sqlspec 0.45.0

SQL Experiments in Python

SQLSpec

SQLSpec is a SQL execution layer for Python. You write the SQL -- as strings, through a builder API, or loaded from files -- and SQLSpec handles connections, parameter binding, SQL injection prevention, dialect translation, and mapping results back to typed Python objects. It uses sqlglot under the hood to parse, validate, and optimize your queries before they hit the database.

It works with PostgreSQL (asyncpg, psycopg, psqlpy), SQLite (sqlite3, aiosqlite), DuckDB, MySQL (asyncmy, mysql-connector, pymysql), Oracle (oracledb), CockroachDB, BigQuery, Spanner, and anything ADBC-compatible. Sync or async, same API. It also includes a built-in storage layer, native and bridged Arrow support for all drivers, and integrations for Litestar, FastAPI, Flask, Sanic, and Starlette.

Quick Start

pip install sqlspec

from pydantic import BaseModel
from sqlspec import SQLSpec
from sqlspec.adapters.sqlite import SqliteConfig

class Greeting(BaseModel):
    message: str

spec = SQLSpec()
db = spec.add_config(SqliteConfig(connection_config={"database": ":memory:"}))

with spec.provide_session(db) as session:
    greeting = session.select_one(
        "SELECT 'Hello, SQLSpec!' AS message",
        schema_type=Greeting,
    )
    print(greeting.message)  # Output: Hello, SQLSpec!

Write SQL, define a schema, get typed objects back. Or use the query builder -- they're interchangeable:

from sqlspec import sql

# Builder API -- same driver, same result mapping
users = session.select(
    sql.select("id", "name", "email")
       .from_("users")
       .where("active = :active")
       .order_by("name")
       .limit(10),
    {"active": True},
    schema_type=User,
)

Features

• Connection pooling -- sync and async adapters with a unified API across all supported drivers
• Parameter binding and dialect translation -- powered by sqlglot, with a fluent query builder and .sql file loader
• Result mapping -- map rows to Pydantic, msgspec, attrs, or dataclass models, or export to Arrow tables for pandas and Polars
• Storage layer -- read and write Arrow tables to local files, fsspec, or object stores
• Framework integrations -- Litestar plugin with DI, Starlette/FastAPI/Sanic middleware, Flask extension
• Observability -- OpenTelemetry and Prometheus instrumentation, structured logging with correlation IDs
• Event channels -- LISTEN/NOTIFY, Oracle AQ, and a portable polling fallback
• Migrations -- schema versioning CLI built on Alembic

Documentation

• Getting Started -- installation, adapter selection, first steps
• Usage Guides -- adapters, configuration, SQL file loader, and more
• Examples Gallery -- working code for common patterns
• API Reference -- full API docs
• CLI Reference -- migration and management commands

Playground

Want to try it without installing anything? The interactive playground runs SQLSpec in your browser with a sandboxed Python runtime.

Reference Applications

• PostgreSQL + Vertex AI Demo -- Vector search with pgvector and real-time chat using Litestar and Google ADK. Shows connection pooling, migrations, type-safe result mapping, vector embeddings, and response caching.
• Oracle + Vertex AI Demo -- Oracle 23ai vector search with semantic similarity using HNSW indexes. Demonstrates NumPy array conversion, large object handling, and real-time performance metrics.

Contributing

Contributions are welcome -- whether that's bug reports, new adapter ideas, or pull requests. Take a look at the contributor guide to get started.

License

MIT