Colnade

A statically type-safe DataFrame abstraction layer for Python.

Colnade replaces string-based column references (pl.col("age")) with typed descriptors (Users.age), so column misspellings, type mismatches, and schema violations are caught by your type checker — before your code runs.

Works with ty, mypy, and pyright. No plugins, no code generation.

Get Started Tutorials

---

Quick Example

Define SchemaRead & TransformOutput Schema

from colnade import Column, Schema, UInt64, Float64, Utf8

class Users(Schema):
    id: Column[UInt64]
    name: Column[Utf8]
    age: Column[UInt64]
    score: Column[Float64]

from colnade_polars import read_parquet

df = read_parquet("users.parquet", Users)

result = (
    df.filter(Users.age > 25)
      .sort(Users.score.desc())
      .select(Users.name, Users.score)
)

class UserSummary(Schema):
    name: Column[Utf8]
    score: Column[Float64]

output = result.cast_schema(UserSummary)
# output is DataFrame[UserSummary]

---

Key Features

• Type-safe column references — Users.naem is a type error, not a runtime crash
• Schema-preserving operations — filter, sort, with_columns preserve DataFrame[S]
• Typed expressions — Users.age > 18 produces Expr[Bool], Users.score * 2 produces Expr[Float64]
• Backend agnostic — works with Polars today, with adapters for other engines
• Generic utility functions — write def f(df: DataFrame[S]) -> DataFrame[S] that works with any schema
• Struct and List support — typed access to nested data structures
• No plugins or codegen — works with standard type checkers out of the box