for upsert logic. Add lifecycle hooks (to_model_on_create, to_model_on_update) for business logic transformations.

Step 4: Wire into Framework

Use the framework plugin (Litestar, FastAPI, Flask, Sanic) to inject sessions and register the service as a dependency.

Step 5: Generate Migration

With Litestar, run litestar database make-migrations -m "description" and then litestar database upgrade. With the standalone CLI, put the required config option before the command: alchemy --config path.to.config make-migrations -m "description".

Guardrails

• Always use the service layer for business logic — never put validation, hashing, or transformation logic directly in route handlers or repositories
• Repositories are for data access only — no business rules, no side effects beyond database operations
• Never bypass the service layer to call repository methods directly from handlers
• Always set match_fields on services that use upsert() to avoid duplicate-key errors
• Use schema_dump() / schema_dump_config for explicit dump behavior — services already convert Pydantic/msgspec/attrs/dataclass inputs during model conversion
• Prefer UUIDAuditBase as default base class — only deviate when you have a concrete reason
• Use advanced_alchemy.* imports — the old litestar.plugins.sqlalchemy paths are deprecated
• Pass stable keys to EncryptedString and EncryptedText. Omitting key= emits a 1.11 deprecation warning and produces data that cannot survive a process restart.
• Use get_many() and get_many_and_count(). list() and list_and_count() are deprecated until 2.0.

Validation Checkpoint

Before delivering code, verify:

• Model inherits from an Advanced Alchemy base class (not raw DeclarativeBase from SQLAlchemy)
• All columns use Mapped[] type annotations
• Service has an inner Repo class with model_type set
• Business logic lives in service lifecycle hooks, not in route handlers
• Imports come from advanced_alchemy.*, not deprecated paths
• Encrypted columns receive a stable explicit key
• New code uses get_many() / get_many_and_count(), not deprecated list aliases

Example

A complete Tag entity with model, repository, and service:

"""Tag domain — model, repository, and service."""

from advanced_alchemy.base import UUIDAuditBase
from advanced_alchemy.repository import SQLAlchemyAsyncRepository
from advanced_alchemy.service import ModelDictT, SQLAlchemyAsyncRepositoryService
from sqlalchemy.orm import Mapped, mapped_column


class Tag(UUIDAuditBase):
    """Tag model with audit trail."""

    __tablename__ = "tag"

    name: Mapped[str] = mapped_column(unique=True)
    description: Mapped[str | None] = mapped_column(default=None)


class TagRepository(SQLAlchemyAsyncRepository[Tag]):
    """Data access for tags."""

    model_type = Tag


class TagService(SQLAlchemyAsyncRepositoryService[Tag]):
    """Business logic for tags."""

    class Repo(SQLAlchemyAsyncRepository[Tag]):
        model_type = Tag

    repository_type = Repo
    match_fields = ["name"]

    async def to_model_on_create(self, data: ModelDictT[Tag]) -> ModelDictT[Tag]:
        """Normalize tag name before creation."""
        if isinstance(data, dict) and "name" in data:
            data["name"] = data["name"].strip().lower()
        return data

---

References Index

Choosing between advanced-alchemy and sqlspec: advanced-alchemy (this skill) gives you an opinionated ORM service layer with UUIDAuditBase, lifecycle hooks, repository / service / Alembic integration, and OffsetPagination[T] out of the box — pick it when you want a complete CRUD surface with attribute-style row access and you're happy inside the SQLAlchemy ecosystem. sqlspec gives you direct SQL control, 15+ driver adapters (asyncpg, oracledb, DuckDB, BigQuery, SQLite, and more), Arrow-native result streams for analytics, and a builder API when you need it — pick it when you want explicit SQL, heterogeneous database backends, or Arrow integration. Both skills integrate with Litestar via first-party plugins; see ../sqlspec/SKILL.md for the raw-SQL / multi-adapter path.

For detailed guides and code examples, refer to the following documents in references/:

• Models Base classes, mixins, special types, relationships, PII tracking, and deferred loading.
• Repositories Async repository variants, configuration, slug repos, and query repos.
• Services Service layer, lifecycle hooks, composite services, filtering, and pagination.
• Litestar Plugin SQLAlchemy plugin config, DTOs, dependency injection, and session management.
• Migrations Alembic integration, CLI commands, metadata registry, and multi-database support.
• Types Complete catalog of custom column types: EncryptedString, FileObject, DateTimeUTC, GUID, PasswordHash, ColorType, and more.
• Base Classes Declarative base classes, UUID/BigInt/Nanoid variants, audit mixins, SlugKey, UniqueMixin, metadata registry, and custom base creation.
• Filters Filter system, pagination, SearchFilter, CollectionFilter, BeforeAfter, OrderBy, LimitOffset, and frontend integration patterns.
• Framework Integrations FastAPI, Flask, Starlette, and Sanic plugin setup, session management, and feature comparison across frameworks.
• Caching Dogpile.cache integration, CacheConfig, CacheManager API, automatic cache invalidation via session events, version-based list cache keys, singleflight stampede protection, and serialization.
• Read Replicas Read/write routing, RoutingConfig, engine groups, RoundRobinSelector/RandomSelector, sticky-after-write consistency, context managers for explicit routing, and RoutingAsyncSessionMaker.
• Storage (obstore) FileObject and StoredObject types, ObstoreBackend and FSSpecBackend configuration (S3, GCS, Azure, local), StorageRegistry, presigned URL generation, automatic file lifecycle via session tracker, and Pydantic integration.
• Operations, Listeners, Serialization OnConflictUpsert / MergeStatement dialect-aware upsert building blocks, session event listeners (FileObject, cache invalidation, touch_updated_timestamp), and the msgspec-first encode_json / decode_json used across the library.

---

Official References

• https://github.com/litestar-org/advanced-alchemy/tree/v1.11.0/advanced_alchemy
• https://github.com/litestar-org/advanced-alchemy/blob/v1.11.0/docs/changelog.rst
• https://github.com/litestar-org/advanced-alchemy/tree/v1.11.0/tests
• https://github.com/litestar-org/advanced-alchemy/tree/v1.11.0/docs/usage
• https://docs.litestar.dev/2/release-notes/changelog.html
• https://docs.sqlalchemy.org/en/20/orm/quickstart.html

Shared Styleguide Baseline

• Use shared styleguides for generic language/framework rules to reduce duplication in this skill.
• General Principles
• Python
• Litestar
• Keep this skill focused on tool-specific workflows, edge cases, and integration details.