Start here

RowGuard turns SQLAlchemy query results into validated Pydantic models.

Day-1 path: Install → Quickstart → skim SQLRules pushdown.

By default, SQLRules pushes supported constraints into SQL. Every row that reaches validation is accepted or explicitly rejected—never ignored after fetch. Use use_sqlrules=False when you need invalid rows classified in Python.

Choose a path

You want to…	Go to
Install and run a first query	Installation → Quickstart
Understand why rejected can be empty	SQLRules pushdown
Upgrade from a prior minor	Upgrading
Rejection policies	Rejection policies
Stream large results	Streaming
AsyncSession	Async
Performance tips	Performance
What is shipped vs planned	Supported vs planned
API contracts	API guide · Errors

What RowGuard is

• Orchestration over SQLAlchemy + SQLRules + Pydantic

• Sync and async buffered and streaming APIs

• Policies: raise (default), collect, skip, log, callback, quarantine

What RowGuard is not

• Not an ORM / persistence layer (it validates ORM and SQLModel reads)

• Not a migration tool

• Not a replacement for SQLAlchemy or Pydantic

• Not an authorization system (pushdown is not a security boundary)

Requirements

• Python 3.10+ (3.10–3.12 tested in CI; 3.13 untested)

• Pydantic v2, SQLAlchemy 2.x, SQLRules ≥ 1.0

Optional: pip install "rowguard[async]" / "rowguard[sqlmodel]".

Next

1. Install

2. Quickstart — default path first, then inspect rejections

3. FAQ / Troubleshooting