metaxy

name	metaxy
description	This skill should be used when the user asks to "define a feature", "create a BaseFeature class", "track feature versions", "set up metadata store", "field-level lineage", "FieldSpec", "FeatureDep", "run metaxy CLI", "metaxy migrations", or needs guidance on metaxy feature definitions, versioning, metadata stores, CLI commands, or testing patterns.

Metaxy

Metaxy is a metadata layer for multimodal Data and ML pipelines that manages and tracks feature versions, dependencies, and data lineage across complex computational graphs.

Core Concepts

Feature Definitions

To define a feature, create a class inheriting from mx.BaseFeature with a FeatureSpec metaclass argument:

import metaxy as mx


class Video(
    mx.BaseFeature,
    spec=mx.FeatureSpec(
        key="media/video",
        id_columns=["video_id"],
        fields=["audio", "frames"],  # Logical fields: describe data contents for versioning
    ),
):
    # Metadata columns: stored in the metadata store, tracked by metaxy
    video_id: str
    path: str
    duration: float
    height: int
    width: int

Important distinction: fields in FeatureSpec are logical field specs that describe the data contents for versioning and lineage tracking. Class attributes are metadata columns stored in the metadata store. They serve different purposes and should not overlap.