Measure data quality. Keep the decisions yours.¶

Kontra is a developer-first measurement engine for data contracts. It evaluates declarative rules against files, databases, and DataFrames, then returns structured results for humans, CI systems, and agents.

Get started Python API View on GitHub

import kontra
from kontra import rules

result = kontra.validate("users.parquet", rules=[
    rules.not_null("user_id"),
    rules.unique("email"),
    rules.range("age", min=0, max=120),
])

result.passed
result.to_dict()   # structured output
result.to_llm()    # compact agent context

Runs where the data lives¶

Kontra resolves rules from metadata first, pushes eligible work into SQL, and uses Polars only for the remaining measurements.

One result model¶

CLI, Python, CI, and MCP consumers receive the same deterministic counts, execution sources, severity metadata, and bounded explanations.

Designed for composition¶

Kontra measures. Your application decides what blocks a deployment, opens an incident, or needs human review.

Choose a path¶

If you want to…	Start here
Validate a file or table	Getting started
Use DataFrames or inline rules	Python API
Define a reusable contract	Contracts
Inspect all built-in measurements	Rules
Compare transformation inputs and outputs	Transformation probes
Connect Claude, Codex, Cursor, or another MCP client	Agents and MCP
Understand execution cost and exact counts	Performance

Supported data sources¶

Parquet, CSV, PostgreSQL, SQL Server, ClickHouse, S3-compatible object storage, Azure ADLS Gen2, Polars DataFrames, pandas DataFrames, and Python records all enter the same measurement API.

pip install kontra
kontra profile users.parquet --draft > contract.yml
kontra validate contract.yml

Kontra is open source under the Apache License 2.0.