Skip to content
ClearFeature

Open source

Production feature infrastructure, built in the open.

Keep feature logic in your own Python project, inspect the execution engine, and run ClearFeature in infrastructure you control.

What is open

ClearFeature is published under the Apache-2.0 license by ClearFeature Labs. The engine you would run in production — API, workers, compute core, storage integrations — is the code in the public repository, together with the documentation, the quickstart, and the engineering demo.

license

Apache-2.0

A permissive license with no copyleft obligations on your feature code or your deployment.

repository

One public codebase

The platform you evaluate is the platform you deploy — not a cut-down community edition next to a closed enterprise fork.

docs

Documentation in the repo

Architecture, API contracts, observability, and the quickstart live beside the code they describe. Read the docs →

Your Feature Project stays yours

The boundary is deliberate: ClearFeature Core is the engine; your Feature Project is a separate Python package in your repository, under your review process and your CI. Adopting the platform never means handing your feature logic to a vendor format you cannot leave.

Feature Project — your repository

snapshot-ratio/
├── feature_project.yaml
└── snapshot_ratio/
    ├── features.py
    ├── registry/features_v1.yaml
    └── tests/golden.yaml
  • features.py — feature UDFs in plain Python
  • registry — sources, features, versions, dependencies, groups
  • golden.yaml — expected values, run as tests

Feature Runtime

ClearFeature Core loads your Feature Project and executes the same registry-defined DAG for historical materialization, point-in-time training datasets, and live request-time computation.

Your code stays your code — Core is the engine, not the owner.

Inspectable execution

For production ML systems, 'trust us' is not an execution model. Because the compute core is open, your engineers can read exactly how source values become feature values: how dependencies resolve, how availability is enforced in training datasets, how materialization stays idempotent. Every claim on this website is checkable against code.

Build and test features like software

Feature UDFs are plain Python functions. The registry is reviewable YAML. Golden tests pin expected values for every feature and run through the same compute core that production uses — so a feature is tested before it is ever deployed, not observed after it misbehaves.

cases:
  - name: ratio
    feature: payment_to_income_ratio
    deps: {active_monthly_payment: 750.0, monthly_income: 3000.0}
    expected: {value: 0.25}

A golden case from the public quickstart: expected values as data, executed as tests.

Run it yourself

ClearFeature is designed for self-hosted deployment on components your team can operate: PostgreSQL, Valkey, S3-compatible object storage, and a Kafka-compatible broker. No required data SaaS sits between your source reports and your features.

The self-hosted deployment model →

Community path, commercial help

The open-source platform is self-service: clone it, run the quickstart, deploy it, file issues. Nothing about evaluation requires talking to us. Organizations that want direct help with production adoption — architecture review, deployment planning, migration — can start that conversation separately.

Start with the Quickstart.

Scaffold a Feature Project, define three features across two source reports, and run golden tests — in about five minutes.