Skip to content

Status and fit

Artigraph is an early-stage, pre-1.0 project hosted by the LF AI & Data Foundation as a Sandbox project. The core model — artifacts, producers, fingerprints, and snapshots — is ready to explore in local workflows, and the examples in these docs are tested. APIs in the curated reference are the public surface to start from; other modules may change faster. Use the project changelog to track changes between releases.

What works today

  • Python: 3.13 and 3.14, both tested in CI.
  • Storage: local files.
  • Formats: JSON and Parquet, with Parquet support focused on Pandas DataFrames.
  • Metadata backends: in-memory by default, SQLite for durable local metadata, and Postgres for a shared durable backend.
  • Core workflow: define artifacts and producers, write raw inputs, build a graph, tag a snapshot, and read a named snapshot back.
  • Ecosystem interop: hand exact snapshot files to outside tools, such as DuckDB, without relying on path globs.
  • Experimental: GCS storage (GCSFile), with JSON and pickle IO only.

Planned / not yet supported

  • Cloud Parquet IO.
  • Additional storage and format implementations.

The examples include short notes showing the intended shape of future cloud storage, but the tested path today is local-first.

Where Artigraph fits

Artigraph is a library, not a scheduler. Use it with a script, cron, Airflow, Dagster, or any other runner when you want reproducible, versioned data artifacts with explicit snapshot lineage.

If you need task scheduling or a managed DAG, keep that tool. Let Artigraph own the data contract: types, storage, fingerprints, and the metadata that says which partitions belong to each snapshot.